A cloud platform operation and maintenance method, device, equipment, medium and product

Abstract

The application discloses a cloud platform operation and maintenance method, device, equipment, medium and product, relates to the operation and maintenance technical field, and is applied to a server and includes the following steps: receiving a natural language instruction input by a user end; performing intention recognition and parameter extraction on the natural language instruction to obtain at least one atomic task, and generating an execution plan according to the atomic tasks and a target dependency relationship between the atomic tasks; screening an optimal execution path for each atomic task from at least two preset execution paths based on a task attribute of each atomic task in the execution plan; performing simulation and drilling analysis on the execution plan based on the optimal execution path to generate a risk prediction report, and sending the risk prediction report to the user end; after obtaining confirmation execution information sent by the user end based on the risk prediction report, initiating a call to a cloud platform by using a task executor corresponding to the optimal execution path to execute the execution plan, and converting an execution result of the cloud platform into a natural language response and sending the natural language response to the user end.

Description

Technical Field

[0001] This invention relates to the field of operation and maintenance technology, and in particular to a cloud platform operation and maintenance method, apparatus, equipment, medium and product. Background Technology

[0002] With the widespread adoption of cloud computing technology, private and hybrid clouds have become core IT (Information Technology) infrastructure for enterprises. Their large scale, complex architecture, and heterogeneous technologies present significant challenges to operation and maintenance management. Current mainstream cloud platform operation and maintenance methods primarily include Graphical User Interface (GUI), command line and API (Application Programming Interface) scripts, and Infrastructure as Code (IaC). While GUI methods are intuitive, they are inefficient and prone to errors; command line and API methods can achieve automation, but they have high technical requirements and learning costs; while IaC methods are advanced, they still require professional development and maintenance, failing to lower the barrier to entry. A common problem with these methods is the significant gap between user business intent and machine technical execution, leading to low operation and maintenance efficiency and high communication costs.

[0003] In recent years, Large Language Models (LLMs) have made natural language interaction possible, but their direct application in cloud operations and maintenance still faces many obstacles, such as a lack of real-time status awareness, inaccurate command understanding, high security risks, and the absence of built-in risk control mechanisms. Specifically, traditional cloud management platforms rely on pre-set forms and blueprint editors, which cannot support flexible natural language interaction and still require users to adapt to complex interfaces; knowledge graph-based AIOps systems focus on read operations and simple queries, offering limited support for write operations and complex workflows, and lacking task planning and risk control capabilities; while general-purpose AI code generation tools can generate scripts, they cannot be directly integrated into enterprise environments, lack real-time data support, and have no security safeguards, still requiring manual intervention.

[0004] In summary, how to perform secure, efficient, and intelligent operation and maintenance of cloud platforms using natural language is a problem that needs to be solved. Summary of the Invention

[0005] In view of this, the purpose of this invention is to provide a cloud platform operation and maintenance method, apparatus, equipment, medium, and product, capable of performing secure, efficient, and intelligent operation and maintenance of the cloud platform through natural language. The specific solution is as follows:

[0006] Firstly, this application discloses a cloud platform operation and maintenance method, applied to a server, including:

[0007] Receive natural language commands input from the user terminal;

[0008] Intent recognition and parameter extraction are performed on natural language instructions to obtain at least one atomic task, and an execution plan is generated based on the target dependencies between atomic tasks.

[0009] Based on the task attributes of each atomic task in the execution plan, the optimal execution path is selected for each atomic task from at least two preset execution paths; wherein, the preset execution paths include at least an API call path based on the control plane and a database query path based on the data plane;

[0010] The execution plan is simulated and analyzed based on the optimal execution path to generate a risk prediction report, which is then sent to the user's terminal.

[0011] After receiving confirmation of execution information from the user based on the risk prediction report, the system uses the task executor corresponding to the optimal execution path to initiate a call to the cloud platform to execute the execution plan, and converts the execution result from the cloud platform into a natural language response and sends it to the user.

[0012] Optionally, the server and the client interact with each other based on a preset integration interface; wherein the preset integration interface supports multiple communication modes, including at least standard input / output mode, server push event mode and HTTP streaming response mode.

[0013] Optionally, intent recognition and parameter extraction are performed on the natural language instructions to obtain at least one atomic task, including:

[0014] The received natural language instructions are preprocessed to obtain processed instructions; the preprocessing includes error correction and instruction format standardization.

[0015] Based on the pre-trained model, the operation type in the processed instruction is identified, and the operation object and configuration parameters corresponding to the operation type are extracted from the processed instruction.

[0016] Generate at least one atomic task based on the operation type, operation object, and configuration parameters.

[0017] Optionally, the process of intent recognition and parameter extraction for natural language instructions may also include:

[0018] Check for missing target information in the natural language instructions;

[0019] If it exists, a question to be clarified is generated based on the missing target information, and the question to be clarified is sent to the user terminal to receive supplementary feedback from the user terminal regarding the question to be clarified.

[0020] Accordingly, the operation object and configuration parameters corresponding to the operation type are extracted from the processed instructions, including:

[0021] Extract the operation object and configuration parameters corresponding to the operation type from the processed instructions and supplementary feedback.

[0022] Optionally, an execution plan is generated based on the atomic tasks and the target dependencies between them, including:

[0023] Retrieve a pre-configured operations and maintenance knowledge base, and obtain the target dependencies between atomic tasks from the operations and maintenance knowledge base; the operations and maintenance knowledge base is used to record the dependencies between different operations.

[0024] An execution plan based on a directed acyclic graph (DAG) is generated based on each atomic task and the target dependency relationship. Each node in the DAG corresponds to an atomic task, and the edges in the DAG correspond to the target dependencies between atomic tasks.

[0025] Optionally, after generating the execution plan based on the directed acyclic graph, the following may also be included:

[0026] The execution plan based on the directed acyclic graph is encapsulated based on a preset data format to obtain a structured execution plan. The structured execution plan includes the execution order between atomic tasks and the task parameters of each atomic task. The execution order includes parallel execution order and serial execution order.

[0027] Optionally, based on the task attributes of each atomic task in the execution plan, the optimal execution path is selected for each atomic task from at least two preset execution paths, including:

[0028] If the atomic task is a write operation, the optimal execution path will be determined as the API call path based on the control plane;

[0029] If the atomic task is a read operation, the read operation is analyzed to select the optimal execution path from at least two preset execution paths based on the analysis results.

[0030] Optionally, based on the analysis results, the optimal execution path can be selected from at least two preset execution paths, including:

[0031] If the read operation is used to trigger the target business logic of the cloud platform or to obtain the real-time status of resources, the optimal execution path will be determined as the API call path based on the control plane; otherwise, the optimal execution path will be determined as the database query path based on the data plane.

[0032] Optionally, the execution plan can be simulated and analyzed based on the optimal execution path to generate a risk prediction report, including:

[0033] Determine the simulation analysis model corresponding to the optimal execution path;

[0034] The execution process of each atomic task in the execution plan is simulated and analyzed using a simulation analysis model to obtain simulation analysis results;

[0035] A risk assessment report is generated based on the simulation analysis results, including descriptions of potential risks and their impacts.

[0036] Optionally, simulation analysis may include one or more of the following: cost analysis, security compliance analysis, and dependency analysis.

[0037] Optional, the cost analysis process includes:

[0038] Call the cloud platform billing system's valuation interface and determine the expected cost of the execution plan based on preset cost calculation rules;

[0039] Compare the expected costs with the budget corresponding to the execution plan to determine whether there is a risk of exceeding the budget, and write the judgment results into the risk assessment report.

[0040] Optional, the security compliance analysis process includes:

[0041] Obtain the pre-set security baseline rule base;

[0042] The operational parameters of the execution plan are compared with the security baseline rule base to determine whether there is a risk of unauthorized configuration in the execution plan, and the judgment results are written into the risk prediction report.

[0043] Optional, the dependency analysis process includes:

[0044] Identify the target resources involved in the execution of the plan;

[0045] Identify the pre-configured configuration management database in the cloud platform and determine whether there is resource information in the configuration management database that has a dependency relationship with the target resource;

[0046] The judgment results will be written into the risk assessment report.

[0047] Optionally, the execution results from the cloud platform can be converted into a natural language response and sent to the user terminal, including:

[0048] Receive the execution results of each atomic task in the execution plan from the cloud platform; wherein the execution results include at least one of the following: execution status, resource identifier, query results, or error information;

[0049] The execution results are converted into natural language responses using a pre-trained model, and then the natural language responses are sent to the user terminal.

[0050] Optionally, the execution plan is executed by initiating a call to the cloud platform using the task executor corresponding to the optimal execution path, including:

[0051] If the optimal execution path is an API call path based on the control plane, then the preset API tool executor is used to initiate an API call to the cloud platform to execute the execution plan;

[0052] The API tool executor invocation process includes:

[0053] Establish a connection with the cloud platform to obtain the API metadata documentation provided by the cloud platform;

[0054] Parse the API metadata document and generate corresponding call functions for each API operation in memory; the call functions include the calling method for the corresponding API operation;

[0055] The API call is initiated to the cloud platform based on the calling function and the task parameters of each atomic task in the execution plan.

[0056] Optionally, the execution plan is executed by initiating a call to the cloud platform using the task executor corresponding to the optimal execution path, including:

[0057] If the optimal execution path is a database query path based on the data plane, then the preset database query executor is used to initiate a query to the target database on the cloud platform.

[0058] The query process of the database query executor includes:

[0059] Utilize pre-trained models to convert execution plans into structured query intents;

[0060] The structured query intent is input into the preset security compiler, which uses the security compiler to verify and securely compile the structured query intent based on a predefined whitelist rule base to generate the target query statement;

[0061] Execute the target query statement and initiate a query to the target database on the cloud platform.

[0062] Secondly, this application discloses a cloud platform operation and maintenance device applied to a server, comprising:

[0063] The instruction receiving module is used to receive natural language instructions input by the user.

[0064] The execution plan generation module is used to perform intent recognition and parameter extraction on natural language instructions to obtain at least one atomic task, and generate an execution plan based on the target dependencies between the atomic tasks.

[0065] The execution path selection module is used to select the optimal execution path for each atomic task from at least two preset execution paths based on the task attributes of each atomic task in the execution plan; wherein, the preset execution paths include at least API call paths based on the control plane and database query paths based on the data plane;

[0066] The risk prediction module is used to simulate and analyze the execution plan based on the optimal execution path to generate a risk prediction report, and then send the risk prediction report to the user terminal.

[0067] The execution module is used to, after receiving the confirmation execution information sent by the user based on the risk prediction report, use the task executor corresponding to the optimal execution path to initiate a call to the cloud platform to execute the execution plan, and convert the execution result of the cloud platform into a natural language response and send it to the user.

[0068] Thirdly, this application discloses an electronic device, including:

[0069] Memory, used to store computer programs;

[0070] A processor is used to execute computer programs to implement the steps of the aforementioned disclosed cloud platform operation and maintenance method.

[0071] Fourthly, this application discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the steps of the aforementioned disclosed cloud platform operation and maintenance method.

[0072] Fifthly, this application discloses a computer program product, including a computer program that, when executed by a processor, implements the steps of the aforementioned disclosed cloud platform operation and maintenance method.

[0073] As can be seen, this application receives natural language instructions input by the user terminal via a server; performs intent recognition and parameter extraction on the natural language instructions to obtain at least one atomic task, and generates an execution plan based on the target dependencies between the atomic tasks; based on the task attributes of each atomic task in the execution plan, selects the optimal execution path for each atomic task from at least two preset execution paths; wherein, the preset execution paths include at least an API call path based on the control plane and a database query path based on the data plane; performs simulation and analysis on the execution plan based on the optimal execution path to generate a risk prediction report, and sends the risk prediction report to the user terminal; after obtaining the confirmation execution information sent by the user terminal based on the risk prediction report, it uses the task executor corresponding to the optimal execution path to initiate a call to the cloud platform to execute the execution plan, and converts the execution result of the cloud platform into a natural language response and sends it to the user terminal.

[0074] Beneficial Effects: This application automatically converts user's natural language commands into execution plans containing atomic tasks and their inter-task dependencies by receiving natural language commands and performing intent recognition and parameter extraction. This eliminates the need for manual script writing and overcomes the limitations of traditional operations and maintenance relying on professional personnel. Furthermore, by distinguishing between the control plane and data plane of the cloud platform, this application can proactively and intelligently select the optimal execution path for different atomic tasks based on task attributes, achieving intelligent task scheduling. This ensures operational security, consistency, and real-time performance while significantly improving the performance of large-scale data querying and analysis. Moreover, before actual task execution, this application simulates the execution plan based on the optimal execution path and generates a risk prediction report. This allows users to understand the risks and make decisions before actual execution, avoiding accidents caused by misoperation or improper configuration in traditional operations and maintenance. This ensures that all operations comply with safety regulations before execution, thereby improving the security, compliance, and reliability of operations and maintenance. Finally, after obtaining the user's confirmation of execution, this application uses the task executor corresponding to the optimal execution path to initiate a call to the cloud platform to execute the execution plan, and converts the execution result into a natural language response and returns it to the user. This makes the operation result clear and easy to understand, reduces the user's understanding cost, and improves the user experience. Through this application's solution, secure, efficient, and intelligent operation and maintenance of the cloud platform is achieved using natural language. Attached Figure Description

[0075] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0076] Figure 1 This application discloses a flowchart of a cloud platform operation and maintenance method.

[0077] Figure 2 This is a schematic diagram of the overall architecture of a server disclosed in this application;

[0078] Figure 3 This application discloses a specific cloud platform operation and maintenance method flowchart;

[0079] Figure 4 This is a schematic diagram of the structure of a cloud platform operation and maintenance device disclosed in this application;

[0080] Figure 5 This is a structural diagram of an electronic device disclosed in this application. Detailed Implementation

[0081] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0082] With the popularization of cloud computing technology, private and hybrid clouds have become core IT infrastructures for enterprises. Their large scale, complex architecture, and heterogeneous technologies present significant challenges to operation and maintenance management. In recent years, large language models have made natural language interaction possible, but their direct application to cloud operations and maintenance still faces many obstacles, such as a lack of real-time status awareness, inaccurate command understanding, high security risks, and the absence of built-in risk control mechanisms. Therefore, this application discloses a cloud platform operation and maintenance method, apparatus, equipment, medium, and product that enables secure, efficient, and intelligent operation and maintenance of the cloud platform through natural language.

[0083] See Figure 1 As shown in the figure, this application discloses a cloud platform operation and maintenance method applied to a server. The method includes:

[0084] Step S11: Receive natural language instructions input by the user.

[0085] This embodiment discloses a server, specifically an MCP (Model Context Protocol) server. As an intelligent middleware, it connects the user interaction front-end with the underlying cloud platform, accurately translating the user's ambiguous, high-level natural language commands into a series of atomic, secure, and compliant operations on the cloud platform. See also... Figure 2 As shown, the server's overall architecture includes: an adaptive integration interface, a natural language processing and task planning layer, an intelligent scheduling and decision engine, an execution and risk prediction layer, and a response generation layer. This architecture ensures the intelligence, automation, and security of the entire process from receiving user instructions to finally returning the execution result.

[0086] First, it should be noted that the server and the user terminal interact with each other based on a preset integration interface. This preset integration interface supports multiple communication modes, including at least standard input / output mode, server push event mode, and HTTP streaming response mode. In this embodiment, the preset integration interface is the aforementioned adaptive integration interface, whose goal is to achieve deep integration and broad adaptability to the heterogeneous and complex application ecosystem in a private cloud environment. It is not a fixed, single communication interface, but a flexible, pluggable, multi-mode interface framework, ensuring that the server can act as a capability hub, seamlessly embedded into the client's existing operation, monitoring, and automation processes. To achieve this goal, the server's preset integration interface provides multiple communication modes, including standard input / output mode, server push event mode, and HTTP (Hypertext Transfer Protocol) streaming response mode.

[0087] The standard input / output (stdio) mode is primarily designed for script automation. Its purpose is to address the numerous text- and script-based automation tools accumulated by operations teams over many years in enterprise-level private cloud environments. To protect these existing investments and achieve the lowest-cost, highest-efficiency integration with these tools, the server provides a low-level interaction mode based on standard input / output. In this mode, the server process can be directly started by any parent process that supports standard pipes (such as a shell script, a Python program, or an execution module of a configuration management tool like Ansible). The parent process sends serialized (e.g., JSON format) request data by connecting its standard output stream to the server's standard input stream. After processing, the server returns the result data to the parent process via its standard output stream. The entire process does not require traversing the network protocol stack, resulting in virtually zero communication overhead.

[0088] For example, suppose a nighttime inspection script needs to check the status of all virtual machines. This script can call this server and pass a natural language command via stdio to "query the running status and CPU utilization of all virtual machines". After the server executes the command, it returns the structured query results via stdout. The script can then receive these results for further analysis and alerting. This model allows the server to function as a powerful "command" that can be easily orchestrated into any existing automated workflow.

[0089] The server-sent event pattern is primarily designed for real-time monitoring. Its purpose is to enable modern operations and maintenance management platforms, especially monitoring dashboards for network operations centers, to provide real-time, visualized process tracking capabilities for long-running tasks. To meet this requirement, the server implements an asynchronous communication interface based on Server-SentEvents (SSE). When a front-end application (such as a web-based management console) submits a complex task (e.g., "deploy a complete test environment") through this interface, the server immediately returns a confirmation message and maintains the persistence of the HTTP connection. Subsequently, during the long task execution process, the server's internal task executor periodically pushes key status changes, log information, or progress percentages as independent events to the front-end through this persistent connection. Upon receiving these events, the front-end application can update the progress bar, display execution logs, or highlight the currently executing step in real time on the interface.

[0090] For example, operations personnel might issue a command via a web interface to "migrate the production database to a new storage device." During execution, the server will push a series of events to the web interface every few seconds via the SSE interface, such as "Task started, pre-check in progress...", "Pre-check passed, creating target storage volume...", and "Data backup 30% complete...". This allows operations personnel to clearly and in real-time monitor the progress of the entire high-risk operation, greatly improving operational transparency and user confidence.

[0091] HTTP streaming response mode is primarily designed for massive data queries. Its purpose is to address the common need in compliance audits or global resource inventory scenarios to export large amounts of configuration data at once, such as "exporting complete configuration information for all virtual machines, cloud disks, and security groups within the data center." Using a traditional request-response model, the server would need to query and load potentially several gigabytes of data into memory before returning it all at once, easily leading to server memory exhaustion and crashes. To solve this problem, the server's HTTP interface implements a streaming response mechanism, typically utilizing chunked transfer encoding in the HTTP / 2.1.1 protocol. When receiving such a massive data query request, the server's executor does not cache the entire result set when retrieving data from the backend database or API. Instead, it uses a "read-and-write" approach, immediately sending each small chunk (e.g., 1000 records) as an HTTP response data block to the client. The client then receives and processes these data blocks streaming, such as writing them to local files.

[0092] For example, a security auditor might need to export all firewall policy change records from the past year. After executing this query, the server will transmit millions of change records in batches without interruption to the auditor's client tools in a streaming manner. Throughout the process, the server's memory usage remains at an extremely low level, ensuring the system's stability and availability even when processing very large datasets.

[0093] Through the combination of the above-mentioned communication modes, the server's preset integration interface can be adaptive, ensuring that it exhibits extremely high flexibility and robustness in integration with traditional scripts, support for modern web applications, and handling extreme data loads, making it a truly plug-and-play engine in the private cloud management ecosystem.

[0094] Step S12: Perform intent recognition and parameter extraction on the natural language instructions to obtain at least one atomic task, and generate an execution plan based on the target dependencies between the atomic tasks.

[0095] In this embodiment, by receiving natural language commands and performing intent recognition and parameter extraction, the user's natural language commands are automatically converted into an execution plan containing atomic tasks and the target dependencies between atomic tasks. This eliminates the need for manual script writing and overcomes the limitations of traditional operations and maintenance that rely on professional personnel. It should be noted that this process is handled by the natural language processing and task planning layer, which converts natural language into machine-executable instructions. This layer is mainly responsible for deeply understanding the user's intent and decomposing and arranging complex instructions that may contain multiple steps into a clearly structured and logically rigorous task execution plan. This layer mainly consists of two core components: an intent recognition module and a multi-step task planner.

[0096] In a specific implementation, the natural language instructions are subjected to intent recognition and parameter extraction to obtain at least one atomic task, including: preprocessing the received natural language instructions to obtain processed instructions; wherein, the preprocessing includes error correction and instruction format standardization; identifying the operation type in the processed instructions based on a pre-trained model, and extracting the operation object and configuration parameters corresponding to the operation type from the processed instructions; and generating at least one atomic task based on the operation type, operation object and configuration parameters.

[0097] Understandably, the intent recognition module integrates one or more advanced pre-trained models built on large language models as its core engine for semantic understanding. Upon receiving a user's natural language instruction from the interface layer, this module executes a series of refined processing steps, specifically including:

[0098] Instruction preprocessing: First, the raw input is cleaned, such as removing unnecessary punctuation marks, correcting common spelling errors, converting colloquial expressions into more formal terms, and standardizing instruction formats.

[0099] Core Intent Recognition: Utilizing the powerful semantic analysis capabilities of pre-trained models, the core verbs or operation types of instructions are identified, such as create, delete, query, expand, backup, etc.

[0100] Entity and Parameter Extraction: Based on the identification of the core intent, this module further extracts all relevant entities (operation objects) and parameters (configuration parameters) from the text, thereby generating at least one atomic task based on the operation type, operation object, and configuration parameters. For example, in the instruction "Upgrade the CPU specifications of two web servers (i-123, i-456) in East China Region 1 from 4 cores to 8 cores, while keeping the memory unchanged at 16GB", this module needs to accurately extract:

[0101] Operation type: Configuration upgrade;

[0102] List of target resource IDs: [i-123, i-456];

[0103] Regional parameters: East China Zone 1;

[0104] Configuration change parameters: {CPU: 8, Memory: 16}.

[0105] Furthermore, the process of intent recognition and parameter extraction for natural language instructions also includes: checking whether there is missing target information in the natural language instructions; if so, generating a question to be clarified based on the missing target information and sending the question to be clarified to the user terminal to receive supplementary feedback from the user terminal regarding the question to be clarified; correspondingly, extracting the operation object and configuration parameters corresponding to the operation type from the processed instructions, including: extracting the operation object and configuration parameters corresponding to the operation type from the processed instructions and supplementary feedback content.

[0106] In other words, the intent recognition module also checks for missing target information in the natural language command, i.e., whether there is ambiguity or incomplete information, thus proactively identifying ambiguity. If such a problem exists, the intent recognition module generates a clarification question based on the missing target information and sends it to the user's end to receive supplementary feedback from the user, thereby guiding the user to complete the command information. Further, it extracts the operation object and configuration parameters corresponding to the operation type from the processed command and supplementary feedback. For example, if the user's command is "Help me create a test server," the module will find that necessary information such as region, specifications, and image are missing. In this case, it will not directly report an error, but will generate a clarifying question, such as "Okay, in which region do you want to create this test server? What specifications and operating system image do you want to use?", and pass this question to the response generation layer, which returns it to the user, thus forming an interactive dialogue process that guides the user to complete the command.

[0107] Furthermore, it should be noted that in real-world private cloud operations and maintenance scenarios, much of the valuable work is not a single operation, but rather a complex workflow consisting of multiple steps with sequential relationships and dependencies. The multi-step task planner in this application is designed to automate the handling of such complex tasks. When the output of the intent recognition module contains multiple atomic operations, the multi-step task planner is activated, with the aim of generating an execution plan based on the target dependencies between atomic tasks.

[0108] Step S13: Based on the task attributes of each atomic task in the execution plan, select the optimal execution path for each atomic task from at least two preset execution paths; wherein, the preset execution paths include at least an API call path based on the control plane and a database query path based on the data plane.

[0109] In this embodiment, by distinguishing between the control plane and data plane of the cloud platform, the optimal execution path can be proactively and intelligently selected for different atomic tasks based on task attributes, achieving intelligent task scheduling. This ensures operational security, consistency, and real-time performance while significantly improving the performance of large-scale data querying and analysis. It should be noted that this process is handled by the intelligent scheduling and decision engine, which is the decision-making center of the server architecture. It embodies the core idea of ​​this invention in balancing operational correctness and query performance. It receives the decomposed atomic tasks from the task planning layer and intelligently selects an optimal execution path for each task based on its inherent properties. The core basis for its decision-making is a profound distinction and application of the two interaction modes of the private cloud platform: the control plane and the data plane.

[0110] In a specific implementation, based on the task attributes of each atomic task in the execution plan, the optimal execution path is selected from at least two preset execution paths for each atomic task, including: if the atomic task is a write operation, the optimal execution path is determined as the API call path based on the control plane; if the atomic task is a read operation, the read operation is analyzed, and the optimal execution path is selected from at least two preset execution paths based on the analysis results.

[0111] It is understood that the embodiments of this application set scheduling rules to specify which atomic tasks should use the API call path based on the control plane and which atomic tasks should use the database query path based on the data plane. First, it should be noted that the control plane is a set of programmatic interfaces provided by the cloud platform to manage the entire lifecycle of resources. It is not merely a simple encapsulation of the database; it contains the platform's complex core business logic, including but not limited to: strict identity authentication and permission verification, resource quota and billing management, idempotency guarantees for operations, resource isolation between multi-tenants, and detailed operation log auditing. The data plane refers to the underlying database system that stores various types of data from the cloud platform. In a typical private cloud environment, for purposes such as performance monitoring, data analysis, and backup auditing, read-only copies of the production database are usually built, or various types of data are synchronized to a centralized data warehouse or data lake through an ETL (Extract, Transform, Load) process. The database query path refers to directly querying these securely accessible, non-production databases.

[0112] In one specific implementation, if the atomic task is a write operation, the optimal execution path is directly determined as the API call path based on the control plane. That is, all write operations—any operation intended to change the state or configuration of a resource, such as creating, deleting, modifying, updating, starting, stopping, restarting, binding, unbinding, and authorizing—must be executed through the control plane API. This is to ensure that all changes undergo legality verification by the platform's business logic, thereby guaranteeing data consistency and state correctness throughout the cloud environment.

[0113] In another specific implementation, if the atomic task is a read operation, the read operation is further analyzed, and the optimal execution path is selected from at least two preset execution paths based on the analysis results. Specifically, selecting the optimal execution path from at least two preset execution paths includes: if the read operation is used to trigger the target business logic of the cloud platform, or to obtain the real-time status of resources, then the optimal execution path is determined to be an API call path based on the control plane; otherwise, the optimal execution path is determined to be a database query path based on the data plane.

[0114] Understandably, when read operations are required to trigger platform business logic, some query operations, while not changing resource status, need to trigger specific platform calculations or checks. For example, "querying whether the current user has permission to create a virtual machine of a specific specification" requires real-time calls to the permission verification interface to obtain an accurate answer; therefore, the control plane-based API call path should be chosen as the optimal execution path. Furthermore, when read operations are required to obtain the absolute real-time status of a resource, such as when a user needs to obtain the precise status of a resource at this very moment, like "querying whether the i-123 virtual machine is currently running," an API call must be made; that is, the control plane-based API call path should be chosen as the optimal execution path. This is because the API directly connects to the cloud platform's source-of-truth database, providing the most authoritative status information without delay.

[0115] In addition, when an atomic task is a read operation and does not meet the mandatory routing conditions of the API call path, the intelligent scheduling and decision engine will further analyze the nature of its query. If it meets the following characteristics, it will be preferentially routed to the database query path based on the data plane:

[0116] 1. Complex, cross-resource type correlation analysis: When a query needs to perform correlation analysis on multiple different types of resources, such as "querying a list of all virtual machines that are mounted with 'SSD high-performance cloud disks,' whose security groups have 'port 80' open, and whose operating system is 'CentOS 7'," implementing this via API would require separately calling interfaces for querying virtual machines, cloud disks, and security groups, and then performing complex matching and filtering on the large amount of data returned in server memory, resulting in extremely low efficiency. However, a single SQL (Structured Query Language) join query can efficiently complete this task using a database query.

[0117] 2. Aggregation and Statistics of Large-Scale Data: When queries require aggregation calculations on massive amounts of data, such as "statistics on the total consumption (or total cost) of computing, storage, and network resources for each business department in the past month," or "calculation of the total number of virtual machines in an 'error' state across all projects," these queries utilize built-in database aggregate functions (such as SUM, COUNT, AVG) and grouping (GROUP BY) capabilities, achieving performance far superior to API calls.

[0118] 3. Historical Snapshots and Time-Series Data Analysis: When queries involve historical status or performance metrics over a specific time period, such as "querying the average, maximum, and minimum CPU utilization of virtual machine i-456 between 2 PM and 4 PM yesterday," cloud platform monitoring systems typically store this type of time-series data in dedicated time-series databases. This historical data is often not accessible through standard resource management APIs. Directly querying these monitoring databases is more efficient.

[0119] In this way, through the scheduling decision mechanism of this application, the server can find the optimal execution path for each user request, achieving a perfect balance between the correctness and security of the operation and the high performance and breadth of the analysis and query.

[0120] Step S14: Based on the optimal execution path, perform simulation and analysis of the execution plan to generate a risk prediction report, and send the risk prediction report to the user terminal.

[0121] In this embodiment, before the actual task execution, the application also simulates the execution plan based on the optimal execution path and generates a risk prediction report. This allows users to be aware of the risks and make decisions before actual execution, avoiding accidents caused by misoperation or improper configuration in traditional operations and maintenance. This ensures that all operations comply with safety regulations before execution, thereby improving the security, compliance, and reliability of operations and maintenance. It should be noted that this process is handled by the execution and risk prediction layer, serving as a mandatory pre-check checkpoint to enhance operational security.

[0122] It should be noted that, in a preferred implementation, simulation analysis can be performed only on atomic tasks related to write operations. That is, when a "write" operation task (such as batch creation of virtual machines) arrives at this layer, it is not immediately sent to the cloud platform. Instead, it is first sent to the risk prediction layer for simulation analysis, and the analysis results—a consulting report containing potential risks and impacts—are returned to the upper layer and finally presented to the user. Only after obtaining explicit authorization from the user will the task be allowed to proceed to the actual executor.

[0123] Step S15: After obtaining the confirmation execution information sent by the user based on the risk prediction report, the task executor corresponding to the optimal execution path is used to initiate a call to the cloud platform to execute the execution plan, and the execution result of the cloud platform is converted into a natural language response and sent to the user.

[0124] In this embodiment, after obtaining the user's confirmation of execution, the application uses the task executor corresponding to the optimal execution path to initiate a call to the cloud platform to execute the execution plan, and converts the execution result into a natural language response and returns it to the user. This makes the operation result clear and easy to understand, reduces the user's understanding cost, and improves the user experience. Through the solution of this application, secure, efficient, and intelligent operation and maintenance of the cloud platform is achieved through natural language.

[0125] It should be noted that, in addition to the aforementioned risk prediction module, the execution and risk prediction layer also includes an API tool executor and a database (DB) query executor.

[0126] In one specific implementation, the execution plan is executed by initiating a call to the cloud platform using a task executor corresponding to the optimal execution path. This includes: if the optimal execution path is an API call path based on the control plane, then initiating an API call to the cloud platform using a preset API tool executor to execute the execution plan; wherein the API tool executor's invocation process includes: establishing a connection with the cloud platform to obtain the API metadata document provided by the cloud platform; parsing the API metadata document and generating corresponding call functions for each API operation in memory; wherein the call functions include the call method for the corresponding API operation; and initiating an API call to the cloud platform based on the call functions and the task parameters of each atomic task in the execution plan.

[0127] In other words, when the optimal execution path is an API call path based on the control plane, the corresponding task executor is the API tool executor. Its core idea is runtime dynamic synthesis. To solve the maintenance problems caused by the rapid iteration of cloud platform APIs in traditional automation tools, the API tool executor adopts a runtime dynamic synthesis method. It does not rely on any pre-written software development kits (SDKs) for specific API versions. Instead, it is designed to directly understand the API metadata published by the cloud platform itself. The API tool executor's invocation process specifically includes the following:

[0128] Metadata Synchronization and Parsing: When the server starts up or receives an update command, it automatically connects to the target cloud platform's metadata service address and retrieves a complete API metadata document describing all available APIs. Then, by parsing the API metadata document, it loads the defined services, resources, operations, parameters, data structures, and other information into the server's memory.

[0129] In-memory tool generation: Based on the parsed API metadata, this embodiment dynamically creates a corresponding, directly callable function object or method in memory for each API operation. The function then calls the calling method containing the API operation, thereby initiating an API call to the cloud platform based on the calling function and the task parameters of each atomic task in the execution plan. This dynamically generated tool has a clear function signature, parameter list, parameter types, and docstring, making it almost indistinguishable from a function written by the developer. This process relies on the metaprogramming capabilities of the target programming language. For example, in Python, this can be achieved by dynamically creating classes (types) and functions (FunctionTypes). When the cloud platform adds a new API, the server only needs to resynchronize the metadata once to automatically learn this new capability, without any code changes or redeployment, achieving true "hot updates."

[0130] Dynamic adaptation during API calls: To address the subtle differences in API design among various cloud services, the API tool executor possesses two key dynamic adaptation capabilities when actually initiating API calls:

[0131] Dynamic Endpoint Calculation: Different cloud services may have different endpoint rules. For example, object storage services may have independent domain names in each region, while identity authentication services may have a single, nationally centralized domain name. API tool executors can automatically and accurately calculate and construct the exact URL (Uniform Resource Locator) required for each API call based on the service attributes declared in the metadata and the user-specified regional parameters.

[0132] Dynamic parameter marshalling: APIs have diverse formatting requirements for complex parameters. For example, a parameter that requires a list of IDs might be a comma-separated string, a JSON array, or multiple repeated lookup parameters with the same name. API tool executors can automatically marshal the user-provided raw data (such as a Python list) into the correct format required by the API before initiating the call, based on the detailed format declarations for each parameter in the metadata, thus shielding the user from the complexities of the underlying implementation.

[0133] In another specific implementation, the execution plan is executed by initiating a call to the cloud platform using the task executor corresponding to the optimal execution path. This includes: if the optimal execution path is a database query path based on the data plane, then initiating a query to the target database of the cloud platform using a preset database query executor; wherein the query process of the database query executor includes: converting the execution plan into a structured query intent using a pre-trained model; inputting the structured query intent into a preset security compiler, so that the security compiler can verify and securely compile the structured query intent based on a predefined whitelist rule base to generate a target query statement; executing the target query statement and initiating a query to the target database of the cloud platform.

[0134] In other words, when the optimal execution path is a database query path based on the data plane, the corresponding task executor is the database query executor. Its core principle is that, because directly allowing a large language model to generate and execute SQL statements poses a significant security risk (SQL injection), the database query executor employs a strict secure compilation model from intent to SQL to completely eliminate this risk. The query process of the database query executor specifically includes the following:

[0135] Structured Query Intent: First, a pre-trained model built on a large language model is needed to convert the execution plan into a structured query intent. In this process, the pre-trained model's task is not to generate SQL, but rather to generate a standardized, intermediate representation (R) structured object that describes the query requirements. This object clearly defines the tables (resource types), fields (attributes), filtering conditions (WHERE clauses), join relationships (JOIN), and aggregation requirements (GROUP BY).

[0136] Rule-Driven Security Compiler: The server incorporates a non-AI-based security compiler based on deterministic rules. This compiler receives the structured query intents mentioned above as input and performs validation and secure compilation based on a predefined whitelist rule base to generate the target query statement. Specifically, the security compiler maintains a whitelist that strictly defines which tables and fields are allowed to be queried, as well as the legitimate relationships between different tables. The compiler verifies that all elements in the query intent are on the whitelist, and then compiles it into a final, executable target query statement based on these secure, validated elements. Any attempt to access unauthorized data or perform dangerous operations is directly rejected during the compilation phase. This design fully leverages the ability of a large language model to understand complex queries while ensuring absolute security for database access through traditional compilation techniques.

[0137] Furthermore, the execution results from the cloud platform are converted into natural language responses and sent to the user terminal. This includes: receiving the execution results from the cloud platform for each atomic task in the execution plan; wherein the execution results include at least one of execution status, resource identifier, query results, or error messages; using a pre-trained model to convert the execution results into natural language responses, and sending the natural language responses to the user terminal. It should be noted that this process is mainly handled by the response generation layer, which is responsible for transforming the machine-generated, impersonal execution results into user-friendly and easily understood natural language responses. The response generation layer mainly includes processes such as result aggregation, natural language generation, and error mitigation.

[0138] Result aggregation: For a multi-step task, this layer collects the execution results from all steps in the execution layer and aggregates these results, which may include various types such as execution status, success information, error codes, resource identifiers, and query results, into a single, structured context object.

[0139] Natural Language Generation: This context object is then passed to the large language model. The task of the large language model is to generate a fluent, coherent, and concise natural language description based on this context, which contains the complete execution process and the final result—that is, to obtain the corresponding natural language response—and send the natural language response to the user. For example, for a successful virtual machine creation task, it will not simply return a JSON containing the resource ID, but will generate: "Okay, a 4-core, 8GB virtual machine with the ID i-xyz has been successfully created for you in East China Zone 1. The virtual machine has started and you can now access it via remote connection."

[0140] Error and Exception Handling: When errors occur during execution, this embodiment can also "translate" low-level, technically jargon-filled error messages (such as "InvalidParameter.ValueNotSupported") into reasons and next steps that users can understand. For example: "Sorry, the operation failed. The reason is that the 'Ultra-High Frequency' specification you selected is sold out in this availability zone. We suggest you try other availability zones or select the 'General' specification."

[0141] In this way, through the work of the response generation layer, the interaction experience between the server and the user is greatly improved, realizing the transformation from "humans adapting to machines" to "machines adapting to humans", enabling non-technical personnel to easily understand and manage complex cloud resources.

[0142] As can be seen, this application automatically converts user's natural language commands into execution plans containing atomic tasks and the target dependencies between atomic tasks by receiving natural language commands and performing intent recognition and parameter extraction. This eliminates the need for manual script writing and overcomes the limitations of traditional operations and maintenance that rely on professional personnel. Furthermore, by distinguishing between the control plane and data plane of the cloud platform, this application can proactively and intelligently select the optimal execution path for different atomic tasks based on task attributes, achieving intelligent task scheduling. This ensures the security, consistency, and real-time performance of operations while significantly improving the performance of large-scale data querying and analysis. Moreover, before the actual execution of the task, this application also simulates the execution plan based on the optimal execution path and generates a risk prediction report. This allows users to be aware of the risks and make decisions before actual execution, avoiding accidents caused by misoperation or improper configuration in traditional operations and maintenance. This ensures that all operations comply with safety regulations before execution, thereby improving the security, compliance, and reliability of operations and maintenance. Finally, after obtaining the user's confirmation of execution, this application uses the task executor corresponding to the optimal execution path to initiate a call to the cloud platform to execute the execution plan, and converts the execution result into a natural language response and returns it to the user. This makes the operation result clear and easy to understand, reduces the user's understanding cost, and improves the user experience. Through this application's solution, secure, efficient, and intelligent operation and maintenance of the cloud platform is achieved using natural language.

[0143] See Figure 3 As shown, this application discloses a specific cloud platform operation and maintenance method. Compared with the previous embodiment, this embodiment further explains and optimizes the technical solution. Specifically, it includes:

[0144] Step S21: Receive natural language instructions input by the user.

[0145] Step S22: Perform intent recognition and parameter extraction on the natural language instructions to obtain at least one atomic task, acquire a pre-set operation and maintenance knowledge base, and obtain the target dependency relationship between each atomic task from the operation and maintenance knowledge base; wherein, the operation and maintenance knowledge base is used to record the dependency relationship between different operations.

[0146] In this embodiment, the multi-step task planner first obtains a list of all identified atomic tasks. Then, it queries a built-in or externally configurable operations and maintenance (O&M) knowledge base. This O&M knowledge base defines standard dependencies between different cloud resources and different operations. For example, it defines that the "mount cloud disk" operation depends on both the "virtual machine" and the "cloud disk" being in an "available" state; the "add virtual machine to load balancer" operation depends on both the "virtual machine" and the "load balancer instance" existing. Therefore, this application obtains the target dependencies between atomic tasks from a pre-set O&M knowledge base.

[0147] Step S23: Generate an execution plan based on a directed acyclic graph (DAG) according to each atomic task and the target dependency relationship; wherein each node of the DAG corresponds to an atomic task, and the edges in the DAG correspond to the target dependency relationships between atomic tasks.

[0148] In this embodiment, the multi-step task planner further generates an execution plan based on a Directed Acyclic Graph (DAG) according to each atomic task and its target dependencies. That is, the multi-step task planner organizes all atomic tasks into a DAG, which precisely describes the execution order and concurrency relationships of the entire workflow. Each node in the DAG represents an atomic task, and the directed edges between nodes represent the target dependencies between tasks. Taking a complex instruction, "Create a highly available cluster for an e-commerce application," as an example, the generated DAG clearly shows which tasks can be executed in parallel to improve efficiency, such as creating multiple virtual machines and multiple cloud disks simultaneously; and which tasks must be strictly sequential, such as creating virtual machines first before subsequent software deployment.

[0149] Furthermore, after generating the execution plan based on the directed acyclic graph (DAG), the process includes encapsulating the DAG-based execution plan using a preset data format to obtain a structured execution plan. This structured execution plan includes the execution order of atomic tasks and the task parameters for each atomic task. The execution order can be either parallel or sequential. In other words, the multi-step task planner encapsulates the constructed DAG in a standardized, machine-readable data structure (usually JSON format) to obtain a structured execution plan. This structured execution plan not only includes the topological order of tasks but also attaches all task parameters extracted from the intent recognition module to each task node. This detailed execution plan is the core data object driving all subsequent processing flows, and it will be passed to the next layer of intelligent scheduling and decision-making engine.

[0150] It should also be noted that the planner is capable of handling more advanced instructions, such as those containing conditional statements ("If server A's load exceeds 80%, then scale up a new server") or loop operations ("Install security patches for all virtual machines labeled 'Web'"). It will also orchestrate this logic into the execution plan, forming a more complex execution graph with conditional branches and loop nodes.

[0151] Step S24: Based on the task attributes of each atomic task in the execution plan, select the optimal execution path for each atomic task from at least two preset execution paths; wherein, the preset execution paths include at least an API call path based on the control plane and a database query path based on the data plane.

[0152] Step S25: Determine the simulation analysis model corresponding to the optimal execution path, and use the simulation analysis model to simulate and analyze the execution process of each atomic task in the execution plan to obtain the simulation analysis results.

[0153] In this embodiment, the simulation exercise analysis includes any one or more of the following analyses: cost analysis, security compliance analysis, and dependency analysis.

[0154] The cost analysis process specifically includes: calling the cloud platform billing system's valuation interface and determining the expected cost of the execution plan based on preset cost calculation rules; comparing the expected cost with the budget corresponding to the execution plan to determine whether the execution plan has a risk of exceeding the budget, and writing the judgment result into a risk prediction report. That is, in private cloud or hybrid cloud scenarios that support internal billing, this embodiment can query the internal resource price catalog or call the billing system's valuation interface based on the operation content to simulate the potential financial impact of this operation based on preset cost calculation rules, and obtain the expected cost. Then, by comparing the expected cost with the budget corresponding to the execution plan, it determines whether the execution plan has a risk of exceeding the budget, and writes the judgment result into a risk prediction report. For example, when a user attempts to create 100 virtual machines, it calculates the expected cost per hour, per day, and per month for these resources and prompts the user: "Warning: The 100 virtual machines you applied to create are expected to incur approximately 150,000 yuan in monthly costs, exceeding your project team's monthly budget. Please confirm whether to continue."

[0155] The security compliance analysis process specifically includes: obtaining a pre-set security baseline rule base; comparing the operation parameters of the execution plan with the security baseline rule base to determine whether the execution plan has any unauthorized configuration risks, and writing the judgment results into a risk prediction report. That is, this embodiment maintains a configurable security baseline rule base, which defines the best practices and security red lines within the enterprise, such as: "Prohibiting the creation of virtual machines without passwords or keys," "Prohibiting the mounting of unencrypted cloud disks to core database instances," and "Prohibiting the opening of high-risk ports such as database (3306) or remote desktop (3389) to the public network (0.0.0.0 / 0)." Before executing an operation, this application compares the operation parameters of the execution plan with the rule base to determine whether the execution plan has any unauthorized configuration risks. Once a violation is detected, it will be immediately intercepted, and the user will be informed of the detailed violation content and modification suggestions.

[0156] The dependency analysis process specifically includes: identifying the target resources involved in the execution plan during execution; identifying the pre-configured configuration management database in the cloud platform and determining whether resource information with dependencies on the target resources exists in the configuration management database; and writing the determination results into a risk assessment report. It is understandable that for dangerous operations such as deletion or modification, this application will determine whether resource information with dependencies on the target resources exists in the pre-configured configuration management database in the cloud platform, thereby analyzing the potential cascading effects. For example, when a user attempts to delete a storage volume shared by multiple virtual machines, it will issue a warning: "Caution: You are attempting to delete the shared storage volume vol-abc, which is being mounted and used by virtual machines i-123 and i-789. Deletion will cause applications on these virtual machines to crash. Please ensure you understand the associated risks."

[0157] Step S26: Generate a risk prediction report containing descriptions of potential risks and impacts based on the simulation analysis results, and send the risk prediction report to the user terminal.

[0158] In this embodiment, a risk prediction report containing descriptions of potential risks and their impacts is generated based on the aforementioned simulation analysis results, and the risk prediction report is sent to the user terminal.

[0159] Step S27: After obtaining the confirmation execution information sent by the user based on the risk prediction report, the task executor corresponding to the optimal execution path is used to initiate a call to the cloud platform to execute the execution plan, and the execution result of the cloud platform is converted into a natural language response and sent to the user.

[0160] For more detailed processing of steps S21 and S27, please refer to the relevant content disclosed in the foregoing embodiments, which will not be repeated here.

[0161] Through the above solution, this application has the following technical effects:

[0162] 1. Significantly lowers the technical barrier and improves operational efficiency and response speed. This application replaces complex GUI clicks and professional script programming with intuitive natural language interaction. This enables non-professional operations and maintenance personnel to safely perform complex cloud resource management tasks, thereby reducing what might have taken hours or even days of manual operation or script development to a few minutes of dialogue, greatly improving the IT department's response speed to business needs and overall operational efficiency.

[0163] 2. Significantly enhances the security, compliance, and predictability of operations and maintenance. This application employs a risk prediction mechanism to mandate cost estimation, security baseline scanning, and impact analysis before any destructive or high-cost operation is executed, and reports potential risks to users in a human-understandable manner for decision-making. This forms a human-machine collaborative security closed loop, effectively preventing major production accidents and financial losses caused by misoperation or misconfiguration, and ensuring that all operations comply with the company's security and compliance policies before execution.

[0164] 3. Achieve adaptive and unified management of heterogeneous cloud environments, reducing maintenance costs. This application's dynamic API tool executor dynamically synthesizes calling tools by directly consuming the cloud platform's API metadata at runtime, completely eliminating dependence on specific SDK versions. When the cloud platform API is upgraded or changed, the server only needs to synchronize the latest metadata to automatically adapt, without requiring code modification or redeployment. This enables it to easily address the challenges of rapid API iteration and differences in multi-cloud and hybrid cloud environments with extremely low maintenance costs, providing a unified, future-oriented management entry point.

[0165] 4. Intelligent Balancing of Operational Correctness and Query Analysis Performance. This application distinguishes between the control plane and data plane of the cloud platform, selecting the optimal execution path for tasks of different natures. This ensures that all write operations undergo rigorous business logic and permission checks by the cloud platform API, guaranteeing absolute data consistency and correctness. Simultaneously, it routes complex, cross-resource, large-scale data analysis and statistical read operations to the higher-performance data plane, avoiding the performance bottlenecks of traditional API polling methods. This hybrid query mode achieves a balance between security and high performance.

[0166] 5. Ensuring absolute security for database queries. In scenarios involving data queries using large language models, this application's "intent-to-SQL secure compiler" mode fundamentally eliminates the risk of SQL injection by having the LLM generate intermediate representations (IRs) instead of directly generating SQL, and then having a compiler based on deterministic rules generate the final SQL. This allows users to benefit from the powerful semantic understanding capabilities of the LLM while providing security guarantees for queries of core enterprise data.

[0167] The implementation of the present invention will be described in detail below with reference to a specific application scenario.

[0168] Scenario: A project manager needs to quickly deploy a standard, highly available web application testing environment on the company's private cloud located in the East China-Shanghai region for an upcoming "Smart Retail Project." The environment requirements include: two application servers (specifications: 4-core CPU, 8GB RAM, CentOS 8 operating system), one database server (specifications: 8-core CPU, 16GB RAM, MySQL 8.0), and connecting the two application servers to a new load balancer instance to provide unified services externally.

[0169] The implementation steps are as follows:

[0170] Step 1: User command input and interface reception

[0171] The project manager logs into the company's unified operations and maintenance portal (a web application) and enters a natural language command in the dialog box: "Help me deploy a high-availability test environment for the smart retail project in the Shanghai region. It requires two 4-core 8G CentOS application servers and one 8-core 16G MySQL database server. Connect the application servers behind a load balancer."

[0172] The front end of the operations and maintenance portal sends a JSON request containing the aforementioned text to the server via the adaptive integration interface provided by the server of this invention, using the HTTP protocol. Since the deployment process may be time-consuming, the portal front end prioritizes establishing a Server Push Event (SSE) connection with the server to receive subsequent progress updates in real time.

[0173] Step 2: Natural Language Processing and Task Planning

[0174] The request reaches the server's natural language processing and task planning layer.

[0175] 1. The intent recognition module starts, and the internal Large Language Model (LLM) analyzes the instructions:

[0176] Core intent identification: Identifying the user's core objective as "deployment environment" is a complex and multifaceted process.

[0177] Entity and parameter extraction: Accurately extract the following information:

[0178] Region: East China - Shanghai;

[0179] Project: Smart Retail Project;

[0180] Resources list:

[0181] Application server (type) x 2 (quantity), 4-core CPU (specification), 8GB memory (specification), CentOS (operating system);

[0182] Database server (type) x 1 (quantity), 8-core CPU (specification), 16GB memory (specification), MySQL (application);

[0183] Load balancer (type) x 1 (quantity);

[0184] Relationship extraction: Identify the dependency that the "application server" needs to be "attached behind the load balancer".

[0185] Ambiguity Resolution: Module checks revealed that creating virtual machines typically requires specifying a network (VPC / subnet) and security group, which the user had not provided. Instead of failing, the module generates a clarifying question: "Okay. In which network (VPC) do you wish to deploy these servers? Is a specific security group required?" This question is returned to the user through the response generation layer. After the user selects the default "dev-vpc" and "web-server-sg" security groups on the interface, the information is completed.

[0186] 2. The multi-step task planner receives the above complete and structured intent information.

[0187] Task decomposition: The planner breaks down the grand goal of "deployment environment" into a series of atomic cloud platform operations, such as: `create_vm`, `create_db_instance`, `create_load_balancer`, and `add_lb_backend`.

[0188] Dependency Analysis: The planner queries the built-in operations and maintenance knowledge base to analyze the dependencies of these atomic tasks: servers and load balancers must be created first before servers can be added as backends to the load balancer. The creation of various servers and load balancers themselves has no dependencies and can be executed in parallel to reduce time.

[0189] Execution plan generation: The planner constructs these tasks and dependencies into a directed acyclic graph (DAG). This graph is represented in JSON format and clearly defines the execution flow, for example:

[0190] Parallel Phase 1:

[0191] Task A: Create Application Server 1 (Parameters: Shanghai, 4c8g, CentOS, dev-vpc...);

[0192] Task B: Create Application Server 2 (Parameters: Shanghai, 4c8g, CentOS, dev-vpc...);

[0193] Task C: Create a database server (parameters: Shanghai, 8c16g, MySQL, dev-vpc...);

[0194] Task D: Create a load balancer (parameter: dev-vpc...);

[0195] Serial Phase 2 (Completed by A and D):

[0196] Task E: Add the instance ID of application server 1 to the backend server group of the load balancer.

[0197] Serial Phase 3 (Completed by B and D):

[0198] Task F: Add the instance ID of application server 2 to the backend server group of the load balancer.

[0199] Step 3: Intelligent Scheduling and Decision-Making

[0200] The execution plan containing the DAG is passed to the intelligent scheduling and decision engine. The engine traverses every atomic task (A to F) in the DAG. Its analysis reveals that all these tasks (`create_...`, `add_...`) are "write" operations that modify resource states. Based on built-in scheduling rules, the engine decides that all tasks must be routed to "path A - control plane API call path" to ensure absolute safety, compliance, and state consistency of the operations.

[0201] Step 4: Execution and Risk Assessment

[0202] The task list is sent to the execution and risk prediction layer.

[0203] 1. The risk prediction module intervenes first to conduct a simulation exercise of the entire DAG:

[0204] Cost simulation: Based on the specifications of the 3 servers and 1 load balancer to be created, the module queries the internal billing system and estimates: "Warning: The resources created in this operation are expected to incur a monthly cost of approximately 3,500 yuan. This cost will be recorded in the 'Smart Retail Project' cost center. Please confirm."

[0205] Security Compliance Scan: The module checks whether the operation parameters comply with the security baseline. For example, it finds that the "web-server-sg" security group selected by the user for the application server has opened the public SSH (22) port, which violates the rule "prohibit test environment servers from exposing management ports to the public network". It will intercept the operation and prompt: "Security Warning: The security group you selected allows access to the SSH port from the public network, which poses a security risk. It is recommended to change to the 'internal-access-only-sg' security group, or confirm acceptance of this risk."

[0206] The project manager saw the above prompt on the web interface, selected to accept the cost, and adopted the suggestion to change the security group. He clicked the "Confirm Execution" button.

[0207] 2. After user authorization, the DAG is passed to the dynamic API tool executor.

[0208] The executor, upon startup, retrieves the latest OpenAPI / Swagger specifications by accessing the cloud platform's metadata service and dynamically generates corresponding, directly callable Python functions in memory for all APIs (such as `CreateVirtualMachine` and `CreateLoadBalancer`).

[0209] The executor begins executing tasks according to the concurrent and serial relationships defined in the DAG. It first initiates API calls in parallel to create two application servers, one database, and one load balancer.

[0210] When `CreateVirtualMachine` is called, the dynamic endpoint calculation logic automatically sends the API request to the correct address `ecs.sh-01.mycloud.com` based on the "East China-Shanghai" regional parameter. Simultaneously, the dynamic parameter encapsulation logic automatically converts the Python dictionary-style server configuration into the nested XML or JSON format required by the cloud platform API.

[0211] Whenever an atomic task (such as task A "Create Application Server 1") completes, the executor pushes a progress message to the user's front end through the previously established SSE connection, such as `{"step": "Create App Server 1", "status": "Success", "instance_id": "i-abcdef12"}`. The user can see a real-time progress bar and logs on the interface.

[0212] Once all tasks in the parallel phase are completed, the executor continues to execute subsequent serial tasks, adding the created server instance IDs to the load balancer.

[0213] Step 5: Response Generation

[0214] After the entire DAG is executed, the successful results of all steps (including the newly created resource ID, the public IP of the load balancer, etc.) are aggregated into the response generation layer.

[0215] This layer invokes LLM to render the structured results data into a summary, user-friendly natural language text. LLM generates the following response: "Deployment successful! The required test environment for the 'Smart Retail Project' has been created in the Shanghai region. Details are as follows:"

[0216] Application servers (2 units): i-abcdef12, i-ghijkl34;

[0217] Database server (1 unit): db-mnopqr56;

[0218] Load balancer IP: 121.xxx;

[0219] You can now access your web application via this IP address.

[0220] This text was finally displayed on the project manager's web interface via the SSE connection's closing message. Thus, a complex, multi-step operations and maintenance task was completed securely, efficiently, and transparently through a single natural language dialogue.

[0221] See Figure 4 As shown in the figure, this application discloses a cloud platform operation and maintenance device applied to a server. The device includes:

[0222] The instruction receiving module 11 is used to receive natural language instructions input by the user terminal;

[0223] The execution plan generation module 12 is used to perform intent recognition and parameter extraction on natural language instructions to obtain at least one atomic task, and generate an execution plan based on the target dependencies between the atomic tasks.

[0224] The execution path selection module 13 is used to select the optimal execution path for each atomic task from at least two preset execution paths based on the task attributes of each atomic task in the execution plan; wherein the preset execution paths include at least API call paths based on the control plane and database query paths based on the data plane.

[0225] The risk prediction module 14 is used to simulate and analyze the execution plan based on the optimal execution path to generate a risk prediction report, and send the risk prediction report to the user terminal.

[0226] The execution module 15 is used to, after obtaining the confirmation execution information sent by the user terminal based on the risk prediction report, use the task executor corresponding to the optimal execution path to initiate a call to the cloud platform to execute the execution plan, and convert the execution result of the cloud platform into a natural language response and send it to the user terminal.

[0227] As can be seen, this application automatically converts user's natural language commands into execution plans containing atomic tasks and the target dependencies between atomic tasks by receiving natural language commands and performing intent recognition and parameter extraction. This eliminates the need for manual script writing and overcomes the limitations of traditional operations and maintenance that rely on professional personnel. Furthermore, by distinguishing between the control plane and data plane of the cloud platform, this application can proactively and intelligently select the optimal execution path for different atomic tasks based on task attributes, achieving intelligent task scheduling. This ensures the security, consistency, and real-time performance of operations while significantly improving the performance of large-scale data querying and analysis. Moreover, before the actual execution of the task, this application also simulates the execution plan based on the optimal execution path and generates a risk prediction report. This allows users to be aware of the risks and make decisions before actual execution, avoiding accidents caused by misoperation or improper configuration in traditional operations and maintenance. This ensures that all operations comply with safety regulations before execution, thereby improving the security, compliance, and reliability of operations and maintenance. Finally, after obtaining the user's confirmation of execution, this application uses the task executor corresponding to the optimal execution path to initiate a call to the cloud platform to execute the execution plan, and converts the execution result into a natural language response and returns it to the user. This makes the operation result clear and easy to understand, reduces the user's understanding cost, and improves the user experience. Through this application's solution, secure, efficient, and intelligent operation and maintenance of the cloud platform is achieved using natural language.

[0228] Since the embodiments of the device part correspond to the embodiments described above, please refer to the embodiments described in the method part for the embodiments of the device part, and will not be repeated here.

[0229] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Specifically, it may include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 stores a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the cloud platform operation and maintenance method performed by the electronic device disclosed in any of the foregoing embodiments.

[0230] In this embodiment, the power supply 23 is used to provide operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of this application, and is not specifically limited here; the input / output interface 25 is used to acquire 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.

[0231] The processor 21 may include one or more processing cores, such as a quad-core processor or an octa-core processor. The processor 21 may be implemented using at least one hardware form selected from DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). The processor 21 may also include a main processor and a coprocessor. The main processor, also known as a CPU (Central Processing Unit), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state. In some embodiments, the processor 21 may integrate a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the screen. In some embodiments, the processor 21 may also include an AI (Artificial Intelligence) processor, which is used to handle computational operations related to machine learning.

[0232] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk or optical disk, etc. The resources stored on it include operating system 221, computer program 222 and data 223, etc., and the storage method can be temporary storage or permanent storage.

[0233] The operating system 221 manages and controls the various hardware devices and computer programs 222 on the electronic device 20 to enable the processor 21 to perform calculations and processing on the massive amounts of data 223 in the memory 22. The operating system 221 can be Windows, Unix, Linux, etc. The computer program 222, in addition to including a computer program capable of performing the cloud platform operation and maintenance method executed by the electronic device 20 as disclosed in any of the foregoing embodiments, may further include computer programs capable of performing other specific tasks. The data 223 may include data received by the electronic device from external devices, as well as data collected by its own input / output interface 25.

[0234] Furthermore, this application also discloses a computer-readable storage medium storing a computer program. When the computer program is loaded and executed by a processor, it implements the cloud platform operation and maintenance method steps disclosed in any of the foregoing embodiments.

[0235] This invention also discloses a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the cloud platform operation and maintenance method disclosed in any of the foregoing embodiments.

[0236] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section.

[0237] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0238] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, compact disc read-only memory (CD-ROM), or any other form of storage medium known in the art.

[0239] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0240] The above provides a detailed description of the cloud platform operation and maintenance method, apparatus, equipment, medium, and product provided by the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only intended to help understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A cloud platform operation and maintenance method, characterized in that, Applied to a server, comprising: receiving a natural language instruction input by a user end; performing intent recognition and parameter extraction on the natural language instruction to obtain at least one atomic task, and generating an execution plan according to the atomic tasks and target dependency relationships between the atomic tasks; based on the task attributes of each atomic task in the execution plan, screening an optimal execution path for each atomic task from at least two preset execution paths; wherein the preset execution paths at least include an API calling path based on a control plane and a database query path based on a data plane; based on the optimal execution path, simulating and analyzing the execution plan to generate a risk prediction report, and sending the risk prediction report to the user end; after obtaining confirmation execution information sent by the user end based on the risk prediction report, initiating a call to a cloud platform using a task executor corresponding to the optimal execution path to execute the execution plan, and converting the execution result of the cloud platform into a natural language response and sending it to the user end; wherein the simulating and analyzing the execution plan based on the optimal execution path to generate a risk prediction report comprises: determining a simulation analysis model corresponding to the optimal execution path; using the simulation analysis model to simulate and analyze the execution process of each atomic task in the execution plan to obtain simulation analysis results; generating a risk prediction report containing potential risks and impact explanations based on the simulation analysis results; the simulation and analysis includes any one or several of cost analysis, security compliance analysis and dependency relationship analysis; the process of cost analysis comprises: calling the valuation interface of the cloud platform billing system, and determining the expected cost of the execution plan based on the preset cost calculation rules; comparing the expected cost with the budget amount corresponding to the execution plan to determine whether the execution plan has an over-budget risk, and writing the determination result into the risk prediction report. 2.The cloud platform operation and maintenance method according to claim 1, characterized in that, The server and the user end perform data interaction based on a preset integration interface; wherein the preset integration interface supports multiple communication modes, and the multiple communication modes at least include a standard input / output mode, a server push event mode and an HTTP streaming response mode. 3.The cloud platform operation and maintenance method according to claim 1, characterized in that, The process of performing intent recognition and parameter extraction on the natural language instruction to obtain at least one atomic task comprises: preprocessing the received natural language instruction to obtain a processed instruction; wherein the preprocessing includes error correction and instruction format standardization; based on a pre-trained model, identifying the operation type in the processed instruction, and extracting the operation object and configuration parameters corresponding to the operation type from the processed instruction; based on the operation type, the operation object and the configuration parameters, generating at least one atomic task.

4. The cloud platform operation and maintenance method according to claim 3, characterized in that, In the process of performing intent recognition and parameter extraction on the natural language instruction, it also comprises: checking whether there is target missing information in the natural language instruction; generating a to-be-clarified question based on the target missing information if the target missing information exists, and sending the to-be-clarified question to the user terminal to receive supplementary feedback content of the user terminal for the to-be-clarified question; Correspondingly, the extracting the operation object and the configuration parameter corresponding to the operation type from the processed instruction includes: extracting the operation object and the configuration parameter corresponding to the operation type from the processed instruction and the supplementary feedback content. 5.The cloud platform operation and maintenance method according to claim 1, characterized in that, The generating an execution plan according to the atomic tasks and the target dependency relationship between the atomic tasks includes: obtaining a pre-set operation and maintenance knowledge base, and obtaining the target dependency relationship between the atomic tasks from the operation and maintenance knowledge base; wherein the operation and maintenance knowledge base is used to record the dependency relationship between different operations; generating a directed acyclic graph-based execution plan according to the atomic tasks and the target dependency relationship; wherein each node of the directed acyclic graph corresponds to an atomic task, and the edge in the directed acyclic graph corresponds to the target dependency relationship between the atomic tasks. 6.The cloud platform operation and maintenance method according to claim 5, characterized in that, After the generating the directed acyclic graph-based execution plan, the method further includes: packaging the directed acyclic graph-based execution plan based on a pre-set data format to obtain a structured execution plan; wherein the structured execution plan includes an execution order between the atomic tasks and a task parameter of each atomic task, and the execution order includes a parallel execution order and a serial execution order. 7.The cloud platform operation and maintenance method according to claim 1, characterized in that, The screening the optimal execution path from at least two pre-set execution paths for each atomic task based on the task attribute of each atomic task in the execution plan includes: if the atomic task is a write operation, determining the optimal execution path as an API call path based on a control plane; if the atomic task is a read operation, analyzing the read operation to screen the optimal execution path from at least two pre-set execution paths according to an analysis result. 8.The cloud platform operation and maintenance method according to claim 7, characterized in that, The screening the optimal execution path from at least two pre-set execution paths according to the analysis result includes: if the read operation is an operation for triggering a target business logic of a cloud platform or an operation for obtaining a real-time state of a resource, determining the optimal execution path as an API call path based on a control plane, otherwise, determining the optimal execution path as a database query path based on a data plane. 9.The cloud platform operation and maintenance method according to claim 1, wherein, The process of the security compliance analysis includes: obtaining a pre-set security baseline rule library; comparing the operation parameter of the execution plan with the security baseline rule library to determine whether the execution plan has a rule violation risk, and writing a determination result into the risk prediction report. 10.The cloud platform operation and maintenance method according to claim 1, characterized in that, The process of the dependency relationship analysis includes: determining a target resource involved in an execution process of the execution plan; determining a pre-set configuration management database in a cloud platform, and determining whether there is resource information having a dependency relationship with the target resource in the configuration management database; writing a determination result into the risk prediction report. 11.The cloud platform operation and maintenance method according to claim 1, characterized in that, The converting the execution result of the cloud platform into a natural language response and sending the natural language response to the user terminal includes: receive an execution result of each atomic task in the execution plan from the cloud platform, wherein the execution result comprises at least one of an execution state, a resource identifier, a query result, or error information; convert the execution result into a natural language response by using a pre-trained model, and send the natural language response to the user terminal. 12.The cloud platform operation and maintenance method according to any one of claims 1 to 11, characterized in that, the task executor corresponding to the optimal execution path is used to initiate a call to the cloud platform to execute the execution plan, including: if the optimal execution path is an API call path based on a control plane, a preset API tool executor is used to initiate an API call to the cloud platform to execute the execution plan; wherein the call process of the API tool executor includes: establishing a connection with the cloud platform to obtain an API metadata document provided by the cloud platform; parsing the API metadata document and generating a corresponding call function in the memory for each API operation; wherein the call function includes a call method of the corresponding API operation; based on the call function and the task parameters of each atomic task in the execution plan, an API call is initiated to the cloud platform. 13.The cloud platform operation and maintenance method according to any one of claims 1 to 11, characterized in that, the task executor corresponding to the optimal execution path is used to initiate a call to the cloud platform to execute the execution plan, including: if the optimal execution path is a database query path based on a data plane, a preset database query executor is used to initiate a query to a target database of the cloud platform; wherein the query process of the database query executor includes: using a pre-trained model to convert the execution plan into a structured query intent; inputting the structured query intent into a preset security compiler to use the security compiler to check and securely compile the structured query intent based on a pre-defined whitelist rule library to generate a target query statement; execute the target query statement and initiate a query to the target database of the cloud platform. 14.A cloud platform operation and maintenance device, characterized in that, application to a server, comprising: an instruction receiving module configured to receive a natural language instruction input by a user terminal; an execution plan generating module configured to perform intent recognition and parameter extraction on the natural language instruction to obtain at least one atomic task, and generate an execution plan according to the atomic tasks and target dependency relationships between the atomic tasks; an execution path selection module configured to select an optimal execution path for each atomic task from at least two preset execution paths based on the task attributes of each atomic task in the execution plan; wherein the preset execution paths include at least an API call path based on a control plane and a database query path based on a data plane; a risk prediction module configured to perform simulation analysis on the execution plan based on the optimal execution path to generate a risk prediction report, and send the risk prediction report to the user terminal; an execution module configured to, after obtaining confirmation execution information sent by the user terminal based on the risk prediction report, use a task executor corresponding to the optimal execution path to initiate a call to the cloud platform to execute the execution plan, and convert the execution result of the cloud platform into a natural language response and send it to the user terminal; wherein the risk prediction module is specifically configured to: determining a simulation analysis model corresponding to the optimal execution path; performing simulation analysis on the execution process of each atomic task in the execution plan by using the simulation analysis model to obtain simulation analysis results; generating a risk prediction report containing potential risks and impact descriptions based on the simulation analysis results The simulation analysis includes any one or several of cost analysis, safety compliance analysis, and dependency analysis. The cost analysis process specifically includes: calling an estimation interface of a cloud platform billing system and determining an expected cost of the execution plan based on a preset cost calculation rule; comparing the expected cost with a budget amount corresponding to the execution plan to determine whether the execution plan has an over-budget risk, and writing the determination result into the risk prediction report.

15. An electronic device, comprising: comprises: a memory for storing a computer program; a processor for executing the computer program to implement the steps of the cloud platform operation and maintenance method according to any one of claims 1 to 13.

16. A computer-readable storage medium, characterized in that, for storing a computer program; wherein the computer program is executed by a processor to implement the steps of the cloud platform operation and maintenance method according to any one of claims 1 to 13.

17. A computer program product comprising a computer program, characterized in that, The computer program is executed by a processor to implement the steps of the cloud platform operation and maintenance method according to any one of claims 1 to 13.

Images