Problem positioning analysis method and device, equipment and medium

Abstract

The invention discloses a problem positioning analysis method and device, equipment and a medium, and relates to the field of cloud storage. Comprising the following steps: constructing a knowledge graph based on data of a plurality of databases in a cloud storage system; based on the knowledge graph, training a to-be-trained graph attention network to obtain a trained graph attention network, the trained graph attention network carrying knowledge in the knowledge graph; in the testing process of the cloud storage system, determining a target task with a failed task, and obtaining task information corresponding to the target task; and inputting task information corresponding to the target task into the trained graph attention network to obtain a failure reason of the target task. By means of the method, the efficiency of positioning personnel in the aspect of problem positioning can be remarkably improved, and then the testing efficiency and accuracy are improved.

Description

Technical Field

[0001] This invention relates to the field of cloud storage, and in particular to a problem localization and analysis method, apparatus, device, and medium. Background Technology

[0002] Cloud storage is a service that stores data on remote servers via the internet, allowing users to access and manage this data anytime, anywhere without storing it on local devices.

[0003] In cloud storage systems, cloud management services and Cinder services each maintain independent database systems, and each database contains multiple data tables. This data structure presents significant challenges in the problem localization process: in relational databases, data is widely distributed across different databases and tables, and querying complex relationships often requires performing multiple JOIN operations, resulting in a lack of intuitive data presentation, degraded query performance, and prolonged problem localization time.

[0004] During the problem localization process, troubleshooters need to start from the cloud management layer and gradually delve into the Cinder layer for detailed investigation. This process is tedious and complex, severely impacting the efficiency of test localization. Furthermore, current problem analysis primarily relies on manual methods, which, limited by the experience and knowledge level of the troubleshooters, not only takes a long time to analyze the problem but is also relatively inefficient.

[0005] In summary, current cloud storage system testing faces multiple challenges, including unintuitive data visualization, low problem localization efficiency, and significant limitations of manual analysis. To improve testing efficiency and accuracy, it is necessary to explore more efficient problem localization methods. Summary of the Invention

[0006] In view of the above-mentioned technical problems, the present invention provides a problem localization analysis method, apparatus, equipment and medium, which aims to overcome the above problems or at least partially solve the above problems.

[0007] The first aspect of this invention provides a problem localization and analysis method, the method comprising: A knowledge graph is constructed based on data from multiple databases in a cloud storage system. Based on the knowledge graph, the graph attention network to be trained is trained to obtain a trained graph attention network, which carries the knowledge in the knowledge graph. During the testing of the cloud storage system, the target task that failed is identified, and the task information corresponding to the target task is obtained; The task information corresponding to the target task is input into the trained graph attention network to obtain the reason for the failure of the target task.

[0008] A second aspect of the present invention provides a problem localization and analysis device, the device comprising: The building module is used to construct knowledge graphs based on data from multiple databases in a cloud storage system; The training module is used to train the graph attention network to be trained based on the knowledge graph to obtain a trained graph attention network, wherein the trained graph attention network carries the knowledge in the knowledge graph. The acquisition module is used to identify the target task that failed during the testing process of the cloud storage system and to acquire the task information corresponding to the target task. The localization module is used to input the task information corresponding to the target task into the trained graph attention network to obtain the failure reason of the target task.

[0009] A third aspect of the present invention provides an electronic device comprising a processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein the program or instructions, when executed by the processor, implement the steps of the problem localization and analysis method as described in the first aspect of the present invention.

[0010] A fourth aspect of the present invention provides a readable storage medium on which a program or instructions are stored, which, when executed by a processor, implement the steps of the problem localization and analysis method as described in the first aspect of the present invention.

[0011] In the problem localization and analysis method proposed in this invention, a knowledge graph of the cloud storage domain is constructed based on data from multiple data streams dispersed in the cloud storage system. A graph attention network is then trained based on this knowledge graph. Therefore, during the testing process of the cloud storage system, when a task fails, knowledge reasoning can be performed based on the trained graph attention network to predict the cause of the target task's failure. Thus, this invention's knowledge reasoning technology based on graph attention networks intelligently completes the constructed knowledge graph. This measure aims to utilize the trained graph attention network to automatically mine and fill in potential connections and missing information in the knowledge graph, enhancing the graph's completeness and depth, and reasoning to predict possible outcomes and corresponding failure reasons. This provides strong support for problem localization personnel, helping them to pinpoint the problem more quickly and accurately, significantly improving their efficiency in problem localization, and consequently enhancing testing efficiency and accuracy. Attached Figure Description

[0012] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is a flowchart illustrating the steps of a problem localization and analysis method according to an embodiment of the present invention; Figure 2 This is a schematic diagram illustrating the construction of a knowledge graph according to an embodiment of the present invention; Figure 3 This is a flowchart illustrating an efficient localization analysis method based on graph attention network knowledge reasoning, according to an embodiment of the present invention. Figure 4 This is a partial schematic diagram of knowledge reasoning shown in an embodiment of the present invention; Figure 5 This is a partial schematic diagram illustrating a successful backup inference according to an embodiment of the present invention; Figure 6 This is a partial schematic diagram illustrating a failure details inference method according to an embodiment of the present invention; Figure 7 This is a structural block diagram of a problem localization and analysis device provided in an embodiment of the present invention; Figure 8 This is a schematic diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0014] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0015] Please refer to Figure 1 , Figure 1 This is a flowchart illustrating the steps of a problem localization and analysis method according to an embodiment of the present invention. Figure 1 As shown, the problem localization and analysis method provided in this embodiment includes at least the following steps: Step S11: Construct a knowledge graph based on data from multiple databases in the cloud storage system.

[0016] In this embodiment, the cloud storage system maintains multiple independent database systems. Based on the data from these databases, a knowledge graph can be constructed. A knowledge graph is a structured semantic knowledge base that organizes, integrates, and stores various entities and their complex relationships in a network structure. An entity is an independent and identifiable object or abstract concept that can be distinguished from other things, such as a person, place, organization, or event. Relationships describe the connections and interactions between these entities, including their various types and attributes, such as "is," "located in," and "belongs to."

[0017] Step S12: Based on the knowledge graph, train the graph attention network to be trained to obtain a trained graph attention network.

[0018] In this embodiment, the graph attention network to be trained can be trained based on the constructed knowledge graph to obtain a trained graph attention network, which carries the knowledge from the knowledge graph. The graph attention network (GAT) is a deep learning model used to process graph-structured data.

[0019] Step S13: During the testing of the cloud storage system, identify the target task that failed and obtain the task information corresponding to the target task.

[0020] In this embodiment, during the testing of the cloud storage system, the target task that failed in the test is identified, and the task information corresponding to the target task during the testing process is obtained. In this embodiment, the target task is the test task that failed. For example, if the test task is to create a backup 2 for cloud disk 4 in the repository, and the test result is that the backup creation failed, then this test task is the target task.

[0021] Step S14: Input the task information corresponding to the target task into the trained graph attention network to obtain the failure reason of the target task.

[0022] In this embodiment, after obtaining the task information corresponding to the target task, the task information can be input into the trained graph attention network for knowledge graph reasoning. This allows the network to output the reasons for the target task's failure, thereby enabling problem localization and analysis during cloud storage system testing. The goal of knowledge graph reasoning is to utilize existing knowledge in the knowledge graph to discover potential, unknown (not yet clearly defined) knowledge through reasoning.

[0023] In this embodiment, knowledge reasoning technology based on graph attention networks is used to intelligently complete the constructed knowledge graph. This measure aims to use the trained graph attention network to automatically mine and fill in potential relationships and missing information in the knowledge graph, enhance the integrity and depth of the graph, and infer and predict possible results and corresponding reasons for failure, thereby providing strong support for problem locators. This helps problem locators to pinpoint the problem more quickly and accurately, significantly improving the efficiency of problem locators and thus improving testing efficiency and accuracy.

[0024] In conjunction with the above embodiments, in one implementation, the present invention also provides a problem localization and analysis method, in which step S14 may specifically include step S21: Step S21: Take the task object of the target task as the first entity, and input the failure details of the target task as the first relation into the trained graph attention network to obtain the second entity output by the trained graph attention network, and take the second entity as the failure reason of the target task.

[0025] In this embodiment, the constructed knowledge graph includes multiple triples; the task information corresponding to the target task includes at least: the task object of the target task and the failure details of the target task. Here, the task object refers to the target object in the cloud storage system test task. For example, if the test task is to create a backup 2 for cloud disk 4 in the repository, and the test result is that the backup creation fails, then this test task is the target task, and the task object of the target task is backup 2. The failure details of the target task are various task detail data recorded during the test process of the target task, such as the task flow, task nodes, etc., and there are no restrictions on this.

[0026] In this embodiment, the task object of the target task can be taken as the first entity, and the failure details of the target task can be taken as the first relation and input into the trained graph attention network for knowledge reasoning to obtain the second entity output by the trained graph attention network. The second entity is then used as the reason for the failure of the target task. Here, the first entity and the second entity are two different entities in a triple. It can be understood that in this embodiment, the first entity, the first relation, and the second entity constitute a triple. That is, the input of the trained graph attention network is the relation in the triple and one of the entities (such as the relation and the head entity in the triple, or the relation and the tail entity in the triple), and the output is the other entity in the triple (such as the tail entity in the triple, or the head entity in the triple).

[0027] In conjunction with any of the above embodiments, the present invention also provides a problem localization analysis method. In this method, the trained graph attention network includes a trained encoder and a trained decoder. Furthermore, step S21 above, "taking the task object of the target task as a first entity and the failure details information of the target task as a first relation input to the trained graph attention network to obtain the second entity output by the trained graph attention network," specifically includes the following steps S31 to S33: Step S31: Preprocess the first entity and the first relation using the TransE model to obtain the first embedding vector of the first entity and the first relation.

[0028] In this embodiment, before inputting the first entity and the first relation into the trained graph attention network, the first entity and the first relation can be preprocessed: the original triplet data (i.e., the first entity and the first relation) is preprocessed using the TransE model to obtain the first embedding vector output by the TransE model. The TransE model is a knowledge representation learning method that can represent entities and relations as low-dimensional embedding vectors while preserving the association information between them. This first embedding vector includes: the low-dimensional embedding vectors of the first entity and the first relation, and the association information between the first entity and the first relation.

[0029] Step S32: Input the first embedding vector into the trained encoder, and convert the first embedding vector into a first encoding result through the trained encoder.

[0030] In this embodiment, the first embedding vector is input into the trained encoder, which encodes the first embedding vector, converting it into a first encoding result, thereby obtaining the first encoding result output by the trained encoder. The first encoding result can be understood as the hidden state of the first embedding vector.

[0031] Step S33: The first encoding result is processed by the trained decoder to predict the position and relationship of the first embedding vector in the knowledge graph, thereby obtaining the second entity.

[0032] In this embodiment, after obtaining the first encoding result output by the trained encoder, the first encoding result can be input into the trained decoder. The trained decoder processes the first encoding result to predict the position and relationship of the first embedding vector in the knowledge graph, obtain the second entity and output it. The second entity is the reason for the failure of the target task obtained by the trained graph attention network through knowledge reasoning.

[0033] In this embodiment, the knowledge reasoning of the trained graph attention network adopts an encoder-decoder framework. The trained encoder is responsible for converting the input graph structure data (i.e., the first embedding vector) into hidden states, while the trained decoder generates reasoning results based on these hidden states. Through the trained graph attention network, it can learn embedding representations of entities and relations that reflect their positions and relationships in the knowledge graph.

[0034] In one alternative embodiment, the knowledge reasoning algorithm uses the PyTorch framework to construct a graph attention network. A graph attention network is a graph neural network based on an attention mechanism that dynamically assigns weights to each neighboring entity based on the relationship between the neighboring entity and the central entity when aggregating neighborhood information. This adaptive weight allocation mechanism allows the graph attention network to better capture the relationships and importance between nodes, thereby improving the accuracy of reasoning.

[0035] In conjunction with any of the above embodiments, in one implementation, the present invention also provides a problem localization and analysis method. In this method, the cloud storage system includes at least: a cloud management server and a Cinder server; and step S11 above specifically includes steps S41 to S42: Step S41: Obtain table data related to the cloud storage field from the databases of the cloud management server and the Cinder server.

[0036] In this embodiment, within the cloud storage architecture, the cloud management server and the Cinder server each maintain an independent database system, and these databases each contain multiple data tables. This embodiment can retrieve table data related to the construction of a knowledge graph in the cloud storage domain from the databases of the cloud management server and the Cinder server. This table data will serve as the foundation for knowledge graph construction, and the table data will include at least the following key information: entities, relationships, and attributes, where attributes represent information describing entities.

[0037] Step S42: Convert the table data into multiple triples, and integrate the multiple triples into a graph database to construct a knowledge graph for the cloud storage domain.

[0038] In this embodiment, the acquired tabular data can be converted into triples, that is, the tabular data can be converted into multiple triples, and the multiple triples can be integrated into a graph database to construct a knowledge graph in the cloud storage domain.

[0039] In a specific example, the tabular data is first transformed: for example, using a specially written script or efficient data processing tools, the tabular data in the database is converted into the form of triples (head entity, relation, tail entity). This transformation process aims to accurately represent the relationships and attributes between entities, laying the foundation for subsequent knowledge graph construction. Then, the transformed triples are integrated: the triples extracted from various databases and tables are integrated into a unified graph database. This step ensures the comprehensiveness and consistency of knowledge, enabling data from different sources to be efficiently managed and analyzed within the same framework.

[0040] In one alternative implementation, after retrieving tabular data related to cloud storage from the databases of the cloud management server and the Cinder server, the tabular data can be cleaned to ensure its accuracy, integrity, and consistency. After obtaining the cleaned tabular data, it is transformed into multiple triples, and these triples from different databases and tables are effectively integrated into a unified graph database.

[0041] In one specific embodiment, the data cleaning step includes at least one or more of the following: 1. Remove duplicate fields and invalid data: Given that the databases of the cloud management server and Cinder server may contain duplicate representations of certain information (such as volume ID and volume name), it is necessary to carefully identify and remove these duplicate fields. A comprehensive check should also be performed on the exported data to remove null values, duplicate values, and any potentially invalid data.

[0042] 2. Data formatting: To facilitate subsequent data processing and analysis, date, numerical, and other fields are formatted to ensure that their formats are consistent and meet the requirements for knowledge graph construction.

[0043] 3. Data type conversion: For string data, it will be converted into an appropriate type (such as date, number, etc.) according to its actual meaning and purpose to ensure data accuracy.

[0044] 4. Field Name and Value Consistency Verification: Carefully compare field names and values ​​across different tables to ensure that the same fields have consistent naming and meanings. This step helps eliminate data ambiguity and improves the accuracy and usability of the knowledge graph.

[0045] In one embodiment, an example of integrating tabular data from the Cinder server and cloud management server into a graph database can be as follows: Figure 2 As shown, Figure 2 This is a schematic diagram illustrating the construction of a knowledge graph according to an embodiment of the present invention. Figure 2In the Cinder server, the volume table contains fields such as volume ID (volumes_id), volume name (volumes_name), volume size (size), and volume status (volumes_status). The cloud management server contains fields such as volume ID (volumes_id), label, and disk status in the cloud management layer (status). Data cleaning is then performed to obtain unique fields: volume ID (volumes_id), volume name (volumes_name), volume size (size), volume status (volumes_status), label, and disk status in the cloud management layer (status). Using the volume ID as a unique identifier, triples can be constructed based on the fields in the Cinder server, such as (id1, hasName, name1), (id1, hasSize, 10), and (id1, hasVolumeStatus, available); and triples can be constructed based on the unique fields in the cloud management server, such as (id1, hasLabel, label) and (id1, hasStatus, available).

[0046] In conjunction with any of the above embodiments, in one implementation, the present invention also provides a problem localization and analysis method. In addition to the steps described above, this method may further include steps S51 to S53: Step S51: Store the cloud storage domain knowledge graph in a graph database platform.

[0047] In this embodiment, after obtaining the cloud storage domain knowledge graph, it can be stored in a graph database platform. In one embodiment, Neo4j can be selected as the graph database platform, as Neo4j can efficiently store and manage graph data and provide fast query performance.

[0048] Step S52: Use the visualization tools provided by the graph database platform to display the knowledge graph of the cloud storage domain in a graphical manner.

[0049] In this embodiment, visualization tools provided by a graph database platform can be used to graphically display the stored cloud storage domain knowledge graph, thus providing an intuitive understanding of the relationships and attributes between entities. In an optional example, visualization tools provided by the graph database platform Neo4j are used to graphically display the integrated cloud storage domain knowledge graph.

[0050] Step S53: In response to the user's query operation, use the Cypher query language to query the target entity and / or target relationship in the cloud storage domain knowledge graph.

[0051] In this embodiment, users can perform relevant query operations on the knowledge graph of the cloud storage domain. In response to the user's query operation, the Cypher query language is used to query the target entity and / or target relationship in the knowledge graph of the cloud storage domain. That is, the Cypher query language is used to retrieve and view the specific entity and all its related entity relationship information, which can quickly and intuitively discover problems or view entity information in a large amount of data.

[0052] In this embodiment, to enhance the intuitive presentation of data, data scattered across multiple databases is transformed into triples (entity-relationship-entity), and these triples are integrated into a unified graph database for centralized display. This allows for a clearer view of the relationships and connections between data, facilitating comprehensive browsing and understanding, improving data readability and comprehension, and achieving efficient data presentation. The intuitiveness of the data can quickly identify problems, reduce misunderstandings and communication costs, and avoid data omissions, offering multiple advantages such as improved testing efficiency, enhanced testing accuracy, optimized testing strategies, and improved decision-making quality. Once a problem (such as a target task) is identified, a trained graph attention network can be used to locate the problem and determine the cause of the target task's failure. Furthermore, the Cypher query language can be used to retrieve and view information about a specific entity and all its related entity relationships, enabling rapid and intuitive problem identification and more effective data analysis and problem localization.

[0053] In conjunction with any of the above embodiments, in one implementation, the present invention also provides a problem localization analysis method. In this method, the knowledge graph includes multiple triples; and step S12 specifically includes steps S61 to S62: Step S61: Preprocess the relations and any entity in the triples using the TransE model to obtain the second embedding vectors of the relations and any entity in the triples.

[0054] In this embodiment, after constructing the knowledge graph, for any triple in the knowledge graph, the TransE model can be used to preprocess the relation and any entity in the triple to obtain the second embedding vector corresponding to the triple output by the TransE model. The second embedding vector can be understood as the first embedding vector during model training, which is then used as input to the model. The second embedding vector includes: the low-dimensional embedding vectors of the relation and any entity in the triple, and the association information between the relation and any entity.

[0055] Step S62: Using the second embedding vector as a training sample and the remaining entities in the triplet as labels, train the graph attention network to be trained by minimizing the loss function to obtain the trained graph attention network.

[0056] In this embodiment, the second embedding vector output by the TransE model can be used as a training sample, and the remaining entities in the triplet can be used as labels. The graph attention network to be trained is then trained by minimizing the loss function to obtain a trained graph attention network. Specifically, the second embedding vector can be input into the graph attention network to be trained for processing to obtain the output of the graph attention network to be trained. Then, based on the output and labels of the graph attention network to be trained, the loss value is calculated by minimizing the loss function. The model parameters of the graph attention network to be trained are updated based on the loss value until the minimization of the loss function converges. The model parameters are then fixed to obtain the trained graph attention network.

[0057] In this embodiment, by minimizing the loss function and continuously adjusting the model parameters, the model's prediction results can be made more accurate.

[0058] In conjunction with any of the above embodiments, in one implementation, the present invention also provides a problem localization analysis method. In this method, the graph attention network to be trained includes: an encoder to be trained and a decoder to be trained; and step S62 specifically includes steps S71 to S74: Step S71: Input the second embedding vector into the encoder to be trained to obtain the second encoding result; process the second encoding result through the decoder to be trained to obtain the first prediction result.

[0059] In this embodiment, in the first round of training, the preprocessed second embedding vector can be input to the encoder to be trained for processing to obtain the second encoding result output by the encoder to be trained. Then, the second encoding result is input to the decoder to be trained. The decoder to be trained processes the second encoding result to obtain the first prediction result output by the decoder to be trained.

[0060] Step S72: Based on the first prediction result and the remaining entities in the triplet, the network parameters of the encoder and the decoder to be trained are updated by minimizing the hinge loss function until the minimized hinge loss function converges, thereby obtaining the trained encoder and intermediate decoder.

[0061] In this embodiment, after obtaining the first prediction result corresponding to the second embedding vector, the first loss value can be calculated by minimizing the hinge loss function based on the remaining entities in the triplet corresponding to the first prediction result and the second embedding vector. The network parameters of the encoder and the decoder to be trained are updated by the first loss value until the minimized hinge loss function converges. The network parameters of the encoder and the decoder to be trained are then fixed to obtain the trained encoder and intermediate decoder, thus completing the first round of training.

[0062] Step S73: Input the second embedding vector into the trained encoder to obtain the third encoding result; process the third encoding result through the intermediate decoder to obtain the second prediction result.

[0063] In this embodiment, after obtaining the trained encoder and intermediate decoder, a second round of training is performed. In the second round of training, the preprocessed second embedding vector can be input into the trained encoder for processing to obtain the third encoding result output by the trained encoder. Then, the third encoding result is input into the intermediate decoder, and the intermediate decoder processes the third encoding result to obtain the second prediction result output by the intermediate decoder.

[0064] It should be noted that the constructed knowledge graph includes multiple triples. The triples used in the first round of training and the triples used in the second round of training can be the same triples in the knowledge graph or different triples in the knowledge graph. This embodiment does not impose any restrictions on this.

[0065] Step S74: Based on the second prediction result and the remaining entities in the triplet, the network parameters of the intermediate decoder are updated by minimizing the soft edge loss function until the minimizing soft edge loss function converges, so as to obtain the trained decoder, and thus obtain the trained graph attention network composed of the trained encoder and the trained decoder.

[0066] In this embodiment, after obtaining the second prediction result corresponding to the second embedding vector, the second loss value can be calculated by minimizing the soft edge loss function based on the second prediction result and the remaining entities in the triplet corresponding to the second embedding vector. The network parameters of the intermediate decoder are updated by minimizing the soft edge loss function until the convergence of the soft edge loss function. The network parameters of the intermediate decoder are fixed to obtain the trained decoder and complete the second round of training, thereby obtaining a trained graph attention network composed of a trained encoder and a trained decoder.

[0067] In this embodiment, a two-step training process is used to train the encoder-decoder framework in the graph attention network to be trained: in the first step of training, the encoder is trained by minimizing the hinge loss function; in the second step of training, the decoder is trained by minimizing the soft edge loss function, thereby obtaining the trained graph attention network.

[0068] In one embodiment, such as Figure 3 As shown, Figure 3 This is a flowchart illustrating an efficient localization analysis method based on graph attention network knowledge reasoning, as shown in an embodiment of the present invention. Figure 3 First, data collection is performed: tabular data related to the construction of knowledge graphs in the cloud storage domain is exported from the databases of the cloud management server and Cinder server. This data will serve as the foundation for knowledge graph construction, containing key information such as entities, relations, and attributes. Second, data cleaning is performed: the collected data is cleaned, such as removing duplicate fields and invalid data, data formatting, data type conversion, and consistency verification of field names and values. Then, the knowledge graph is constructed and visualized: the cleaned data is first converted into triples, and then the triples are integrated to construct the knowledge graph, which is stored in the graph database platform. Then, the knowledge graph is displayed graphically using the visualization tools provided by the graph database platform. Third, knowledge reasoning based on graph attention networks is performed: knowledge reasoning based on graph attention networks adopts an encoder-decoder framework. The encoder is responsible for converting the input graph structure data (i.e., the embedding vectors of entities and relations) into hidden states, while the decoder generates reasoning results based on these hidden states. By training the network, it can learn the embedding representations of entities and relations that reflect their positions and relationships in the knowledge graph. Finally, knowledge graph completion and prediction results are applied.

[0069] In one embodiment, cloud disk backup is used as an example to further illustrate the process of knowledge graph completion and its predictive effect in practical applications, in the application of knowledge graph completion and prediction results. In the knowledge graph completion stage, the goal is to use existing knowledge relationships in the graph to fill information gaps through logical reasoning. For example, it is known that backup 1 and backup 2 are both associated with cloud disk 1, and they belong to the same repository as a full backup and incremental backup, respectively; meanwhile, backup 3 and backup 4 are associated with cloud disk 2 and cloud disk 3, respectively, and they are both full backups in their corresponding repositories. Based on this known information, a complete knowledge graph about the relationships between cloud disks, backup types, and repositories can be constructed. Figure 4 As shown, Figure 4 This is a partial schematic diagram illustrating knowledge reasoning according to an embodiment of the present invention. Figure 4In this scenario, suppose a backup needs to be created for cloud disk 4 in the repository. Since multiple full backup relationships between cloud disks and the repository already exist in the knowledge graph, and there is no backup information for cloud disk 4, inference can be made using the knowledge relationships in the knowledge graph. Based on the existing patterns in the graph—that backup relationships between cloud disks and the repository are mostly full backups—it can be reasonably inferred that if a backup is created for cloud disk 4, then the backup type is likely also a full backup. This reasoning process not only fills the information gap in the knowledge graph regarding backups of cloud disk 4 but also provides the locator with predictive results about future backup operations. In practical applications, these predictive results can guide the locator to conduct more intelligent backup management, improving the efficiency and accuracy of backup operations.

[0070] In one embodiment, when further expanding the application scenarios of knowledge graphs, combined with actual business logic, the outcomes of specific events can be predicted. Taking cloud disk backup as an example, knowledge graphs can integrate and associate various related information, including the current status of cloud disks, database stability, and the running status of the Cinder-backup service. This information exists in the graph as entities and is interconnected with other entities through relationships.

[0071] When predicting the success rate of backup creation, such as Figure 5 As shown, Figure 5 This is a partial schematic diagram illustrating a successful backup inference according to an embodiment of the present invention. Figure 5 The first step is to confirm the status of these key entities and their related attributes. If the disk status is normal, the repository is running stably, and the Cinder-backup service is also healthy, logical reasoning can be performed based on existing knowledge relationships in the graph, such as common characteristics of successful backup cases. Since past successful backup operations are often accompanied by positive status of these key factors, when these factors are simultaneously positive again, the current backup operation will also be successful. This reasoning process is not only based on statistical regularities but also incorporates the complex relationships between various entities in the graph, thereby improving the accuracy of prediction.

[0072] When predicting the reasons for backup creation failure, such as Figure 6 As shown, Figure 6 This is a partial schematic diagram illustrating a failure details inference method according to an embodiment of the present invention. Figure 6This approach fully leverages the established knowledge relationships within the knowledge graph, particularly information that records backup failure cases and their common characteristics. These cases typically encompass multiple failure causes, such as disk failure, non-existent repositories, network issues, and insufficient permissions. By combining this with the specific context of the current backup task, the most probable cause of the backup failure can be inferred, and the corresponding task details can be completed. For example, inputting Backup 2 and the failure details into a trained graph neural network predicts the failure cause as server shutdown.

[0073] In summary, this invention considers introducing automated analysis tools to assist human decision-making: it utilizes knowledge reasoning technology based on graph attention networks to perform reasoning completion processing on the aggregated knowledge graph. This enhances the completeness and depth of the graph, predicts possible outcomes and corresponding reasons for failure, and provides strong support for problem-solving personnel, helping them quickly locate and deeply analyze potential issues. Through the application of knowledge graph completion and prediction results, this invention can more comprehensively understand and utilize the knowledge relationships within the graph, providing more intelligent decision support for real-world business scenarios.

[0074] It should be noted that, for the sake of simplicity, the method embodiments are all described as a series of actions. However, those skilled in the art should understand that the embodiments of the present invention are not limited to the described order of actions, because according to the embodiments of the present invention, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions involved are not necessarily essential to the embodiments of the present invention.

[0075] Based on the same inventive concept, one embodiment of the present invention provides a problem localization and analysis device. (Reference) Figure 7 , Figure 7 This is a structural block diagram of a problem localization and analysis device provided in an embodiment of the present invention. Figure 7 As shown, the device includes: The building module is used to construct knowledge graphs based on data from multiple databases in a cloud storage system; The training module is used to train the graph attention network to be trained based on the knowledge graph to obtain a trained graph attention network, wherein the trained graph attention network carries the knowledge in the knowledge graph. The acquisition module is used to identify the target task that failed during the testing process of the cloud storage system and to acquire the task information corresponding to the target task. The localization module is used to input the task information corresponding to the target task into the trained graph attention network to obtain the failure reason of the target task.

[0076] Optionally, the knowledge graph includes multiple triples; the task information corresponding to the target task includes at least: the task object of the target task and the failure details of the target task; The positioning module includes: The cause analysis module is used to take the task object of the target task as the first entity, and the failure details of the target task as the first relation as input to the trained graph attention network to obtain the second entity output by the trained graph attention network, and use the second entity as the failure cause of the target task; The first entity and the second entity are two different entities in a triplet.

[0077] Optionally, the trained graph attention network includes: a trained encoder and a trained decoder; The root cause analysis module includes: The first processing module is used to preprocess the first entity and the first relation through the TransE model to obtain the first embedding vector of the first entity and the first relation; The second processing module is used to input the first embedding vector into the trained encoder and convert the first embedding vector into a first encoding result through the trained encoder; The third processing module is used to process the first encoding result through the trained decoder to predict the position and relationship of the first embedding vector in the knowledge graph, and obtain the second entity.

[0078] Optionally, the cloud storage system includes at least: a cloud management server and a Cinder server; the building module includes: The data acquisition module is used to acquire tabular data related to the cloud storage field from the databases of the cloud management server and the Cinder server. The tabular data includes at least: entities, relationships and attributes. The data transformation module is used to convert the tabular data into multiple triples and integrate the multiple triples into a graph database to construct a knowledge graph in the cloud storage domain.

[0079] Optionally, the device further includes: The storage module is used to store the cloud storage domain knowledge graph in the graph database platform; The display module is used to display the knowledge graph of the cloud storage domain in a graphical way using the visualization tools provided by the graph database platform; The query module is used to respond to user query operations and use the Cypher query language to query target entities and / or target relationships in the cloud storage domain knowledge graph.

[0080] Optionally, the knowledge graph includes multiple triples; the training module includes: The fourth processing module is used to preprocess the relations and any entity in the triplet using the TransE model to obtain the second embedding vector of the relations and any entity in the triplet. The model training module is used to train the graph attention network to be trained by minimizing the loss function, using the second embedding vector as a training sample and the remaining entities in the triples as labels, to obtain the trained graph attention network.

[0081] Optionally, the graph attention network to be trained includes: an encoder to be trained and a decoder to be trained; The model training module includes: The fifth processing module is used to input the second embedding vector into the encoder to be trained to obtain a second encoding result; and to process the second encoding result through the decoder to be trained to obtain a first prediction result. The first training module is used to update the network parameters of the encoder and the decoder to be trained based on the first prediction result and the remaining entities in the triplet, using the minimized hinge loss function, until the minimized hinge loss function converges, thereby obtaining the trained encoder and intermediate decoder. The sixth processing module is used to input the second embedding vector into the trained encoder to obtain a third encoding result; and to process the third encoding result through the intermediate decoder to obtain a second prediction result. The second training module is used to update the network parameters of the intermediate decoder based on the second prediction result and the remaining entities in the triplet, using the minimized soft edge loss function, until the minimized soft edge loss function converges to obtain a trained decoder, so as to obtain the trained graph attention network composed of the trained encoder and the trained decoder.

[0082] The terms "first," "second," etc., used in the specification and claims of this invention are used to distinguish similar objects and are not used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the invention can be implemented in orders other than those illustrated or described herein. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0083] The problem location and analysis device in this embodiment of the invention can be a device, or a component, integrated circuit, or chip in a terminal. The device can be a mobile electronic device or a non-mobile electronic device. For example, mobile electronic devices can be mobile phones, tablets, laptops, PDAs, in-vehicle electronic devices, wearable devices, ultra-mobile personal computers (UMPCs), netbooks, or personal digital assistants (PDAs), etc., while non-mobile electronic devices can be servers, network-attached storage (NAS), personal computers (PCs), televisions (TVs), ATMs, or self-service machines, etc. This embodiment of the invention does not impose specific limitations.

[0084] The problem location and analysis device in this embodiment of the invention can be a device with an operating system. This operating system can be Android, iOS, or other possible operating systems; this embodiment of the invention does not impose specific limitations.

[0085] Based on the same inventive concept, another embodiment of the present invention provides an electronic device, such as... Figure 8 As shown, Figure 8 This is a schematic diagram of an electronic device according to an embodiment of the present invention. The electronic device includes a memory, a processor, and a program or instructions stored in the memory and executable on the processor. When the program or instructions are executed by the processor, they implement the steps in the problem localization and analysis method described in any of the above embodiments of the present invention.

[0086] It should be noted that the electronic devices in the embodiments of the present invention include the mobile electronic devices and non-mobile electronic devices described above.

[0087] Based on the same inventive concept, another embodiment of the present invention provides a readable storage medium storing a program or instructions that, when executed by a processor, implement the steps in the problem localization and analysis method described in any of the above embodiments of the present invention. The processor is the processor in the electronic device described in the above embodiments. The readable storage medium includes a computer-readable storage medium, such as a computer read-only memory (ROM), random access memory (RAM), a magnetic disk, or an optical disk.

[0088] As the device embodiment is basically similar to the method embodiment, the description is relatively simple, and relevant parts can be found in the description of the method embodiment.

[0089] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments of the present invention is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

[0090] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0091] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of the present invention.

Claims

1. A problem localization and analysis method, characterized in that, The method includes: A knowledge graph is constructed based on data from multiple databases in a cloud storage system. Based on the knowledge graph, the graph attention network to be trained is trained to obtain a trained graph attention network, which carries the knowledge in the knowledge graph. During the testing of the cloud storage system, the target task that failed is identified, and the task information corresponding to the target task is obtained; The task information corresponding to the target task is input into the trained graph attention network to obtain the reason for the failure of the target task.

2. The problem localization and analysis method according to claim 1, characterized in that, The knowledge graph includes multiple triples; the task information corresponding to the target task includes at least: the task object of the target task and the failure details of the target task; The task information corresponding to the target task is input into the trained graph attention network to obtain the reasons for the failure of the target task, including: The task object of the target task is taken as the first entity, and the failure details of the target task are taken as the first relation and input into the trained graph attention network to obtain the second entity output by the trained graph attention network. The second entity is taken as the failure reason of the target task. The first entity and the second entity are two different entities in a triplet.

3. The problem localization and analysis method according to claim 2, characterized in that, The trained graph attention network includes: a trained encoder and a trained decoder; The task object of the target task is taken as the first entity, and the failure details of the target task are taken as the first relation and input into the trained graph attention network to obtain the second entity output by the trained graph attention network, including: The first entity and the first relation are preprocessed using the TransE model to obtain the first embedding vectors of the first entity and the first relation; The first embedding vector is input into the trained encoder, and the trained encoder converts the first embedding vector into a first encoding result. The trained decoder processes the first encoding result to predict the position and relationship of the first embedding vector in the knowledge graph, thereby obtaining the second entity.

4. The problem localization and analysis method according to claim 1, characterized in that, The cloud storage system includes at least: a cloud management server and a Cinder server; based on data from multiple databases within the cloud storage system, a knowledge graph is constructed, including: Obtain tabular data related to the cloud storage field from the databases of the cloud management server and the Cinder server. The tabular data includes at least: entities, relationships, and attributes. The table data is converted into multiple triples, and the multiple triples are integrated into a graph database to construct a knowledge graph for the cloud storage domain.

5. The problem localization and analysis method according to claim 4, characterized in that, The method further includes: The cloud storage domain knowledge graph is stored in a graph database platform; The visualization tools provided by the graph database platform are used to display the knowledge graph of the cloud storage domain in a graphical manner; In response to user queries, the Cypher query language is used to query target entities and / or target relationships in the cloud storage domain knowledge graph.

6. The problem localization and analysis method according to claim 1, characterized in that, The knowledge graph includes multiple triples; Based on the knowledge graph, the graph attention network to be trained is trained to obtain a trained graph attention network, including: The relations and any entity in the triples are preprocessed using the TransE model to obtain the second embedding vectors of the relations and any entity in the triples. Using the second embedding vector as a training sample and the remaining entities in the triplet as labels, the graph attention network to be trained is trained by minimizing the loss function to obtain the trained graph attention network.

7. The problem localization and analysis method according to claim 6, characterized in that, The graph attention network to be trained includes: an encoder to be trained and a decoder to be trained; Using the second embedding vector as training samples and the remaining entities in the triples as labels, the graph attention network to be trained is trained by minimizing the loss function to obtain the trained graph attention network, including: The second embedding vector is input into the encoder to be trained to obtain a second encoding result; the second encoding result is processed by the decoder to be trained to obtain a first prediction result; Based on the first prediction result and the remaining entities in the triplet, the network parameters of the encoder and the decoder to be trained are updated by minimizing the hinge loss function until the minimization hinge loss function converges, thus obtaining the trained encoder and intermediate decoder. The second embedding vector is input into the trained encoder to obtain a third encoding result; the third encoding result is processed by the intermediate decoder to obtain a second prediction result; Based on the second prediction result and the remaining entities in the triplet, the network parameters of the intermediate decoder are updated by minimizing the soft edge loss function until the minimizing soft edge loss function converges, thus obtaining a trained decoder, which is composed of the trained encoder and the trained decoder.

8. A problem localization and analysis device, characterized in that, The device includes: The building module is used to construct knowledge graphs based on data from multiple databases in a cloud storage system; The training module is used to train the graph attention network to be trained based on the knowledge graph to obtain a trained graph attention network, wherein the trained graph attention network carries the knowledge in the knowledge graph. The acquisition module is used to identify the target task that failed during the testing process of the cloud storage system and to acquire the task information corresponding to the target task. The localization module is used to input the task information corresponding to the target task into the trained graph attention network to obtain the failure reason of the target task.

9. An electronic device, characterized in that, It includes a processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein the program or instructions, when executed by the processor, implement the steps of the problem localization analysis method as described in any one of claims 1 to 7.

10. A readable storage medium, characterized in that, The readable storage medium stores a program or instructions that, when executed by a processor, implement the steps of the problem localization and analysis method as described in any one of claims 1 to 7.

Images