Cloud platform fault repair method, device, host equipment, program product and system
By combining pre-trained language models with preset knowledge bases and operation and maintenance tools, fault detection text is generated and operation and maintenance tools are called for fault repair, which solves the problem of poor fault detection and repair results of cloud platform in the existing technology, and improves the Adaptability and fault repair efficiency of cloud platform operation and maintenance.
Patent Information
- Application Number
- CN202411017309.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-29
- Publication Date
- 2024-11-05
- Estimated Expiration
- 2044-07-29
AI Technical Summary
The existing cloud platform automatic operation and maintenance methods rely on rule engines or basic machine learning, and cannot effectively deal with complex and dynamic cloud platform environments, resulting in poor fault detection and repair results.
The pre-trained language model is used to combine preset knowledge base and operation and maintenance tool information to generate fault detection text for analysis, output fault type and repair task information, and call corresponding operation and maintenance tools for fault repair, ensuring the output reliability of the model through constraint information .
It improves the reliability of cloud platform fault detection and repair and the ability to adapt to complex environments, and improves the adaptability and fault repair efficiency of cloud platform operation and maintenance methods.
Smart Images

Figure CN118550753B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of cloud platform operation and maintenance, and in particular to a cloud platform fault repair method, device, host equipment, program product and medium. Background Art
[0002] With the rapid development of cloud computing technology, enterprises are increasingly relying on cloud platforms to support their businesses and applications. As the scale and complexity of the platform increase, the operation and maintenance of the cloud platform becomes increasingly arduous and complex, and is no longer suitable for heavy manual operation and maintenance. Therefore, automatic operation and maintenance methods for cloud platforms have emerged.
[0003] In the related technologies, the automatic operation and maintenance methods for cloud platforms mainly rely on rule engines or basic machine learning methods for fault detection and repair. However, these methods have certain limitations when dealing with complex and dynamic cloud platform environments and cannot achieve good fault detection and repair effects. Summary of the invention
[0004] The purpose of the present invention is to provide a cloud platform fault repair method, apparatus, host device, program product and medium, which can improve the reliability of cloud platform fault detection and repair by leveraging the text comprehension capability of a pre-trained language model.
[0005] In order to solve the above technical problems, the present invention provides a cloud platform fault repair method, comprising:
[0006] Obtain log alarm information of the cloud platform and operation and maintenance tool information of the operation and maintenance tool, and obtain operation and maintenance knowledge information corresponding to the log alarm information in a preset knowledge base;
[0007] Generate a fault detection text using the operation and maintenance knowledge information, the operation and maintenance tool information, the log alarm information and the preset constraint information, and input the fault detection text into a pre-trained language model, so that the pre-trained language model analyzes the log alarm information according to the operation and maintenance knowledge information and the operation and maintenance tool information under the constraints of the preset constraint information, and outputs the fault type of the cloud platform, the selected operation and maintenance tool information used to repair the fault, and the repair task information;
[0008] The operation and maintenance tool corresponding to the selected operation and maintenance tool information is called to perform fault repair on the cloud platform according to the repair task information.
[0009] Optionally, the preset knowledge base stores the operation and maintenance knowledge representation vector corresponding to each of the operation and maintenance knowledge information;
[0010] Obtaining operation and maintenance knowledge information corresponding to the log alarm information in a preset knowledge base includes:
[0011] Performing vector conversion on the log alarm information to obtain a log alarm representation vector;
[0012] Inputting the log alarm representation vector into the preset knowledge base, so that the preset knowledge base uses the log alarm representation vector to perform vector retrieval on each of the operation and maintenance knowledge representation vectors, and obtains a selected operation and maintenance knowledge representation vector whose vector similarity with the log alarm representation vector is greater than a preset threshold;
[0013] The operation and maintenance knowledge information corresponding to the selected operation and maintenance knowledge representation vector is used as the operation and maintenance knowledge information corresponding to the log alarm information.
[0014] Optionally, the log alarm information includes log information and alarm information;
[0015] The log alarm information is converted into a vector to obtain a log alarm representation vector, including:
[0016] The alarm information is vector-converted to obtain an alarm representation vector, so as to use the alarm representation vector as the log alarm representation vector.
[0017] Optionally, the process of constructing the preset knowledge base includes:
[0018] Obtaining public operation and maintenance knowledge and historical operation and maintenance data of the cloud platform; wherein the historical operation and maintenance data includes historical log alarm information and corresponding fault repair operation information;
[0019] The historical operation and maintenance data is converted into a question-answer pair by using the historical log alarm information as a question and the fault repair operation information as an answer;
[0020] The public operation and maintenance knowledge and the question-answer pair are both set as the operation and maintenance knowledge information, and each of the operation and maintenance knowledge information is vector-converted to obtain an operation and maintenance knowledge representation vector;
[0021] The operation and maintenance knowledge information and the operation and maintenance knowledge representation vector are both added to the preset knowledge base.
[0022] Optionally, obtain log alarm information from the cloud platform, including:
[0023] Obtain all log alarm information generated by the cloud platform within a preset time window.
[0024] Optionally, inputting the fault detection text into a pre-trained language model includes:
[0025] Sending the fault detection text to a model deployment device so that the model deployment device inputs the fault detection text into a pre-trained language model deployed in the model deployment device;
[0026] Receive the fault type, the selected operation and maintenance tool information, and the repair task information returned by the model deployment device.
[0027] Optionally, a fault detection text is generated by using the operation and maintenance knowledge information, the operation and maintenance tool information, the log alarm information and the preset constraint information, and the fault detection text is input into a pre-trained language model, so that the pre-trained language model analyzes the log alarm information according to the operation and maintenance knowledge information and the operation and maintenance tool information under the restriction of the preset constraint information, and outputs the fault type of the cloud platform, the selected operation and maintenance tool information used to repair the fault, and the repair task information, including:
[0028] Generate a fault location text using the operation and maintenance knowledge information, the log alarm information and the first preset constraint information, and input the fault location text into the pre-trained language model, so that the pre-trained language model analyzes the log alarm information according to the operation and maintenance knowledge information under the restriction of the first preset constraint information and outputs the fault type of the cloud platform;
[0029] A fault repair text is generated using the operation and maintenance tool information, the fault type, the log alarm information and the second preset constraint information, and the fault repair text is input into the pre-trained language model, so that the pre-trained language model analyzes the fault type and the log alarm information according to the operation and maintenance tool information under the restriction of the second preset constraint information, and outputs the selected operation and maintenance tool information and the repair task information.
[0030] Optionally, generating a fault location text using the operation and maintenance knowledge information, the log alarm information and the first preset constraint information includes:
[0031] Fill the operation and maintenance knowledge information and the log alarm information into corresponding positions in the first preset template to obtain the fault location text; wherein the first preset template includes the first preset constraint information;
[0032] The generating of the fault repair text by using the operation and maintenance tool information, the fault type, the log warning information and the second preset constraint information includes:
[0033] Querying a second preset template corresponding to the fault type; wherein the second preset template includes the second preset constraint information;
[0034] Fill the operation and maintenance tool information and the log alarm information into corresponding positions in the second preset template to obtain the fault repair text.
[0035] Optionally, the first preset constraint information includes:
[0036] A first constraint text is used to restrict the pre-trained language model to analyze the log alarm information only according to the operation and maintenance knowledge information;
[0037] A second constraint text is used to limit the output format used by the pre-trained language model to output the fault type;
[0038] A third constraint text is used to specify the text content output by the pre-trained language model when the fault type of the cloud platform cannot be determined;
[0039] The second preset constraint information includes:
[0040] A fourth constraint text is used to restrict the pre-trained language model to analyze the log alarm information only according to the operation and maintenance tool information;
[0041] A fifth constraint text, used to limit the output format used by the pre-trained language model to output the selected operation and maintenance tool information and the repair task information;
[0042] The sixth constraint text is used to specify the text content output by the pre-trained language model when the selected operation and maintenance tool information and the repair task information cannot be determined.
[0043] Optionally, after inputting the fault location text into the pre-trained language model, the method further includes:
[0044] When it is determined that the pre-trained language model outputs the third constraint text, the log alarm information is saved to a preset manual processing database so that operation and maintenance personnel can manually process the fault corresponding to the log alarm information;
[0045] After inputting the fault repair text into the pre-trained language model, the method further includes:
[0046] When it is determined that the pre-trained language model outputs the sixth constraint text, the log alarm information is saved to a preset manual processing database so that operation and maintenance personnel can manually process the fault corresponding to the log alarm information.
[0047] Optionally, it also includes:
[0048] When it is determined that the log alarm information recorded in the manual processing database has been added with corresponding fault repair operation information, the preset knowledge base is updated using the log alarm information and the corresponding fault repair operation information.
[0049] Optionally, before calling the operation and maintenance tool corresponding to the selected operation and maintenance tool information to perform fault repair on the cloud platform according to the repair task information, the method further includes:
[0050] According to the output format specified by the fifth constraint text, the selected operation and maintenance tool information and the repair task information are parsed from the text content output by the pre-trained language model.
[0051] Optionally, after calling the operation and maintenance tool corresponding to the selected operation and maintenance tool information to perform fault repair on the cloud platform according to the repair task information, the method further includes:
[0052] Determine whether the repair task is successfully executed;
[0053] If the repair task is successfully executed, the log alarm information, the selected operation and maintenance tool information, and the repair task information are integrated into new operation and maintenance knowledge information, and the preset knowledge base is updated using the new operation and maintenance knowledge information;
[0054] If the repair task is not successfully executed, task adjustment information is generated, and the task adjustment information is input into the pre-trained language model, so that the pre-trained language model analyzes the fault type, the log alarm information and the task adjustment information according to the operation and maintenance tool information, and outputs updated selected operation and maintenance tool information and updated repair task information;
[0055] The operation and maintenance tool corresponding to the updated selected operation and maintenance tool information is called to perform fault repair on the cloud platform according to the updated repair task information.
[0056] The present invention also provides a cloud platform fault repair device, comprising:
[0057] An acquisition module is used to acquire log alarm information of the cloud platform and operation and maintenance tool information of the operation and maintenance tool, and acquire operation and maintenance knowledge information corresponding to the log alarm information in a preset knowledge base;
[0058] A fault detection module is used to generate a fault detection text using the operation and maintenance knowledge information, the operation and maintenance tool information, the log alarm information and the preset constraint information, and input the fault detection text into a pre-trained language model, so that the pre-trained language model analyzes the log alarm information according to the operation and maintenance knowledge information and the operation and maintenance tool information under the restriction of the preset constraint information, and outputs the fault type of the cloud platform, the selected operation and maintenance tool information used to repair the fault, and the repair task information;
[0059] A fault repair module is used to call the operation and maintenance tool corresponding to the selected operation and maintenance tool information to perform fault repair on the cloud platform according to the repair task information.
[0060] The present invention also provides a host device, comprising:
[0061] Memory for storing computer programs;
[0062] A processor is used to implement the cloud platform fault repair method as described above when executing the computer program.
[0063] The present invention also provides a cloud platform fault recovery system, comprising:
[0064] A host device, used to execute the cloud platform fault repair method as described above;
[0065] A model deployment device, in which a pre-trained language model is deployed, is used to receive a fault detection text sent by the host device, input the fault detection text into the pre-trained language model, and send the fault type, selected operation and maintenance tool information, and repair task information output by the pre-trained language model to the host device.
[0066] The present invention also provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the cloud platform fault repair method as described above.
[0067] The present invention also provides a computer-readable storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are loaded and executed by a processor, the cloud platform fault repair method as described above is implemented.
[0068] The present invention provides a cloud platform fault repair method, comprising: obtaining log alarm information of the cloud platform and operation and maintenance tool information of an operation and maintenance tool, and obtaining operation and maintenance knowledge information corresponding to the log alarm information in a preset knowledge base; generating a fault detection text by using the operation and maintenance knowledge information, the operation and maintenance tool information, the log alarm information and preset constraint information, and inputting the fault detection text into a pre-trained language model, so that the pre-trained language model analyzes the log alarm information according to the operation and maintenance knowledge information and the operation and maintenance tool information under the restriction of the preset constraint information, outputs the fault type of the cloud platform, selected operation and maintenance tool information and repair task information used to repair the fault; and calling the operation and maintenance tool corresponding to the selected operation and maintenance tool information to repair the cloud platform according to the repair task information.
[0069] The beneficial effect of the present invention is that the present invention can first obtain the log alarm information of the cloud platform and the operation and maintenance tool information of the operation and maintenance tool, and can also obtain the operation and maintenance knowledge information corresponding to the log alarm information in the preset knowledge base. Subsequently, the present invention can generate a fault detection text using the operation and maintenance knowledge information, the operation and maintenance tool information, the log alarm information and the preset constraint information, and input the fault detection text into the pre-trained language model, so that the pre-trained language model analyzes the log alarm information according to the operation and maintenance knowledge information and the operation and maintenance tool information under the restriction of the preset constraint information, and outputs the fault type of the cloud platform, the selected operation and maintenance tool information used to repair the fault, and the repair task information. In other words, the present invention can use the text understanding ability of the pre-trained language model to input the above-mentioned operation and maintenance knowledge information, operation and maintenance tool information, and log alarm information into the pre-trained language model, so that the control pre-trained language model automatically analyzes the log alarm information according to the operation and maintenance knowledge information and the operation and maintenance tool information, and outputs the fault type of the cloud platform, the selected operation and maintenance tool information used to repair the fault, and the repair task information; in addition, in order to ensure that the pre-trained language model strictly analyzes the log alarm information according to the given operation and maintenance knowledge information and the operation and maintenance tool information, the present invention can also add preset constraint information in the fault detection text to impose restrictions on the pre-trained language model, so as to ensure the reliability of using the pre-trained language model to detect cloud platform faults and formulate repair tasks. Furthermore, the present invention only needs to call the operation and maintenance tool corresponding to the selected operation and maintenance tool information according to the output of the pre-trained language model to repair the cloud platform according to the repair task information, thereby improving the reliability of cloud platform fault repair and improving the adaptability of the cloud platform operation and maintenance method to the complex cloud platform environment. The present invention also provides a cloud platform fault repair device, a host device, a computer program product, and a computer-readable storage medium, which have the above-mentioned beneficial effects. BRIEF DESCRIPTION OF THE DRAWINGS
[0070] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.
[0071] Figure 1 A flow chart of a cloud platform fault repair method provided by an embodiment of the present invention;
[0072] Figure 2 A schematic diagram of a knowledge base construction process provided by an embodiment of the present invention;
[0073] Figure 3 A schematic diagram of an information retrieval and fault diagnosis process provided by an embodiment of the present invention;
[0074] Figure 4 A schematic diagram of a task formulation and execution process provided by an embodiment of the present invention;
[0075] Figure 5 A schematic diagram of a cloud platform fault recovery system provided by an embodiment of the present invention;
[0076] Figure 6 A structural block diagram of a cloud platform fault repair device provided by an embodiment of the present invention;
[0077] Figure 7 A structural block diagram of a host device provided by an embodiment of the present invention;
[0078] Figure 8 A structural block diagram of a cloud platform fault recovery system provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0079] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0080] With the rapid development of cloud computing technology, enterprises are increasingly relying on cloud platforms to support their businesses and applications. As the scale and complexity of the platform increase, the operation and maintenance of the cloud platform has become increasingly arduous and complex, and it is no longer suitable to rely heavily on manual operation and maintenance. Therefore, automatic operation and maintenance methods for cloud platforms have emerged. In the related art, the automatic operation and maintenance methods for cloud platforms mainly rely on rule engines or basic machine learning methods for fault detection and repair. However, these methods have certain limitations when dealing with complex and dynamic cloud platform environments, and cannot achieve better fault detection and repair effects. In view of this, the present invention can use the text understanding ability of the pre-trained language model to improve the reliability of cloud platform fault detection and repair, and can improve the adaptability of the cloud platform operation and maintenance method to the complex cloud platform environment.
[0081] For ease of understanding, the following will first briefly introduce the pre-trained language model. The pre-trained language model can be a large language model (LLM), which refers to a machine learning model trained with a large amount of text data that can generate natural language text or understand the meaning of language text. Since the large language model is significantly different from the general language model in terms of model scale (such as the number of parameters), amount of training data, and complexity, it has significant performance advantages in semantic understanding and text generation. Common open source large language models include: ChatGLM-6B (Chat General Language Model with 6 Billion parameters, a conversational general language model with 6 billion parameters), GLM-130B (General Language Model with 130 Billion parameters, a conversational general language model with 130 billion parameters), etc.
[0082] In order to meet the question-answering needs in a specific field, the relevant technology can usually use the knowledge of the specific field to fine-tune the pre-trained language model to ensure that the model fully learns the knowledge of the specific field and embeds the knowledge of the specific field into its own model parameters. In this way, when the model receives questions in the specific field, it can understand the user's intention based on the learned knowledge and generate corresponding answers for the user. For example, the pre-trained language model can be fine-tuned using the cloud platform operation and maintenance domain knowledge so that when the model receives cloud platform operation and maintenance questions, it can answer based on the learned knowledge. However, the above-mentioned fine-tuning training scheme has poor adaptability to cloud platform operation and maintenance scenarios. On the one hand, it may be difficult for the pre-trained language model to stably output the correct answer to the user based on the learned knowledge. For example, since the question answering ability of the pre-trained language model is closely related to its ability to master knowledge, if its mastery of knowledge is poor, it will not be able to stably output the correct answer when facing questions raised by users, and may even output the wrong answer. For another example, in order to enhance the generalization ability, the pre-trained language model usually needs to learn knowledge from multiple fields. If it refers to knowledge from other unrelated fields when answering cloud platform operation and maintenance questions, it may also cause it to output wrong answers. In the cloud platform operation and maintenance scenario, it is usually necessary to ensure that the detected cloud platform fault information and the cloud platform fault repair plan are reliable and effective, otherwise it will directly affect the cloud platform operation and maintenance effect, and may even affect the normal operation of the cloud platform. On the other hand, it is difficult to collect enough operation and maintenance knowledge for dynamic and complex cloud platform environments, which will also interfere with the fine-tuning training effect of the pre-trained language model, and thus cause the model to be unable to correctly answer cloud platform operation and maintenance questions. Therefore, using cloud platform operation and maintenance knowledge to fine-tune the pre-trained language model not only fails to give full play to the advantages of the pre-trained language model, but may even interfere with the normal operation of the cloud platform, and has poor adaptability to cloud platform operation and maintenance scenarios.
[0083] In view of this, the embodiment of the present invention will not adopt the above-mentioned fine-tuning training scheme, but will improve the accuracy, professionalism and stability of the answers to questions by limiting the scope of knowledge that the pre-trained language model can refer to when answering cloud platform operation and maintenance questions. Specifically, before each cloud platform operation and maintenance question is input into the pre-trained language model, the embodiment of the present invention will pre-search for operation and maintenance knowledge information related to the cloud platform operation and maintenance question as known information, and provide the known information and the cloud platform operation and maintenance question to the pre-trained language model at the same time. In addition, the embodiment of the present invention can also input preset constraint information into the pre-trained language model (for example, input "please answer based on known information" into the pre-trained language model) so that the pre-trained language model answers cloud platform operation and maintenance questions only based on the known information. In this way, the embodiment of the present invention can strictly control the scope of knowledge referenced by the pre-trained language model within the scope of known information input for this question, and only rely on the powerful understanding and analysis ability of the pre-trained language model for complex texts to avoid using the knowledge that the pre-trained language model has learned, thereby avoiding the pre-trained language model from referring to knowledge in other irrelevant fields when answering questions, and can improve the accuracy, professionalism and stability of the pre-trained language model in answering cloud platform operation and maintenance issues. At the same time, it can effectively exert the analysis and understanding ability of the pre-trained language model for complex semantic logic, making it more adaptable to dynamic and complex cloud platform environments, and thus effectively exert the advantages of the pre-trained language model in cloud platform operation and maintenance scenarios. In addition, since the embodiment of the present invention does not require fine-tuning training of the pre-trained language model, and does not require baseline testing of the pre-trained language model after fine-tuning training (i.e., testing the accuracy of the answers output by the pre-trained language model), it can also reduce the deployment cost and difficulty of the cloud platform operation and maintenance solution.
[0084] Based on the above introduction to the pre-trained language model, the cloud platform fault repair method provided by the embodiment of the present invention will be introduced below. It should be noted that the embodiment of the present invention does not limit the hardware device for executing the present method, and can be set according to actual application requirements, such as a personal computer, a server, etc. In addition, the present method can also be executed by multiple hardware devices. For example, the cloud platform fault repair system may include a host device and a model deployment device, the host device is responsible for collecting the cloud platform operation and maintenance problems input by the user, and sending the cloud platform operation and maintenance problems to the model deployment device, and is also responsible for feeding back the cloud platform operation and maintenance answers returned by the model deployment device to the user; the model deployment device is provided with the above-mentioned pre-trained language model, which is responsible for receiving the cloud platform operation and maintenance problems sent by the host device, and passing the problems into the pre-trained language model, and is also responsible for returning the cloud platform operation and maintenance answers generated by the pre-trained language model to the host device. It should be noted that the embodiment of the present invention does not limit the communication method between the host device and the model deployment device, such as network communication. The embodiment of the present invention also does not limit the specific hardware types of the host device and the model deployment device, and can be set according to actual application requirements. In addition, in a distributed computing scenario, there may also be multiple model deployment devices.
[0085] For easier understanding, please refer to Figure 1 , Figure 1 A flowchart of a cloud platform fault repair method provided by an embodiment of the present invention, the method may include:
[0086] S101, obtaining log alarm information of a cloud platform and operation and maintenance tool information of an operation and maintenance tool, and obtaining operation and maintenance knowledge information corresponding to the log alarm information in a preset knowledge base.
[0087] In this step, the obtained log alarm information will be used as part of the cloud platform operation and maintenance problem, while the operation and maintenance tool information and operation and maintenance knowledge information will be used as known information. It is worth noting that in this embodiment, a preset knowledge base is provided, which can save all the operation and maintenance knowledge information required for the cloud platform operation and maintenance, such as the instructions for use of each device in the cloud platform, troubleshooting and repair instructions, and can also include the historical log alarm information, fault type, operation and maintenance tools used for repair, and repair task information corresponding to the cloud platform when it encountered a fault before. It is also worth noting that the embodiment of the present invention does not send the full amount of operation and maintenance knowledge information in the preset knowledge base to the pre-trained language model, but can query the operation and maintenance knowledge information corresponding to the current log alarm information in the preset knowledge base, and only send the operation and maintenance knowledge information corresponding to the log alarm information to the pre-trained language model. In this way, the embodiment of the present invention can control the length of the input text, and at the same time improve the correlation between the operation and maintenance knowledge information and the log alarm information, thereby further improving the accuracy of the answer.
[0088] It should be noted that the embodiments of the present invention do not limit the method for obtaining log alarm information, and can be set according to actual application requirements. In addition, the log alarm information can be divided into log information and alarm information, the log information is used to record the operating status of the cloud platform, and the alarm information is used to record the abnormal conditions encountered by the cloud platform. In addition, in order to improve the detection accuracy, the embodiments of the present invention can also obtain multiple log alarm information. For example, all log alarm information generated by the cloud platform within a preset time window can be obtained, so that the pre-trained language model can analyze the faults encountered by the cloud platform within the preset time window. The embodiments of the present invention also do not limit the specific size of the preset time window range, and can be set according to actual application requirements.
[0089] Based on this, the log alarm information of the cloud platform can be obtained, including:
[0090] Step 11: Obtain all log alarm information generated by the cloud platform within a preset time window.
[0091] Furthermore, the embodiments of the present invention do not limit specific operation and maintenance tools, for example, they may be monitoring tools, automated scripts, etc.; the embodiments of the present invention do not limit the content and form of the operation and maintenance tool information, for example, the operation and maintenance tool information may include tool identification, tool name, tool execution mode, execution parameters, and tool functions, etc., which may be set according to actual application requirements. For ease of understanding, the following is a possible storage format for operation and maintenance tool information:
[0092] {
[0093] "ID": "The unique ID of the tool",
[0094] "Tool Name": "Tool Name",
[0095] "Execution Method": "Execution method",
[0096] "Execution Parameters": "Execution parameters",
[0097] "Tool Description": "Tool Function"
[0098] }
[0099] Furthermore, the embodiments of the present invention do not limit how to obtain the operation and maintenance knowledge information corresponding to the log alarm information in the preset knowledge base. For example, based on the similarity between the log alarm information and the operation and maintenance knowledge information, the operation and maintenance knowledge information that is close to the log alarm information can be found. The above similarity can be text similarity or vector similarity, which can be set according to actual application requirements, and the calculation method of text similarity and vector similarity can refer to related technologies. It should be pointed out that for vector similarity, both the log alarm information and the operation and maintenance knowledge information need to be converted into representation vector forms, and then the vector similarity between the log alarm representation vector and the operation and maintenance knowledge representation vector is calculated.
[0100] S102. Generate a fault detection text using the operation and maintenance knowledge information, the operation and maintenance tool information, the log alarm information and the preset constraint information, and input the fault detection text into a pre-trained language model, so that the pre-trained language model analyzes the log alarm information according to the operation and maintenance knowledge information and the operation and maintenance tool information under the restriction of the preset constraint information, and outputs the fault type of the cloud platform, the selected operation and maintenance tool information used to repair the fault, and the repair task information.
[0101] In this step, the operation and maintenance knowledge information, operation and maintenance tool information, log alarm information and preset constraint information can be used to generate a fault detection text, and the fault detection text can be input into the pre-trained language model, so that the pre-trained language model can analyze the log alarm information according to the operation and maintenance knowledge information and the operation and maintenance tool information under the constraints of the preset constraint information, and output the cloud platform fault type, the selected operation and maintenance tool information used to repair the fault, and the repair task information. Among them, the above-mentioned operation and maintenance knowledge information and operation and maintenance tool information will be used as known information, and the log alarm information can be used as the cloud platform operation and maintenance problem part in the fault detection text. The fault type can be one of several fault types specified in the operation and maintenance knowledge information. The selected operation and maintenance tool information is the operation and maintenance tool information corresponding to the operation and maintenance tool selected to repair the cloud platform. The repair task information can include both the execution parameters to be used by the operation and maintenance tool and the operation steps to be performed by the operation and maintenance tool.
[0102] It is worth noting that the embodiment of the present invention also adds preset constraint information to the fault detection text, and the preset constraint information is at least used to limit the pre-trained language model to analyze the log alarm information only based on the operation and maintenance knowledge information and the operation and maintenance tool information. For example, the preset constraint information at least includes the text content "Please answer based on the known information", so that the pre-trained language model can answer questions only based on the known information entered by the user (i.e., operation and maintenance knowledge information, operation and maintenance tool information) after parsing the text content. In this way, the present invention can make use of the powerful understanding and analysis capabilities of the pre-trained language model for complex texts, and can avoid the pre-trained language model from referring to knowledge in other unrelated fields when answering questions, thereby improving the accuracy, professionalism and stability of the pre-trained language model in answering cloud platform operation and maintenance problems, and thus can effectively give play to the advantages of the pre-trained language model in cloud platform operation and maintenance scenarios.
[0103] Of course, the above-mentioned preset constraint information may also include other content to meet other application requirements. For example, the preset constraint information may also include format constraint text to limit the format of the answer output by the pre-trained language model; it may also include specified output text to limit the answer text to be output by the pre-trained language model when it cannot answer the question based on the known information; it may also include role constraint text to limit the role played by the pre-trained language model when analyzing the problem.
[0104] Furthermore, it should be noted that the embodiments of the present invention do not limit the specific form of the above-mentioned fault detection text. It can also be composed of multiple parts of text, as long as it can ensure that the pre-trained language model analyzes the log alarm information according to the operation and maintenance knowledge information and operation and maintenance tool information under the constraints of preset constraint information, outputs the fault type of the cloud platform, the selected operation and maintenance tool information and the repair task information used to repair the fault, and can be set according to actual application needs.
[0105] S103: Calling the operation and maintenance tool corresponding to the selected operation and maintenance tool information to perform fault repair on the cloud platform according to the repair task information.
[0106] In this step, after obtaining the selected operation and maintenance tool information and repair task information output by the pre-trained language model, it is only necessary to call the operation and maintenance tool corresponding to the selected operation and maintenance tool information to repair the cloud platform according to the repair task information, thereby realizing automatic operation and maintenance of the cloud platform.
[0107] Of course, when controlling the operation and maintenance tool to perform the repair task, the execution status of the repair task can also be automatically detected to perform corresponding processing according to the different execution status of the task. For example, if the repair task is determined to be successfully executed, the log alarm information, selected operation and maintenance tool information, and repair task information of this time can be integrated into new operation and maintenance knowledge information, and the preset knowledge base can be updated using the new operation and maintenance knowledge information. If it is determined that the repair task is not successfully executed, task adjustment information can also be generated, and the task adjustment information can be input into the pre-trained language model, so that the pre-trained language model re-analyzes the fault type, log alarm information and task adjustment information according to the operation and maintenance tool information, outputs the updated selected operation and maintenance tool information and the updated repair task information, so that the operation and maintenance tool corresponding to the re-updated selected operation and maintenance tool information performs fault repair on the cloud platform according to the updated repair task information, so as to automatically try different repair methods, thereby improving the reliability of cloud platform repair.
[0108] Based on this, after calling the operation and maintenance tool corresponding to the selected operation and maintenance tool information to perform fault repair on the cloud platform according to the repair task information, the following may also be included:
[0109] Step 21: Determine whether the repair task is successfully executed; if the repair task is successfully executed, proceed to step 22; if the repair task is not successfully executed, proceed to step 23.
[0110] Step 22: Integrate the log alarm information, the selected operation and maintenance tool information, and the repair task information into new operation and maintenance knowledge information, and use the new operation and maintenance knowledge information to update the preset knowledge base.
[0111] Step 23: Generate task adjustment information and input the task adjustment information into the pre-trained language model, so that the pre-trained language model analyzes the fault type, the log alarm information and the task adjustment information according to the operation and maintenance tool information, and outputs updated selected operation and maintenance tool information and updated repair task information.
[0112] Step 24: Calling the operation and maintenance tool corresponding to the updated selected operation and maintenance tool information to perform fault repair on the cloud platform according to the updated repair task information.
[0113] Based on the above embodiments, the present invention can first obtain the log alarm information of the cloud platform and the operation and maintenance tool information of the operation and maintenance tool, and can also obtain the operation and maintenance knowledge information corresponding to the log alarm information in the preset knowledge base. Subsequently, the present invention can generate a fault detection text using the operation and maintenance knowledge information, the operation and maintenance tool information, the log alarm information and the preset constraint information, and input the fault detection text into the pre-trained language model, so that the pre-trained language model analyzes the log alarm information according to the operation and maintenance knowledge information and the operation and maintenance tool information under the constraints of the preset constraint information, and outputs the fault type of the cloud platform, the selected operation and maintenance tool information used to repair the fault, and the repair task information. In other words, the present invention can use the text understanding ability of the pre-trained language model to input the above-mentioned operation and maintenance knowledge information, operation and maintenance tool information, and log alarm information into the pre-trained language model, so that the control pre-trained language model automatically analyzes the log alarm information according to the operation and maintenance knowledge information and the operation and maintenance tool information, and outputs the cloud platform fault type, the selected operation and maintenance tool information used to repair the fault, and the repair task information; in addition, in order to ensure that the pre-trained language model strictly analyzes the log alarm information according to the given operation and maintenance knowledge information and the operation and maintenance tool information, the present invention can also add preset constraint information to the fault detection text to impose restrictions on the pre-trained language model, so as to ensure the reliability of using the pre-trained language model to detect cloud platform faults and formulate repair tasks. Furthermore, the present invention only needs to call the operation and maintenance tool corresponding to the selected operation and maintenance tool information according to the output of the pre-trained language model to repair the cloud platform according to the repair task information, thereby improving the reliability of cloud platform fault repair and improving the adaptability of the cloud platform operation and maintenance method to the complex cloud platform environment.
[0114] Based on the above embodiments, the process of using a pre-trained language model for automatic operation and maintenance of a cloud platform will be further introduced below based on a possible form of fault detection text. In one possible scenario, the fault detection text can be divided into fault location text and fault repair text. The fault location text is used to analyze the fault type of the cloud platform from the log alarm information, and may include operation and maintenance knowledge information and log alarm information; the fault repair text is used to analyze the selected operation and maintenance tools and repair task information used to repair the cloud platform fault based on the fault type and log alarm information, and may include operation and maintenance tool information, fault type and log alarm information. In one possible scenario, the method may also include:
[0115] S201. Obtain log alarm information of a cloud platform and operation and maintenance tool information of an operation and maintenance tool, and obtain operation and maintenance knowledge information corresponding to the log alarm information from a preset knowledge base.
[0116] The relevant description of this step is consistent with the description of step S101 and will not be repeated here.
[0117] S202. Generate a fault location text using the operation and maintenance knowledge information, the log alarm information and the first preset constraint information, and input the fault location text into the pre-trained language model, so that the pre-trained language model analyzes the log alarm information according to the operation and maintenance knowledge information under the restriction of the first preset constraint information and outputs the fault type of the cloud platform.
[0118] In this step, the operation and maintenance knowledge information, log alarm information and the first preset constraint information are first used to generate a fault location text, and the fault location text is input into the pre-trained language model, so that the pre-trained language model analyzes the log alarm information according to the operation and maintenance knowledge information under the restriction of the first preset constraint information and outputs the fault type of the cloud platform. Among them, the first preset constraint information is at least used to limit the pre-trained language model to analyze the log alarm information only according to the operation and maintenance knowledge information. Of course, the first preset constraint information may also contain other content. For example, the first preset constraint information may include:
[0119] The first constraint text is used to limit the pre-trained language model to analyze the log alarm information only according to the operation and maintenance knowledge information.
[0120] The second constraint text is used to limit the output format used by the pre-trained language model to output the fault type.
[0121] The third constraint text is used to specify the text content output by the pre-trained language model when the fault type of the cloud platform cannot be determined.
[0122] After the above three constraint texts are set, the pre-trained language model can analyze the log alarm information only according to the operation and maintenance knowledge information under the restriction of the first constraint text, and can output the fault type of the cloud platform according to the specified output format under the restriction of the second constraint text, and can output the third constraint text when it is determined that the current operation and maintenance knowledge information is insufficient to determine the fault type corresponding to the log alarm information under the restriction of the third constraint text. Furthermore, when the embodiment of the present invention obtains the fault type in the standard format output by the pre-trained language model, it can continue to step S203; when the third constraint text output by the pre-trained language model is obtained, the log alarm information of this time can be input into the preset manual processing database (long-term memory library) so that the operation and maintenance personnel can manually process the fault corresponding to the log alarm information.
[0123] Based on this, after inputting the fault location text into the pre-trained language model, the method may further include:
[0124] Step 31: When it is determined that the pre-trained language model outputs the third constraint text, save the log alarm information to a preset manual processing database so that the operation and maintenance personnel can manually process the faults corresponding to the log alarm information.
[0125] Furthermore, to facilitate the generation of standardized fault location texts, an embodiment of the present invention can also set a first preset template, which includes the above first preset constraint information, and corresponding filling positions can be set for the operation and maintenance knowledge information and the log alarm information. Furthermore, it is only necessary to fill the operation and maintenance knowledge information and the log alarm information into the corresponding positions in the first preset template.
[0126] Based on this, generating a fault location text using the operation and maintenance knowledge information, the log alarm information, and the first preset constraint information may include:
[0127] Step 41: Fill the operation and maintenance knowledge information and the log alarm information into the corresponding positions in the first preset template to obtain the fault location text; wherein, the first preset template includes the first preset constraint information;
[0128] For ease of understanding, the following is a specific form of a first preset template:
[0129] <Instruction>You are a senior software engineer. Based on the known information, answer the question concisely and professionally. If you cannot get the answer from it, say 'The question cannot be answered based on the known information', and do not allow adding fabricated content to the answer. The answer should be in Chinese.< / Instruction> / / ;
[0130] <Known Information>{{ context}}< / Known Information> / / ;
[0131] <Question>Log information is as follows {{log}}, alarm information is as follows {{alert}}. Is there a fault in the current cloud platform? If so, please give the fault information, analyze the cause of the fault, and give repair suggestions.< / Question> / / ;
[0132] Among them, "Based on the known information, answer the question concisely and professionally" and "Do not allow adding fabricated content to the answer" are the first constraint texts; "If so, please give the fault information, analyze the cause of the fault, and give repair suggestions" is the second constraint text; "If you cannot get the answer from it, say 'The question cannot be answered based on the known information'" is the third constraint text. context is the filling position of the operation and maintenance knowledge information, log is the filling position of the log information, and alert is the filling position of the alarm information.
[0133] S203. Generate a fault repair text using the operation and maintenance tool information, the fault type, the log alarm information and the second preset constraint information, and input the fault repair text into the pre-trained language model, so that the pre-trained language model analyzes the fault type and the log alarm information according to the operation and maintenance tool information under the restriction of the second preset constraint information, and outputs the selected operation and maintenance tool information and the repair task information.
[0134] In this step, the operation and maintenance tool information, fault type, log alarm information and the second preset constraint information will be further used to generate a fault repair text, and the fault location text will be input into the pre-trained language model, so that the pre-trained language model analyzes the fault type and log alarm information according to the operation and maintenance tool information under the restriction of the second preset constraint information, and outputs the selected operation and maintenance tool information and repair task information. Among them, the second preset constraint information is at least used to limit the pre-trained language model to analyze the log alarm information only according to the operation and maintenance tool information. Of course, the second preset constraint information may also contain other content. For example, the second preset constraint information may include:
[0135] A fourth constraint text is used to restrict the pre-trained language model to analyze the log alarm information only according to the operation and maintenance tool information;
[0136] A fifth constraint text, used to limit the output format used by the pre-trained language model to output the selected operation and maintenance tool information and the repair task information;
[0137] The sixth constraint text is used to specify the text content output by the pre-trained language model when the selected operation and maintenance tool information and the repair task information cannot be determined.
[0138] After the above three constraint texts are set, the pre-trained language model can analyze the log alarm information only according to the operation and maintenance tool information under the restriction of the fourth constraint text, and can output the selected operation and maintenance tool information and repair task information according to the specified output format under the restriction of the fifth constraint text, and can output the sixth constraint text when it is determined that the current operation and maintenance tool information is insufficient to generate the selected operation and maintenance tool information and repair task information under the restriction of the sixth constraint text. Furthermore, when the embodiment of the present invention obtains the selected operation and maintenance tool information and repair task information in the standard format output by the pre-trained language model, it can continue to step S204; when the sixth constraint text output by the pre-trained language model is obtained, the log alarm information of this time can be input into the preset manual processing database (long-term memory library) so that the operation and maintenance personnel can manually process the fault corresponding to the log alarm information.
[0139] Based on this, after inputting the fault repair text into the pre-trained language model, the method may further include:
[0140] Step 51: When it is determined that the pre-trained language model outputs the sixth constraint text, save the log alarm information to a preset manual processing database so that the operation and maintenance personnel can manually process the faults corresponding to the log alarm information.
[0141] Of course, if it is detected that the fault repair operation information has been added to the log alarm information in the preset manual processing database, that is, the fault corresponding to the log alarm information has been manually investigated and repaired, the preset knowledge base can be updated by using the log alarm information and the corresponding fault repair operation information to facilitate subsequent automatic operation and maintenance.
[0142] Based on this, the method may further include:
[0143] Step 61: When it is determined that the fault repair operation information has been added to the log alarm information recorded in the manual processing database, update the preset knowledge base by using the log alarm information and the corresponding fault repair operation information.
[0144] Further, to facilitate the generation of standardized fault location texts, the embodiment of the present invention may further set a second preset template, which corresponds to the fault type, may include the above second preset constraint information, and may set corresponding filling positions for the operation and maintenance tool information and the log alarm information. Furthermore, only need to query the corresponding second preset template according to the fault type and fill the operation and maintenance knowledge information and the log alarm information into the corresponding positions in the first preset template.
[0145] Based on this, the generation of the fault repair text by using the operation and maintenance tool information, the fault type, the log alarm information, and the second preset constraint information may include:
[0146] Step 71: Query the second preset template corresponding to the fault type; wherein, the second preset template includes the second preset constraint information;
[0147] Step 72: Fill the operation and maintenance tool information and the log alarm information into the corresponding positions in the second preset template to obtain the fault repair text.
[0148] For easy understanding, the following is a specific form of a second preset template:
[0149] <Instruction>You are a senior software engineer. According to the existing tools in the tool list, formulate a repair plan for the fault. If you cannot specify a repair plan, say "The problem cannot be repaired with the existing tools". If you can output the execution commands of the tools, including the execution method and actual parameters, which are required to be directly executable in the Linux system.< / Instruction> / / ;
[0150] <Tool list>{{ tools}}< / Tool list> / / ;
[0151] <Problem>The log information is as follows {{log}}, and the alert information is as follows {{alert}}. Please formulate a repair plan for the fault based on the above content< / Problem> / / ;
[0152] Among them, "formulate a repair plan for the fault based on the existing tools" and "it is not allowed to add fabricated content to the answer" are the fourth constraint texts; "if the execution commands of the tools can be output, including the execution method and actual parameters, it is required to be directly executable in the Linux system" is the fifth constraint text; "if the repair plan cannot be specified, please say 'The problem cannot be repaired using the existing tools'" is the sixth constraint text. tools is the filling position of the operation and maintenance tool information, log is the filling position of the log information, and alert is the filling position of the alert information.
[0153] Furthermore, since the output format of the pre-trained language model for outputting the selected operation and maintenance tool information and repair task information has been pre-constrained, the output of the pre-trained language model can be directly parsed according to the specified output format to obtain the selected operation and maintenance tool information and the repair task information.
[0154] Based on this, before calling the operation and maintenance tool corresponding to the selected operation and maintenance tool information to repair the fault of the cloud platform according to the repair task information, it may further include:
[0155] Step 81: Parse the selected operation and maintenance tool information and the repair task information from the text content output by the pre-trained language model according to the output format specified by the fifth constraint text.
[0156] For easy understanding, the following is a parsing method for the selected operation and maintenance tool information and repair task information:
[0157] execution_plan = model.get_execution_plan() / / Get the execution plan, and the execution_plan contains the selected operation and maintenance tool information and repair task information;
[0158] / / Select the tool and perform diagnosis or repair;
[0159] tool = select_tool(execution_plan['tool']) / / Parse the selected operation and maintenance tool information from the tool field of execution_plan;
[0160] if execution_plan['task'] == 'diagnose':
[0161] diagnostic_results = tool.run_diagnostics() / / Task type from the task field of execution_plan. If it is diagnostic type (diagnose), control the selected operation and maintenance tool for diagnosis;
[0162] elif execution_plan['task'] == 'repair':
[0163] tool.execute_repair(execution_plan['parameters']) / / Task type from the task field of execution_plan. If it is repair type (repair), the control selects the operation and maintenance tool for repair. The parameters used for repair are obtained from the parameters field of execution_plan.
[0164] S204: Calling the operation and maintenance tool corresponding to the selected operation and maintenance tool information to perform fault repair on the cloud platform according to the repair task information.
[0165] Based on the above embodiment, the specific process of obtaining the operation and maintenance knowledge information corresponding to the log alarm information in the preset knowledge base is introduced below. In one possible situation, the preset knowledge base stores the operation and maintenance knowledge representation vector corresponding to each of the operation and maintenance knowledge information;
[0166] Acquiring operation and maintenance knowledge information corresponding to the log alarm information in a preset knowledge base may include:
[0167] S301: Perform vector conversion on the log alarm information to obtain a log alarm representation vector.
[0168] In this step, in order to find the operation and maintenance knowledge information similar to the log alarm information by means of vector query, the log alarm information needs to be converted into a log alarm representation vector first. The embodiment of the present invention does not limit how to convert the text content into a vector form, and reference may be made to the relevant technologies of artificial intelligence and embedding vectors. It is understandable that the method of vectorizing the log alarm information should be the same as the method of vectorizing the operation and maintenance knowledge information.
[0169] In addition, due to the particularity and uncertainty of the log, the embodiment of the present invention may also only vectorize the alarm information through the embedded model.
[0170] Based on this, the log alarm information includes log information and alarm information; the log alarm information is vector-converted to obtain a log alarm representation vector, which may include:
[0171] Step 91: Perform vector conversion on the alarm information to obtain an alarm representation vector, and use the alarm representation vector as the log alarm representation vector.
[0172] S302. Input the log alarm representation vector into the preset knowledge base, so that the preset knowledge base uses the log alarm representation vector to perform vector retrieval on each of the operation and maintenance knowledge representation vectors to obtain a selected operation and maintenance knowledge representation vector whose vector similarity with the log alarm representation vector is greater than a preset threshold.
[0173] After obtaining the log alarm representation vector, the embodiment of the present invention can calculate the similarity between the log alarm representation vector and each operation and maintenance knowledge representation vector, and use the operation and maintenance knowledge representation vector whose vector similarity with the log alarm representation vector is greater than a preset threshold as the selected operation and maintenance knowledge representation vector.
[0174] It should be noted that the embodiments of the present invention do not limit how to calculate vector similarity, nor do they limit how to search for operation and maintenance knowledge representation vectors based on vector similarity. For example, an approximate nearest neighbor search algorithm (ANN, Approximate Nearest Neighbor) can be used for searching, and reference can be made to related technologies. The embodiments of the present invention also do not limit the specific value of the preset threshold, which can be set according to actual application requirements.
[0175] S303: Use the operation and maintenance knowledge information corresponding to the selected operation and maintenance knowledge representation vector as the operation and maintenance knowledge information corresponding to the log alarm information.
[0176] The following is an introduction to the construction process of the preset knowledge base. In one possible case, the construction process of the preset knowledge base may include:
[0177] S401. Obtain public operation and maintenance knowledge and historical operation and maintenance data of the cloud platform; wherein the historical operation and maintenance data includes historical log alarm information and corresponding fault repair operation information.
[0178] In this embodiment, a preset knowledge base can be constructed using public operation and maintenance knowledge and historical operation and maintenance data of the cloud platform. Among them, public operation and maintenance knowledge can be divided into private domain knowledge and public domain knowledge. Private domain knowledge includes internal enterprise documents, technical manuals, customer support records, and operation and maintenance logs, etc.; public domain knowledge includes public technical blogs, scientific research papers, technical forums, and standard documents, etc. The collected knowledge needs to be classified, screened, and organized to ensure its relevance and effectiveness. The historical operation and maintenance data contains historical log alarm information and corresponding fault repair operation information.
[0179] S402: Using the historical log alarm information as a question and the fault repair operation information as an answer, convert the historical operation and maintenance data into a question-answer pair.
[0180] To facilitate the understanding of the pre-trained language model, the historical operation and maintenance data can be converted into question-answer pairs, with the historical log alarm information as questions and the fault repair operation information as answers. In this way, the pre-trained language model can understand the fault corresponding to the historical log alarm information and the fault repair operation required to repair the fault according to the context.
[0181] S403: Set the public operation and maintenance knowledge and the question-answer pair as the operation and maintenance knowledge information, and perform vector conversion on each piece of the operation and maintenance knowledge information to obtain an operation and maintenance knowledge representation vector.
[0182] It should be noted that the embodiments of the present invention do not limit how to convert the operation and maintenance knowledge information into vectors, and reference may be made to related technologies of artificial intelligence and embedding vectors.
[0183] S404: Add the operation and maintenance knowledge information and the operation and maintenance knowledge representation vector to the preset knowledge base.
[0184] Based on the above embodiments, in order to give full play to the computing advantages of different devices, the embodiments of the present invention can also deploy the pre-trained language model separately in the model deployment device, and then after the host device obtains the fault detection text, it can send it to the model deployment device for processing, and can receive the fault type, the selected operation and maintenance tool information and the repair task information returned by the model deployment device.
[0185] Based on this, inputting the fault detection text into the pre-trained language model may include:
[0186] S501, sending the fault detection text to a model deployment device, so that the model deployment device inputs the fault detection text into a pre-trained language model deployed in the model deployment device;
[0187] S502: Receive the fault type, the selected operation and maintenance tool information, and the repair task information returned by the model deployment device.
[0188] Based on the above embodiments, the above cloud platform fault repair method will be fully introduced below based on a specific schematic diagram. The present invention proposes a cloud platform operation and maintenance intelligent body construction method and a fault repair system, including a knowledge base construction process, an information retrieval and fault diagnosis process, and a task formulation and execution process.
[0189] Knowledge base construction process:
[0190] The knowledge base construction process is an offline process that quantifies the private domain knowledge, public domain knowledge and question-answer pairs generated by online long-term memory and constructs a knowledge index. For easier understanding, please refer to Figure 2 , Figure 2 A schematic diagram of a knowledge base construction process provided by an embodiment of the present invention, the process includes:
[0191] A. Collect private domain knowledge and public domain knowledge. Private domain knowledge includes internal documents, technical manuals, customer support records, and operation and maintenance logs; public domain knowledge includes public technical blogs, scientific research papers, technical forums, and standard documents. The collected knowledge needs to be classified, screened, and organized to ensure its relevance and effectiveness.
[0192] B. Extract historical operation and maintenance data and operation logs from the online long-term memory module (i.e., the preset manual processing database) to generate question-answer pairs, i.e., the problems encountered in the actual operation and maintenance process and their solutions. This includes two parts: question extraction and answer generation. Question extraction: extract the part describing the problem from the logs and indicators. Answer generation: extract or generate the corresponding solution or operation steps from the same record.
[0193] C. The processed private domain knowledge, public domain knowledge, and question-answer pairs are vectorized, that is, the knowledge in text form is converted into vector representation, which retains the semantic information of the text and facilitates subsequent similarity calculation and index construction.
[0194] D. Use the vectorized knowledge and question-answer pairs to build an efficient vector index, that is, store all vectorized knowledge into a vector database to support fast similarity search.
[0195] Information retrieval and troubleshooting process:
[0196] Information retrieval and fault diagnosis processes are practical online applications that can quickly and efficiently retrieve relevant knowledge and then quickly diagnose faults through pre-trained language models. Figure 3 , Figure 3 A schematic diagram of an information retrieval and fault diagnosis process provided by an embodiment of the present invention, the process includes:
[0197] A. For real-time alarms and alarm logs, set the size and sliding step of the time sliding window to analyze the log and alarm data occurring within the window period.
[0198] B. Due to the particularity and uncertainty of the log, only the alarm items are vectorized through the embedded model, and then the vectorized indicators are vectorized through the knowledge base to obtain the corresponding operation and maintenance knowledge.
[0199] C. Bring the windowed alarms, logs, and retrieved operation and maintenance knowledge into the fault location text template to construct the fault location text.
[0200] D. The fault location text is brought into the pre-trained language model (operation and maintenance pre-trained language model). The pre-trained language model serves as the brain of the operation and maintenance intelligent agent and provides feedback to obtain fault information.
[0201] Task formulation and execution process:
[0202] Fault information is instantiated into a specific prompt through the fault repair prompt template, and the execution task is generated by the pre-trained language model, realizing efficient and accurate automatic fault handling. For easy understanding, please refer to Figure 4 , Figure 4 A schematic diagram of a task formulation and execution process provided by an embodiment of the present invention, the process includes:
[0203] A. Fault repair template instantiation: According to the fault type and description, select a suitable template from the predefined fault repair text template library, fill the key information in the fault description into the template, and instantiate it into a specific fault repair text. For fault types that do not have a suitable template in the template library, store them in long-term memory.
[0204] B. Input the instantiated fault repair text into the pre-trained language model. The pre-trained language model generates specific fault handling tasks and operation steps based on the template information and fault information.
[0205] C. Assign the generated tasks to appropriate operation and maintenance personnel or automated systems to ensure that the tasks can be executed in a timely manner. According to the generated tasks, call the corresponding operation and maintenance tools to perform specific operations.
[0206] D. Progress control and feedback,During the task execution process, the progress control module monitors the task execution status in real time, ensures that the processing is completed on time, and records the entire process of fault handling, including the diagnosis process, execution steps and final results, and updates it to the long-term memory module to improve the knowledge base content.
[0207] The present invention proposes a specific implementation method of a cloud platform operation and maintenance intelligent body construction method and system, including a knowledge base construction module, a knowledge retrieval module, a pre-trained language model question and answer module, an operation and maintenance tool module, a tool execution module, and a long-term memory module. For easy understanding, please refer to Figure 5 , Figure 5 A schematic diagram of a cloud platform fault recovery system provided by an embodiment of the present invention, the system includes:
[0208] Knowledge base building blocks:
[0209] The knowledge base construction module collects, processes, and stores various knowledge related to the operation and maintenance of the cloud platform. It obtains operation and maintenance related text information from private and public data sources, including log data, system configurations, operation manuals, common problems and their solutions sorted manually according to long-term memory, etc.; the private and public domains involve processing data in different formats, such as word, markdown, pdf, etc. This invention is developed by using different classes of python libraries, such as python-docx, markdown, PyPDF2, etc. for content extraction; then uses text vectorization technology (such as vectorization models, m3e-base, bge-large-zh, etc.) to convert the extracted content into numerical vectors; uses data structures such as FAISS to organize knowledge vectors to form an efficient vectorized index.
[0210] Knowledge retrieval module:
[0211] The information retrieval module is responsible for the vectorization processing of alarm and log data, and realizes fast retrieval through the vectorized index. Convert real-time alarm and log data into numerical vectors (the vectorization technology used is the same as that when constructing the knowledge base, such as vectorization models, m3e-base, bge-large-zh, etc.), ensuring that these data can be compared and retrieved in the vector space; through the approximate nearest neighbor search (ANN) algorithm, find the alarm knowledge most similar to the query vector in the vector space to realize fault diagnosis and location.
[0212] Pre-trained language model question-answering module:
[0213] The retrieved knowledge base content, current time window log, and current time window alarm are used to instantiate the fault detection text through a text template, and the fault detection text is passed into the pre-trained language model. The pre-trained language model locates system problems and provides solutions. Preset fault location templates and fault repair templates.
[0214] The fault location template is:
[0215] <Instruction>You are a senior software engineer. Based on the known information, answer the question concisely and professionally. If you cannot get an answer from it, say "The question cannot be answered based on the known information", and do not allow adding fabricated content in the answer. The answer should be in Chinese.< / Instruction> / / ;
[0216] <Known information>{{ context}}< / Known information> / / ;
[0217] <Question>Log information is as follows {{log}}, alarm information is as follows {{alert}}. Is there a fault in the current cloud platform? If so, please give the fault information, analyze the cause of the fault and give repair suggestions.< / Question> / / ;
[0218] Among them, the first constraint text is "Answer questions concisely and professionally based on known information" and "Do not allow adding fabricated content to the answer"; the second constraint text is "If there is any, please provide the fault information, analyze the cause of the fault and give repair suggestions"; the third constraint text is "If no answer can be obtained from it, please say 'The question cannot be answered based on the known information'". context is the filling position for operation and maintenance knowledge information, log is the filling position for log information, and alert is the filling position for alarm information.
[0219] The fault repair template is:
[0220] <Instruction>You are a senior software engineer. Based on the existing tools in the tool list, formulate a repair plan for the fault. If no repair plan can be specified, please say 'The problem cannot be repaired using the existing tools'. If the execution commands of the tools can be output, including the execution method and actual parameters, they should be directly executable in the Linux system.< / Instruction> / / ;
[0221] <Tool List>{{ tools}}< / Tool List> / / ;
[0222] <Question>The log information is as follows {{log}}, and the alarm information is as follows {{alert}}. Please formulate a repair plan for the fault based on the above content< / Question> / / ;
[0223] Among them, the fourth constraint text is "Formulate a repair plan for the fault based on the existing tools" and "Do not allow adding fabricated content to the answer"; the fifth constraint text is "If the execution commands of the tools can be output, including the execution method and actual parameters, they should be directly executable in the Linux system"; the sixth constraint text is "If no repair plan can be specified, please say 'The problem cannot be repaired using the existing tools'". tools is the filling position for operation and maintenance tool information, log is the filling position for log information, and alert is the filling position for alarm information.
[0224] Operation and maintenance tool module:
[0225] A module used to provide operation and maintenance tools for the cloud platform. The tools include monitoring tools, automation scripts, etc. The intelligent agent will use these tools to perform various operation and maintenance tasks. The storage format of the operation and maintenance tools is:
[0226] {
[0227] "ID": "The unique identifier of the tool",
[0228] "Tool Name": "Tool name",
[0229] "Execution Method": "Execution method",
[0230] "Execution Parameters": "Execution parameters",
[0231] "Tool Description": "Tool Function"
[0232] }
[0233] The operation and maintenance tool list will be used as the tool list in the fault repair template in the previous step.
[0234] Tool execution module:
[0235] The tool execution module interacts with the pre-trained language model to obtain the execution plan, and specifically executes the operation and maintenance tasks and operations to ensure that the tasks are completed as planned. It selects appropriate tools for diagnosis and repair operations based on the execution plan provided by the pre-trained language model, and monitors the progress of the task in real time. After the task is completed, the module generates a detailed report and feedback on the execution results. If an abnormal situation occurs, the module will adopt a predetermined strategy to handle it and notify the pre-trained language model to adjust the plan to ensure that the repair is completed smoothly. Through precise execution and real-time feedback, the tool execution module effectively ensures the reliable operation of the system.
[0236] execution_plan = model.get_execution_plan() / / Get the execution plan, which contains the selected operation and maintenance tool information and repair task information;
[0237] / / Select a tool and perform diagnostics or repairs;
[0238] tool = select_tool(execution_plan['tool']) / / Parse the selected operation and maintenance tool information from the tool field of execution_plan;
[0239] if execution_plan['task'] == 'diagnose':
[0240] diagnostic_results = tool.run_diagnostics() / / Task type from the task field of execution_plan. If it is diagnostic type (diagnose), control the selected operation and maintenance tool for diagnosis;
[0241] elif execution_plan['task'] == 'repair':
[0242] tool.execute_repair(execution_plan['parameters']) / / Task type from the task field of execution_plan. If it is repair type (repair), the control selects the operation and maintenance tool for repair. The parameters used for repair are obtained from the parameters field of execution_plan.
[0243] Long-term memory database (manually processed database):
[0244] The long-term memory module is responsible for recording and storing faults in the system that cannot be automatically repaired, for subsequent manual analysis and the formation of operation and maintenance knowledge. It stores log data and monitoring data related to the fault, provides detailed fault background information, and supports subsequent analysis and troubleshooting. In addition, it records fault information that cannot be automatically repaired in the system, including the time of occurrence, scope of impact, specific manifestations, and preliminary diagnosis information of the pre-trained language model.
[0245] The cloud platform fault repair device, cloud platform fault repair system, host device, computer program product, and computer-readable storage medium provided in the embodiments of the present invention are introduced below. The smart gateway described below and the message information processing method of the home smart gateway described above can be referenced to each other.
[0246] Please refer to Figure 6 , Figure 6 A structural block diagram of a cloud platform fault repair device provided by an embodiment of the present invention, the device may include:
[0247] The acquisition module 601 is used to acquire the log alarm information of the cloud platform and the operation and maintenance tool information of the operation and maintenance tool, and acquire the operation and maintenance knowledge information corresponding to the log alarm information in a preset knowledge base;
[0248] The fault detection module 602 is used to generate a fault detection text using the operation and maintenance knowledge information, the operation and maintenance tool information, the log alarm information and the preset constraint information, and input the fault detection text into the pre-trained language model, so that the pre-trained language model analyzes the log alarm information according to the operation and maintenance knowledge information and the operation and maintenance tool information under the restriction of the preset constraint information, and outputs the fault type of the cloud platform, the selected operation and maintenance tool information used to repair the fault, and the repair task information;
[0249] The fault repair module 603 is used to call the operation and maintenance tool corresponding to the selected operation and maintenance tool information to perform fault repair on the cloud platform according to the repair task information.
[0250] Optionally, the preset knowledge base stores the operation and maintenance knowledge representation vector corresponding to each of the operation and maintenance knowledge information;
[0251] The acquisition module 601 includes:
[0252] A vector conversion submodule, used to perform vector conversion on the log alarm information to obtain a log alarm representation vector;
[0253] A query submodule, used for inputting the log alarm representation vector into the preset knowledge base, so that the preset knowledge base uses the log alarm representation vector to perform vector retrieval on each of the operation and maintenance knowledge representation vectors, and obtains a selected operation and maintenance knowledge representation vector whose vector similarity with the log alarm representation vector is greater than a preset threshold;
[0254] The operation and maintenance knowledge information acquisition submodule is used to use the operation and maintenance knowledge information corresponding to the selected operation and maintenance knowledge representation vector as the operation and maintenance knowledge information corresponding to the log alarm information.
[0255] Optionally, the log alarm information includes log information and alarm information;
[0256] The vector conversion submodule can be used for:
[0257] The alarm information is vector-converted to obtain an alarm representation vector, so as to use the alarm representation vector as the log alarm representation vector.
[0258] Optionally, the device may further include:
[0259] An operation and maintenance knowledge acquisition module, used to acquire public operation and maintenance knowledge and historical operation and maintenance data of the cloud platform; wherein the historical operation and maintenance data includes historical log alarm information and corresponding fault repair operation information;
[0260] A question-answer pair conversion module, configured to convert the historical operation and maintenance data into a question-answer pair using the historical log alarm information as a question and the fault repair operation information as an answer;
[0261] A vector conversion module, used to set the public operation and maintenance knowledge and the question-answer pair as the operation and maintenance knowledge information, and perform vector conversion on each of the operation and maintenance knowledge information to obtain an operation and maintenance knowledge representation vector;
[0262] The preset knowledge base adding module is used to add the operation and maintenance knowledge information and the operation and maintenance knowledge representation vector to the preset knowledge base.
[0263] Optionally, the acquisition module 601 includes:
[0264] The log alarm information acquisition submodule is used to obtain all log alarm information generated by the cloud platform within a preset time window.
[0265] Optionally, the fault detection module 602 includes:
[0266] An input submodule, used for sending the fault detection text to a model deployment device, so that the model deployment device inputs the fault detection text into a pre-trained language model deployed in the model deployment device;
[0267] The receiving submodule is used to receive the fault type, the selected operation and maintenance tool information and the repair task information returned by the model deployment device.
[0268] Optionally, the fault detection module 602 includes:
[0269] A fault location submodule, used to generate a fault location text using the operation and maintenance knowledge information, the log alarm information and the first preset constraint information, and input the fault location text into the pre-trained language model, so that the pre-trained language model analyzes the log alarm information according to the operation and maintenance knowledge information under the restriction of the first preset constraint information and outputs the fault type of the cloud platform;
[0270] The task formulation submodule is used to generate a fault repair text using the operation and maintenance tool information, the fault type, the log alarm information and the second preset constraint information, and input the fault repair text into the pre-trained language model, so that the pre-trained language model analyzes the fault type and the log alarm information according to the operation and maintenance tool information under the restriction of the second preset constraint information, and outputs the selected operation and maintenance tool information and the repair task information.
[0271] Optionally, the fault location submodule includes:
[0272] A fault location text generating unit is used to fill the operation and maintenance knowledge information and the log alarm information into corresponding positions in a first preset template to obtain the fault location text; wherein the first preset template includes the first preset constraint information;
[0273] The task formulation submodule includes:
[0274] A fault repair template query unit, used to query a second preset template corresponding to the fault type; wherein the second preset template includes the second preset constraint information;
[0275] The fault repair text generating unit is used to fill the operation and maintenance tool information and the log alarm information into the corresponding positions in the second preset template to obtain the fault repair text.
[0276] Optionally, the device may further include:
[0277] A first manual processing output module is used to save the log alarm information to a preset manual processing database when it is determined that the pre-trained language model outputs the third constraint text, so that the operation and maintenance personnel can manually process the fault corresponding to the log alarm information;
[0278] The second manual processing output module is used to save the log alarm information to a preset manual processing database when it is determined that the pre-trained language model outputs the sixth constraint text, so that operation and maintenance personnel can manually process the fault corresponding to the log alarm information.
[0279] Optionally, the device may further include:
[0280] The first knowledge base updating module is used to update the preset knowledge base using the log alarm information and the corresponding fault repair operation information when it is determined that the log alarm information recorded in the manual processing database has been added with the corresponding fault repair operation information.
[0281] Optionally, the device may further include:
[0282] A parsing module is used to parse the selected operation and maintenance tool information and the repair task information from the text content output by the pre-trained language model according to the output format specified by the fifth constraint text.
[0283] Optionally, the device may further include:
[0284] The task execution judgment module is used to judge whether the repair task is successfully executed;
[0285] A second knowledge base updating module is used to integrate the log alarm information, the selected operation and maintenance tool information, and the repair task information into new operation and maintenance knowledge information if the repair task is successfully executed, and update the preset knowledge base with the new operation and maintenance knowledge information;
[0286] a third manual processing output module, configured to generate task adjustment information if the repair task is not successfully executed, and input the task adjustment information into the pre-trained language model, so that the pre-trained language model analyzes the fault type, the log alarm information and the task adjustment information according to the operation and maintenance tool information, and outputs updated selected operation and maintenance tool information and updated repair task information;
[0287] The task re-execution module is used to call the operation and maintenance tool corresponding to the updated selected operation and maintenance tool information to perform fault repair on the cloud platform according to the updated repair task information.
[0288] Please refer to Figure 7 , Figure 7This is a structural block diagram of a host device provided in an embodiment of the present invention. The embodiment of the present invention provides a host device 10, including a processor 11 and a memory 12; wherein the memory 12 is used to store a computer program; and the processor 11 is used to execute the cloud platform fault repair method provided in the aforementioned embodiment when executing the computer program.
[0289] For the specific process of the above-mentioned cloud platform fault repair method, please refer to the corresponding content provided in the above-mentioned embodiment, which will not be repeated here.
[0290] Furthermore, the memory 12 as a carrier for storing resources may be a read-only memory, a random access memory, a magnetic disk or an optical disk, etc., and the storage method may be temporary storage or permanent storage.
[0291] In addition, the host device 10 also includes a power supply 13, a communication interface 14, an input / output interface 15 and a communication bus 16; wherein the power supply 13 is used to provide working voltage for each hardware device on the host device 10; the communication interface 14 can create a data transmission channel between the host device 10 and an external device, and the communication protocol it follows is any communication protocol that can be applied to the technical solution of the present invention, and is not specifically limited here; the input / output interface 15 is used to obtain external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs, and is not specifically limited here.
[0292] Please refer to Figure 8 , Figure 8 A structural block diagram of a cloud platform fault recovery system provided by an embodiment of the present invention includes:
[0293] The host device 10 is used to execute the cloud platform fault repair method as described above;
[0294] A model deployment device 20, in which a pre-trained language model is deployed, is used to receive a fault detection text sent by the host device, input the fault detection text into the pre-trained language model, and send the fault type, selected operation and maintenance tool information, and repair task information output by the pre-trained language model to the host device.
[0295] Since the embodiments of the cloud platform fault repair system part correspond to the embodiments of the cloud platform fault repair method part, please refer to the description of the embodiments of the cloud platform fault repair method part for the embodiments of the cloud platform fault repair system part, and will not be repeated here.
[0296] An embodiment of the present invention further provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the cloud platform fault repair method described in the above embodiment.
[0297] Since the embodiments of the computer program product part correspond to the embodiments of the cloud platform fault repair method part, please refer to the description of the embodiments of the cloud platform fault repair method part for the embodiments of the computer program product part, and will not be repeated here.
[0298] An embodiment of 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 cloud platform fault repair method described in the above embodiment is implemented.
[0299] Since the embodiments of the computer-readable storage medium part correspond to the embodiments of the cloud platform fault repair method part, please refer to the description of the embodiments of the cloud platform fault repair method part for the embodiments of the storage medium part, which will not be repeated here.
[0300] The various embodiments in the specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or 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.
[0301] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.
[0302] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be implemented directly using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.
[0303] The cloud platform fault repair method, apparatus, host device, computer program product, system and computer-readable storage medium provided by the present invention are introduced in detail above. Specific examples are used herein to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. It should be pointed out that for ordinary technicians in this technical field, without departing from the principles of the present invention, several improvements and modifications can be made to the present invention, and these improvements and modifications also fall within the scope of protection of the present invention.
Claims
1. A cloud platform fault repair method, characterized in that: include: Obtain log alarm information of the cloud platform and operation and maintenance tool information of the operation and maintenance tool, and obtain operation and maintenance knowledge information corresponding to the log alarm information in a preset knowledge base; Generate a fault location text using the operation and maintenance knowledge information, the log alarm information and the first preset constraint information, and input the fault location text into a pre-trained language model, so that the pre-trained language model analyzes the log alarm information according to the operation and maintenance knowledge information under the restriction of the first preset constraint information and outputs the fault type of the cloud platform; Generate a fault repair text using the operation and maintenance tool information, the fault type, the log warning information, and the second preset constraint information, and input the fault repair text into the pre-trained language model, so that the pre-trained language model analyzes the fault type and the log warning information according to the operation and maintenance tool information under the restriction of the second preset constraint information, and outputs the selected operation and maintenance tool information and the repair task information; Calling the operation and maintenance tool corresponding to the selected operation and maintenance tool information to perform fault repair on the cloud platform according to the repair task information; The first preset constraint information includes: A first constraint text is used to restrict the pre-trained language model to analyze the log alarm information only according to the operation and maintenance knowledge information; A second constraint text is used to limit the output format used by the pre-trained language model to output the fault type; A third constraint text is used to specify the text content output by the pre-trained language model when the fault type of the cloud platform cannot be determined; The second preset constraint information includes: A fourth constraint text is used to restrict the pre-trained language model to analyze the log alarm information only according to the operation and maintenance tool information; A fifth constraint text, used to limit the output format used by the pre-trained language model to output the selected operation and maintenance tool information and the repair task information; The sixth constraint text is used to specify the text content output by the pre-trained language model when the selected operation and maintenance tool information and the repair task information cannot be determined.
2. The cloud platform fault repair method according to claim 1, characterized in that: The preset knowledge base stores the operation and maintenance knowledge representation vector corresponding to each of the operation and maintenance knowledge information; Obtaining operation and maintenance knowledge information corresponding to the log alarm information in a preset knowledge base includes: Performing vector conversion on the log alarm information to obtain a log alarm representation vector; Inputting the log alarm representation vector into the preset knowledge base, so that the preset knowledge base uses the log alarm representation vector to perform vector retrieval on each of the operation and maintenance knowledge representation vectors, and obtains a selected operation and maintenance knowledge representation vector whose vector similarity with the log alarm representation vector is greater than a preset threshold; The operation and maintenance knowledge information corresponding to the selected operation and maintenance knowledge representation vector is used as the operation and maintenance knowledge information corresponding to the log alarm information.
3. The cloud platform fault repair method according to claim 2, characterized in that: The log alarm information includes log information and alarm information; The log alarm information is converted into a vector to obtain a log alarm representation vector, including: The alarm information is vector-converted to obtain an alarm representation vector, so as to use the alarm representation vector as the log alarm representation vector.
4. The cloud platform fault repair method according to claim 2, characterized in that: The process of constructing the preset knowledge base includes: Obtaining public operation and maintenance knowledge and historical operation and maintenance data of the cloud platform; wherein the historical operation and maintenance data includes historical log alarm information and corresponding fault repair operation information; The historical operation and maintenance data is converted into a question-answer pair by using the historical log alarm information as a question and the fault repair operation information as an answer; The public operation and maintenance knowledge and the question-answer pair are both set as the operation and maintenance knowledge information, and each of the operation and maintenance knowledge information is vector-converted to obtain an operation and maintenance knowledge representation vector; The operation and maintenance knowledge information and the operation and maintenance knowledge representation vector are both added to the preset knowledge base.
5. The cloud platform fault repair method according to claim 1, characterized in that: Obtain log alarm information from the cloud platform, including: Obtain all log alarm information generated by the cloud platform within a preset time window.
6. The cloud platform fault repair method according to claim 1, characterized in that: Inputting the fault detection text into the pre-trained language model includes: Sending the fault detection text to a model deployment device so that the model deployment device inputs the fault detection text into a pre-trained language model deployed in the model deployment device; Receive the fault type, the selected operation and maintenance tool information, and the repair task information returned by the model deployment device.
7. The cloud platform fault repair method according to claim 1, characterized in that: Generating a fault location text using the operation and maintenance knowledge information, the log alarm information and the first preset constraint information includes: Fill the operation and maintenance knowledge information and the log alarm information into corresponding positions in the first preset template to obtain the fault location text; wherein the first preset template includes the first preset constraint information; Generating a fault repair text using the operation and maintenance tool information, the fault type, the log warning information, and the second preset constraint information includes: Querying a second preset template corresponding to the fault type; wherein the second preset template includes the second preset constraint information; Fill the operation and maintenance tool information and the log alarm information into corresponding positions in the second preset template to obtain the fault repair text.
8. The cloud platform fault repair method according to claim 1, characterized in that: After inputting the fault location text into the pre-trained language model, the method further includes: When it is determined that the pre-trained language model outputs the third constraint text, the log alarm information is saved to a preset manual processing database so that operation and maintenance personnel can manually process the fault corresponding to the log alarm information; After inputting the fault repair text into the pre-trained language model, the method further includes: When it is determined that the pre-trained language model outputs the sixth constraint text, the log alarm information is saved to a preset manual processing database so that operation and maintenance personnel can manually process the fault corresponding to the log alarm information.
9. The cloud platform fault repair method according to claim 8, characterized in that: Also includes: When it is determined that the log alarm information recorded in the preset manual processing database has been added with corresponding fault repair operation information, the preset knowledge base is updated using the log alarm information and the corresponding fault repair operation information.
10. The cloud platform fault repair method according to claim 1, characterized in that: Before calling the operation and maintenance tool corresponding to the selected operation and maintenance tool information to perform fault repair on the cloud platform according to the repair task information, the method further includes: According to the output format specified by the fifth constraint text, the selected operation and maintenance tool information and the repair task information are parsed from the text content output by the pre-trained language model.
11. The cloud platform fault repair method according to claim 1, characterized in that: After calling the operation and maintenance tool corresponding to the selected operation and maintenance tool information to perform fault repair on the cloud platform according to the repair task information, the method further includes: Determine whether the repair task is successfully executed; If the repair task is successfully executed, the log alarm information, the selected operation and maintenance tool information, and the repair task information are integrated into new operation and maintenance knowledge information, and the preset knowledge base is updated using the new operation and maintenance knowledge information; If the repair task is not successfully executed, task adjustment information is generated, and the task adjustment information is input into the pre-trained language model, so that the pre-trained language model analyzes the fault type, the log alarm information and the task adjustment information according to the operation and maintenance tool information, and outputs updated selected operation and maintenance tool information and updated repair task information; The operation and maintenance tool corresponding to the updated selected operation and maintenance tool information is called to perform fault repair on the cloud platform according to the updated repair task information.
12. A cloud platform fault repair device, characterized in that: include: An acquisition module is used to acquire log alarm information of the cloud platform and operation and maintenance tool information of the operation and maintenance tool, and acquire operation and maintenance knowledge information corresponding to the log alarm information in a preset knowledge base; Fault detection module, including: A fault location submodule is used to generate a fault location text using the operation and maintenance knowledge information, the log alarm information and the first preset constraint information, and input the fault location text into a pre-trained language model, so that the pre-trained language model analyzes the log alarm information according to the operation and maintenance knowledge information under the restriction of the first preset constraint information and outputs the fault type of the cloud platform; the first preset constraint information includes: a first constraint text, used to restrict the pre-trained language model to analyze the log alarm information only according to the operation and maintenance knowledge information; a second constraint text, used to restrict the output format used by the pre-trained language model to output the fault type; a third constraint text, used to specify the text content output by the pre-trained language model when the fault type of the cloud platform cannot be determined; A task formulation submodule is used to generate a fault repair text using the operation and maintenance tool information, the fault type, the log alarm information and the second preset constraint information, and input the fault repair text to the pre-trained language model, so that the pre-trained language model analyzes the fault type and the log alarm information according to the operation and maintenance tool information under the restriction of the second preset constraint information, and outputs the selected operation and maintenance tool information and the repair task information; the second preset constraint information includes: a fourth constraint text, which is used to limit the pre-trained language model to analyze the log alarm information only according to the operation and maintenance tool information; a fifth constraint text, which is used to limit the output format used by the pre-trained language model to output the selected operation and maintenance tool information and the repair task information; a sixth constraint text, which is used to specify the text content output by the pre-trained language model when the selected operation and maintenance tool information and the repair task information cannot be determined; A fault repair module is used to call the operation and maintenance tool corresponding to the selected operation and maintenance tool information to perform fault repair on the cloud platform according to the repair task information.
13. A host device, characterized in that: include: Memory for storing computer programs; A processor, configured to implement the cloud platform fault repair method as described in any one of claims 1 to 11 when executing the computer program.
14. A cloud platform fault recovery system, characterized in that: include: A host device, used to execute the cloud platform fault repair method according to any one of claims 1 to 11; A model deployment device, in which a pre-trained language model is deployed, is used to receive a fault detection text sent by the host device, input the fault detection text into the pre-trained language model, and send the fault type, selected operation and maintenance tool information, and repair task information output by the pre-trained language model to the host device.
15. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by the processor, the steps of the cloud platform fault repair method described in any one of claims 1 to 11 are implemented.
16. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are loaded and executed by the processor, the cloud platform fault repair method according to any one of claims 1 to 11 is implemented.
