Intelligent operation and maintenance management system for server cluster in cloud computing environment

Abstract

The invention discloses an intelligent operation and maintenance management system for a server cluster in a cloud computing environment, and relates to the technical field of cloud computing operation and maintenance management. The business intention modeling module analyzes the business event and constructs a business resource association graph; the dynamic baseline generation module generates a dynamic threshold with a service context according to the atlas; the fault prediction module predicts potential faults based on the long short-term memory network model; the intelligent decision-making module generates a self-healing strategy according to the prediction result and the map; the security execution module executes the strategy and realizes lossless operation; according to the invention, predictive self-healing closed-loop management under service intention perception is realized, normal fluctuation and abnormal premonition of the service are effectively distinguished, invalid alarms are reduced, the operation and maintenance automation level and the system reliability are improved, and the system reliability is improved. And meanwhile, the long-term adaptability and stability of the system are enhanced through a continuous self-optimization mechanism.

Description

Technical Field

[0001] This invention relates to the field of cloud computing operation and maintenance management technology, specifically to an intelligent operation and maintenance management system for server clusters in a cloud computing environment. Background Technology

[0002] With the rapid development of cloud computing technology, server clusters have become the core infrastructure supporting various Internet applications and business systems. Their ever-expanding scale and increasingly complex architecture have brought enormous challenges to operation and maintenance management. Traditional operation and maintenance models heavily rely on manual experience, making it difficult to cope with the dynamic changes in massive monitoring data and rapidly changing business states. Therefore, achieving automation and intelligence in operation and maintenance management has become a key technical problem that the industry urgently needs to solve.

[0003] Currently, various automated operation and maintenance management solutions have been proposed in the industry. For example, some existing technologies generate dynamic thresholds by analyzing historical monitoring data to reduce false alarms from static threshold alarms. This technology can adjust alarm thresholds according to load patterns, but it is essentially limited to the perception of fluctuations at the resource indicator level and cannot distinguish between normal resource consumption within the business plan and true anomalies. Other existing technologies achieve automated recovery after a failure occurs by executing repair actions through predefined recovery scripts. However, this method is a typical post-event remediation model, lacking the ability to predict potential failures, unable to intervene before a failure occurs, and its decision-making process does not consider the business context, potentially interfering with normal business activities due to repair operations.

[0004] A common technical deficiency faced by these existing technologies is the lack of business intent understanding in the operations and maintenance management system. The system can only perceive the "physiological signals" of underlying resource indicators, but cannot interpret the business logic behind these signal changes. This leads to the system's inability to effectively distinguish between resource fluctuations caused by planned business activities such as "e-commerce promotions" and "data backups" and genuine early signs of failure, resulting in a large number of invalid alarms and wasting operational resources. Furthermore, their response mechanisms are all based on the assumption that anomalies or failures have already occurred, representing a passive response, lacking the ability to predict based on business context and proactively execute precise self-healing before a failure occurs. Therefore, how to enable operations and maintenance systems to possess business intent perception capabilities, achieving a fundamental shift from "passive alarm-manual handling" to "proactive prediction-intelligent self-healing," is a pressing technical problem that needs to be overcome.

[0005] In summary, existing server cluster operation and maintenance management systems in cloud computing environments still have significant shortcomings in terms of intelligence, particularly in the areas of integrating business semantic understanding and achieving predictive self-healing. Developing an intelligent operation and maintenance management system capable of understanding business intent and, based on this, making forward-looking decisions and achieving lossless self-healing is of great significance for improving the reliability of cloud computing services and reducing operation and maintenance costs. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of existing technologies and provide an intelligent operation and maintenance management system for server clusters in a cloud computing environment. By constructing a business resource association graph to understand business intent, and based on this, dynamic baseline management and fault prediction are realized. This can improve operation and maintenance decision-making from passive resource indicator response to proactive prediction and intelligent self-healing combined with business context, thereby effectively reducing false alarms, realizing a lossless operation and maintenance closed loop for service instances, and significantly improving the level of operation and maintenance automation and system reliability.

[0007] To address the aforementioned technical problems, this invention provides the following technical solution: an intelligent operation and maintenance management system for server clusters in a cloud computing environment, comprising a data acquisition module, a business intent modeling module, a dynamic baseline generation module, a fault prediction module, an intelligent decision-making module, and a security execution module. The data acquisition module is used to collect resource indicator data and business indicator data of the server cluster; The business intent modeling module is used to parse business event information and construct a business resource association graph. The dynamic baseline generation module is used to generate dynamic threshold ranges with business context labels based on the business resource association graph; The fault prediction module is used to predict potential fault modes based on resource indicator data and business indicator data. The intelligent decision-making module is used to generate a self-healing strategy based on the prediction results and the business resource association graph. The security execution module is used to call the cloud platform interface to execute the self-healing strategy and achieve lossless operation of the service instance; The system achieves closed-loop operation and maintenance management from resource monitoring to predictive self-healing through business intent awareness.

[0008] Furthermore, the business intent modeling module includes a business event parsing unit and an association graph construction unit; The business event parsing unit extracts event type, event time, and expected load characteristics from the business calendar using natural language processing technology; The association graph construction unit establishes a mapping relationship between business events and resource pools, and the mapping relationship is represented as a graph structure. , where the vertex set Includes business event nodes and resource nodes edge set Includes directed edges connecting business event nodes and resource nodes. ; The business event node The attributes include event type. Start time End time and load level ; The resource node The attributes include resource type Resource Identifier and resource capacity ; The directed edge weight This represents the impact factor of business events on resources, calculated through historical data analysis.

[0009] Furthermore, the fault prediction module employs a time-series prediction algorithm to analyze resource indicator sequences. This time-series prediction algorithm includes a Long Short-Term Memory (LSTM) network model, and the forward computation process of the LTM network model is expressed as follows:

[0010]

[0011]

[0012]

[0013]

[0014]

[0015] in Indicates time step The input feature vector contains resource indicator data and business indicator data. Indicates time step The hidden state vector, Indicates time step The cell state vector, This represents the output vector of the forget gate. This represents the input gate output vector. This represents the output vector of the output gate. Represents the candidate cell state vector. , , , Represents the weight matrix. , , , This represents the bias vector. This represents the sigmoid activation function. Represents the hyperbolic tangent activation function; The Long Short-Term Memory (LSTM) network model outputs a sequence of predicted resource indicators and a fault probability distribution for future time periods.

[0016] Furthermore, the intelligent decision-making module includes a root cause localization unit and a strategy generation unit; The root cause localization unit determines the root cause of the anomaly through multi-dimensional data fusion analysis, and the multi-dimensional data includes resource indicator data, business indicator data, and business resource correlation map information. The strategy generation unit generates self-healing strategies based on fault probability distribution and business impact assessment. The self-healing strategies include instance migration strategies, resource adjustment strategies, and service restart strategies. The strategy generation unit calculates the cost function value of each strategy. The cost function is a weighted summation function of resource utilization, service level agreement compliance and operational risk coefficient. The strategy generation unit selects the strategy with the minimum cost function value as the optimal self-healing strategy.

[0017] Furthermore, the safety execution module includes an operation orchestration unit and a fuse protection unit; The operation orchestration unit decomposes the self-healing strategy into an ordered sequence of atomic operations, including instance creation, instance destruction, service registration, and service discovery. The circuit breaker protection unit monitors the execution status of atomic operations and triggers the circuit breaker mechanism when an abnormal execution mode is detected. The abnormal execution modes include continuous operation failures, resource allocation timeouts, and service response delays. The circuit breaker protection unit selects to continue execution, suspend execution, or switch to manual intervention mode based on the severity of the abnormality.

[0018] Furthermore, the dynamic baseline generation module adopts a multi-scenario baseline management method, wherein the multi-scenario includes weekday scenarios, weekend scenarios, peak business scenarios, and system maintenance scenarios; The dynamic baseline generation module establishes an independent threshold calculation model for each scenario, and the threshold calculation model obtains the normal fluctuation range based on time series analysis of historical data. The dynamic baseline generation module automatically switches the corresponding scenario baseline based on the current business event. The scenario switching is achieved by matching the event type in the business resource association graph.

[0019] Furthermore, the system also includes a feedback learning module, which collects execution result data of the self-healing strategy, including strategy effectiveness indicators and system stability indicators; The feedback learning module uses a reinforcement learning algorithm to update the strategy generation rules of the intelligent decision-making module. The state space of the reinforcement learning algorithm includes the system resource state and the business event state, the action space includes a set of self-healing strategies, and the reward function is calculated based on the improvement of system performance after the strategy is executed. The feedback learning module periodically retrains the time-series prediction algorithm model of the fault prediction module.

[0020] Furthermore, the business intent modeling module also includes a business indicator correlation analysis unit. This unit calculates the correlation coefficient matrix between business indicators and resource indicators. The correlation coefficient matrix is ​​used to optimize the edge weights of the business resource correlation graph. The correlation coefficient is calculated using the Pearson correlation coefficient formula.

[0021] in Represents a sequence of business metrics. Represents a sequence of resource indicators. Indicates the number of data points. This represents the mean of the business indicator series. This represents the mean of the resource indicator series. Indicates the first Individual business metrics values, Indicates the first Individual resource indicator values; The business indicator correlation analysis unit adjusts the weight of the impact factor of business events on resources based on the correlation coefficient value.

[0022] Compared with existing technologies, this intelligent operation and maintenance management system for server clusters in a cloud computing environment has the following advantages: I. This invention parses business event information and constructs a business resource association graph through a business intent modeling module. Combined with a dynamic baseline generation module, it generates labeled dynamic threshold ranges based on business context, enabling the system to accurately distinguish between resource fluctuations caused by planned business activities and actual fault precursors, effectively reducing invalid alarms. Furthermore, the fault prediction module performs time-series analysis on resource indicator sequences based on a long short-term memory network model to identify potential fault modes in advance. The intelligent decision-making module generates self-healing strategies that match the business scenario, and the secure execution module achieves lossless operation. This realizes a fundamental shift from passive response to proactive predictive self-healing, significantly improving the accuracy and automation level of operation and maintenance management, reducing manual intervention costs, and enhancing the overall reliability of cloud computing services.

[0023] Second, this invention continuously collects execution result data of the self-healing strategy through a feedback learning module, including strategy effectiveness indicators and system stability indicators, and uses reinforcement learning algorithms to dynamically update the strategy generation rules of the intelligent decision-making module. At the same time, it periodically retrains the time-series prediction model of the fault prediction module, enabling the system to adapt to changes in the business environment and resource status, continuously optimize the fault prediction accuracy and the applicability of the self-healing strategy, thereby achieving continuous self-improvement of the system's operation and maintenance capabilities and long-term stable operation.

[0024] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0026] Figure 1 This is a data interaction diagram of the system modules of the present invention; Figure 2 This is a business resource association graph structure diagram of the present invention; Figure 3 This is a flowchart of the process of the present invention. Detailed Implementation

[0027] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0028] Example 1 like Figures 1 to 3 As shown, the server cluster intelligent operation and maintenance management system of this embodiment is applicable to server clusters in public cloud, private cloud or hybrid cloud environments. It covers the entire closed-loop management process from data collection to self-healing execution. Its core lies in realizing forward-looking operation and maintenance decisions through business intent perception. The following is a detailed description of the specific implementation, collaboration process and key algorithm details of each module of the system.

[0029] This embodiment of the intelligent operation and maintenance management system includes a data acquisition module, a business intent modeling module, a dynamic baseline generation module, a fault prediction module, an intelligent decision-making module, a security execution module, and a feedback learning module. Each module interacts via a data bus, adhering to a unified interface specification to ensure the stability and scalability of inter-module collaboration. The core working logic of the system provides basic data support for the data acquisition module, the business intent modeling module constructs the correlation between business and resources, the dynamic baseline generation module determines a reasonable monitoring threshold range, the fault prediction module identifies potential fault risks, the intelligent decision-making module generates the optimal self-healing strategy, the security execution module implements self-healing operations, and the feedback learning module continuously optimizes system performance, forming a complete operation and maintenance closed loop.

[0030] In this embodiment, the data acquisition module is implemented as follows: The core function of the data acquisition module is to collect resource and business metrics data from the server cluster, providing foundational data for data analysis and decision-making in subsequent modules.

[0031] The resource metrics data collection targets include computing resources, storage resources, and network resources within the server cluster. Computing resource metrics include CPU utilization, memory utilization, and the number of processes; storage resource metrics include disk utilization, disk I / O rate, and storage read / write latency; network resource metrics include network bandwidth utilization, data packet transmission rate, network latency, and packet loss rate. Business metric data collection targets various business systems running on the server cluster, including transaction success rate, request response time, concurrent users, business throughput, and task completion rate.

[0032] Data collection is implemented using a combination of agent-based collection and API calls. A lightweight collection agent is deployed on each node of the server cluster. This agent collects local resource metrics data at a preset frequency, which can be dynamically adjusted based on system load to ensure real-time data while avoiding excessive system resource consumption. For business metrics data, it is obtained by calling standard monitoring interfaces provided by the business system or by parsing the business system's runtime logs. Log parsing uses streaming processing to ensure real-time data collection.

[0033] The collected raw data undergoes preprocessing, including data cleaning, data normalization, and data alignment. Data cleaning removes outliers and missing values, using mean imputation to handle missing values ​​and the 3σ criterion to identify and remove outliers. Data normalization maps all indicator data to the same numerical range, preventing differences in indicator magnitude from affecting subsequent algorithm calculations. Data alignment uses timestamps as a benchmark to ensure consistency in the time dimension of resource indicator data and business indicator data, guaranteeing the accuracy of subsequent correlation analysis. The preprocessed data is stored in a time-series database, supporting high-throughput writes and fast queries to meet the data analysis needs of subsequent modules.

[0034] This module provides high-quality foundational data for modules such as business intent modeling, fault prediction, and intelligent decision-making, ensuring the accuracy of subsequent analysis and decision-making results.

[0035] In this embodiment, the business intent modeling module is implemented as follows: The business intent modeling module is used to parse business event information and construct a business resource association graph to realize the association mapping between business intent and underlying resources. It includes a business event parsing unit, an association graph construction unit, and a business indicator association analysis unit.

[0036] The core of the business event analysis unit is to extract key information from the business calendar using natural language processing (NLP) technology. The business calendar is a document used by enterprises to record various business plans and system arrangements, including information related to various business events such as e-commerce promotional activities, system maintenance, data backups, and version updates. The implementation process of NLP technology includes word segmentation, part-of-speech tagging, entity recognition, and keyword extraction. First, the business calendar text is segmented into words, breaking down continuous text into independent words; then, part-of-speech tagging is performed to identify nouns, verbs, time words, etc.; next, entity recognition algorithms are used to identify key entities such as event type, time information, load, and related descriptions; finally, keyword extraction algorithms are used to extract event type, event time, and expected load features.

[0037] The extracted event types include categories such as promotional activities, system maintenance, data backup, and version updates; the extracted event time includes the start and end times of the event, accurate to the specific moment; the extracted expected load characteristics include three levels: high load, medium load, and low load. These levels are based on the description of the event scale in the business calendar and the load data of similar historical events. For example, promotional activities are usually marked as high load, routine system maintenance as medium load, and data backup as low load.

[0038] The association graph construction unit is used to establish the mapping relationship between business events and resource pools. This mapping relationship is achieved through a graph structure. Implementation. Vertex set. Includes business event nodes and resource nodes Business event nodes The attributes include event type, start time, end time, and load level. The event type corresponds to the result extracted by the business event parsing unit, the start and end times are the extracted time information, and the load level is the extracted expected load characteristic. Resource Node The attributes include resource type, resource identifier, and resource capacity. Resource type is divided into computing resources, storage resources, and network resources. Resource identifier is a unique identifier assigned to each resource by the cloud platform, and resource capacity is the maximum available quota of the corresponding resource.

[0039] edge set Includes directed edges connecting business event nodes and resource nodes. A directed edge points from a business event node to a resource node, indicating that the business event affects the corresponding resource. The weight of the directed edge... This represents the impact factor of business events on resources. It is calculated by statistically analyzing the changes in corresponding resource indicators during similar historical business events, and using the average of these changes as the initial weight. For example, if the average increase in CPU utilization of a certain type of computing resource during multiple historical promotional events is a certain value, this value will be used as the initial weight of the edge between the promotional event node and the computing resource node.

[0040] The business indicator correlation analysis unit is used to optimize the edge weights of the business resource correlation graph. Its core is to calculate the correlation coefficient matrix between business indicators and resource indicators, using the Pearson correlation coefficient formula, as follows:

[0041] in Represents a sequence of business metrics. Represents a sequence of resource indicators. Indicates the number of data points. This represents the mean of the business indicator series. This represents the mean of the resource indicator series. Indicates the first Individual business metrics values, Indicates the first Individual resource indicator values.

[0042] During the calculation process, the business indicator series and resource indicator series from the past period are first selected to determine the number of data points. Then calculate the mean of the business indicator series. and the mean of the resource indicator series Next, calculate the sum of the products of the deviations of each data point from its corresponding mean, and the sum of the squares of the deviations of the two sequences. Finally, substitute these values ​​into the formula to obtain the correlation coefficient. The correlation coefficient ranges from -1 to 1, and the larger the absolute value, the stronger the correlation between business indicators and resource indicators.

[0043] The business indicator correlation analysis unit adjusts the edge weights based on the correlation coefficient values. The larger the absolute value of the correlation coefficient, the larger the edge weight W, and vice versa. Through this adjustment process, the business resource correlation graph can more accurately reflect the correlation strength between business events and resources, providing a more precise basis for subsequent dynamic baseline generation and intelligent decision-making.

[0044] The purpose of this module is to transform business intent into relationships that the system can understand, enabling subsequent modules to analyze and make decisions in conjunction with the business context, thus avoiding misjudgments caused by relying solely on resource metrics.

[0045] In this embodiment, the dynamic baseline generation module is implemented as follows: The dynamic baseline generation module is used to generate dynamic threshold ranges with business context labels based on the business resource association graph, and adopts a multi-scenario baseline management method.

[0046] The categorization of scenarios is based on business activity patterns and system operating status, specifically including weekday scenarios, weekend scenarios, peak business scenarios, and system maintenance scenarios. Weekday scenarios correspond to the business operating status on normal working days, when the business load is relatively stable; weekend scenarios correspond to the business operating status on weekends, where the business load is usually lower than on weekdays; peak business scenarios correspond to the state when business events such as promotional activities occur, and the business load is high; system maintenance scenarios correspond to the state when system maintenance operations are performed, and some resources may be under low load or in a paused state.

[0047] The threshold calculation model for each scenario is based on time series analysis of historical data to obtain the normal fluctuation range. Specifically, for each scenario, historical resource indicator data from the past six months are selected and analyzed using a sliding window method. A fixed-length sliding window is set, and the historical data series is traversed. The mean and standard deviation of the data within each window are calculated. The mean plus or minus three times the standard deviation is taken as the normal fluctuation range corresponding to that window. The normal fluctuation ranges of all windows are statistically merged to obtain the dynamic threshold range of resource indicators for that scenario.

[0048] The dynamic baseline generation module automatically switches to the corresponding scenario baseline based on the current business event. Scenario switching is achieved through event type matching in the business resource association graph. The system monitors the status of business event nodes in the business resource association graph in real time. When a business event is detected to be starting, the system extracts the event type and matches the corresponding scenario baseline based on the event type. For example, if the event type is detected as a promotional activity, it automatically switches to the peak business scenario baseline; if the event type is detected as system maintenance, it automatically switches to the system maintenance scenario baseline; if there is no specific business event, it determines whether it is a weekday or weekend scenario based on the current time and switches to the corresponding baseline.

[0049] The purpose of this module is to provide dynamic judgment criteria for the fault prediction module, enabling the system to distinguish between normal resource fluctuations within the business plan and true abnormal precursors, thereby reducing invalid alarms.

[0050] In this embodiment, the fault prediction module is implemented as follows: The fault prediction module is used to predict potential fault modes based on resource indicator data and business indicator data. It uses a time-series prediction algorithm to analyze the resource indicator sequence, and the core algorithm is a long short-term memory network model.

[0051] The network structure of a Long Short-Term Memory (LSTM) network model includes an input layer, hidden layers, and an output layer. The input feature vector of the input layer is... The system contains preprocessed resource and business indicator data, with each indicator serving as a feature dimension. The dimension of the input feature vector equals the total number of indicator types. The hidden layer contains multiple hidden units, each whose state is determined by both the cell state and the hidden state. Selective memorization and forgetting of information are achieved through forgetting gates (input and output). The output layer outputs a sequence of predicted resource indicator values ​​and a fault probability distribution for future time periods.

[0052] The forward computation process of the Long Short-Term Memory (LSTM) network model is as follows: Calculation of the output vector of the forget gate:

[0053] Calculation of the input gate output vector:

[0054] Calculation of candidate cell state vectors:

[0055] Update of cell state vector:

[0056] Calculation of the output vector of the output gate:

[0057] Update of hidden state vector:

[0058] in Indicates time step The hidden state vector is initialized to an all-zero vector; Indicates time step The cell state vector is initialized to an all-zero vector; , , , The weight matrices representing the input gate, candidate cell state, and output gate of the forget gate are respectively initialized using the Xavier initialization method. , , , These represent the bias vectors of the forget gate, the input gate, the candidate cell state, and the output gate, respectively, initialized to zero vectors; This represents the sigmoid activation function, whose output ranges from 0 to 1, and is used to control the degree to which each gate is open or closed. This represents the hyperbolic tangent activation function, whose output ranges from -1 to 1, and is used to generate candidate cell states and hidden states. This indicates element-wise multiplication.

[0059] The training process of the Long Short-Term Memory (LSTM) network model is as follows: Historical indicator data from the past year is selected as training data and divided into training and validation sets in chronological order. The training set accounts for 80% of the total data, and the validation set accounts for 20%. Mean squared error (MSE) is used as the loss function to measure the difference between the model's predicted values ​​and the actual values. The Adam optimizer is used to minimize the loss function, and the learner rate of the optimizer is adaptively adjusted. During training, the weight matrix and bias vector are iteratively updated until the loss function value of the validation set tends to stabilize, at which point training stops.

[0060] After model training is complete, inputting time-series data up to the current moment will output a sequence of predicted resource indicators for a preset future time period. Simultaneously, by calculating the deviation between the predicted values ​​and a threshold range generated by the dynamic baseline, and using the sigmoid function to map the deviation to a range of 0 to 1, a fault probability distribution is obtained. When the fault probability exceeds a set threshold, a potential fault mode is identified.

[0061] The purpose of this module is to identify potential faults in the server cluster in advance, provide a basis for prediction for the intelligent decision-making module, and enable proactive intervention in faults.

[0062] In this embodiment, the intelligent decision-making module is implemented as follows: The intelligent decision-making module is used to generate self-healing strategies based on prediction results and business resource correlation maps, including a root cause localization unit and a strategy generation unit.

[0063] The root cause localization unit determines the root cause of anomalies through multi-dimensional data fusion analysis. This multi-dimensional data includes resource indicator data, business indicator data, and business resource association graph information. Specifically, the process involves first acquiring abnormal resource indicator data corresponding to potential fault modes output by the fault prediction module, and combining this with concurrent business indicator data to analyze the business operation status at the time of the anomaly. Then, the business resource association graph is queried to obtain the business event nodes and associated resource nodes corresponding to this business status, as well as edge weight information. Finally, a causal inference algorithm is used, combined with the magnitude of the edge weights, to identify the business events and resource nodes that contribute the most to the anomaly, thus determining the root cause. For example, when a potential fault with persistently high CPU utilization is detected, combined with business indicator data revealing a surge in concurrent users, and the association graph indicating the presence of a promotional event with a large edge weight between this event and that type of CPU resource, the root cause is determined to be insufficient CPU resources due to the promotional event.

[0064] The strategy generation unit generates self-healing strategies based on fault probability distribution and business impact assessment. These strategies include instance migration, resource adjustment, and service restart strategies. Instance migration strategies are suitable for scenarios where nodes are overloaded but other nodes have idle resources, migrating service instances from overloaded nodes to idle nodes. Resource adjustment strategies are suitable for scenarios where a certain type of resource is insufficient overall, increasing the quota for that type of resource through the cloud platform interface. Service restart strategies are suitable for scenarios where stateless services experience resource leaks or other anomalies, releasing occupied resources by restarting the service.

[0065] The strategy generation unit selects the optimal self-healing strategy by calculating the cost function value of each strategy. The cost function is a weighted sum of resource utilization, service level agreement compliance, and operational risk coefficient. Resource utilization measures the efficiency of resource utilization after strategy execution; service level agreement compliance measures whether the service quality meets the agreed requirements after strategy execution; and the operational risk coefficient measures the probability of anomalies occurring during strategy execution. The expression for the cost function is:

[0066] in , , These are weighting coefficients, set according to business needs and system characteristics, and , , The sum is 1. Service level agreement compliance typically has the highest weighting, prioritizing ensuring service quality.

[0067] The strategy generation unit calculates the cost function value of each candidate self-healing strategy and selects the strategy with the smallest cost function value as the optimal self-healing strategy.

[0068] The role of this module is to generate accurate self-healing strategies based on fault prediction results and business context, ensuring the effectiveness and rationality of the strategies.

[0069] In this embodiment, the specific implementation of the secure execution module is as follows: The security execution module is used to call the cloud platform interface to execute the self-healing strategy and achieve lossless operation of service instances, including the operation orchestration unit and the circuit breaker protection unit.

[0070] The operation orchestration unit decomposes the self-healing strategy into an ordered sequence of atomic operations. These atomic operations include instance creation, instance destruction, service registration, and service discovery. Different self-healing strategies correspond to different atomic operation sequences. For example, the atomic operation sequence for the instance migration strategy is: service registration pause, instance creation, data synchronization, traffic switching, original instance destruction, service registration recovery; the atomic operation sequence for the resource adjustment strategy is: resource quota application, resource creation, resource configuration, and service adaptation; and the atomic operation sequence for the service restart strategy is: service registration pause, traffic switching, service restart, service verification, traffic recovery, and service registration recovery. The execution of atomic operations is achieved by calling standard APIs provided by the cloud platform, ensuring the compatibility and reliability of the operations.

[0071] The circuit breaker protection unit monitors the execution status of atomic operations to prevent system failures from escalating due to operational anomalies. It acquires the execution result of each atomic operation in real time, including statuses such as successful execution, execution failure, timeout, and no response. When an abnormal execution pattern is detected, the circuit breaker mechanism is triggered. Abnormal execution patterns include consecutive operation failures, resource allocation timeouts, and service response delays.

[0072] Continuous operation failure refers to the failure of multiple executions of the same atomic operation; resource allocation timeout refers to the failure to receive a response after a preset time has elapsed since the resource allocation interface was called; service response delay refers to the response time of the business service exceeding a set threshold during policy execution. The circuit breaker protection unit selects different handling methods according to the severity of the anomaly: for minor anomalies, subsequent atomic operations continue to be executed; for moderate anomalies, execution is paused and retried after a preset time; for severe anomalies, automatic execution is stopped and manual intervention mode is switched to, and operations and maintenance personnel are notified to handle the situation.

[0073] The purpose of this module is to ensure the safe execution of the self-healing strategy, avoid impacting business services during operation, and achieve lossless operation and maintenance.

[0074] In this embodiment, the feedback learning module is implemented as follows: The feedback learning module is used to collect execution result data of the self-healing strategy and optimize system performance. The collected execution result data includes strategy effectiveness indicators and system stability indicators. Strategy effectiveness indicators include whether the fault has been resolved, whether resource indicators have returned to normal, and the time consumed by the self-healing operation; system stability indicators include whether services were interrupted during operation execution, the fluctuation range of business response time, and whether user experience was affected.

[0075] The feedback learning module uses reinforcement learning algorithms to update the policy generation rules of the intelligent decision-making module. The state space of the reinforcement learning algorithm includes system resource states and business event states. System resource states are described by indicators such as the utilization rate of each resource, while business event states are described by information such as whether a business event has occurred, its type, and load level. The action space contains a set of all self-healing strategies. The reward function is calculated based on the system performance improvement after strategy execution: System performance improvement = System performance improvement after strategy execution - System performance loss during strategy execution. The performance improvement includes resource indicator recovery degree and fault resolution effect, while the performance loss includes operation time and resource consumption.

[0076] Reinforcement learning algorithms iteratively learn and update the weight coefficients of the cost function and the policy selection rules in the intelligent decision-making module, making the generated self-healing strategy more closely match the actual operating state of the system. Simultaneously, the feedback learning module periodically retrains the time-series prediction algorithm model of the fault prediction module, typically once a month, adding the latest historical data to the training set and updating the model's weight matrix and bias vector to improve the model's prediction accuracy and generalization ability.

[0077] The purpose of this module is to enable the system to self-optimize and continuously improve the accuracy of fault prediction and the effectiveness of self-healing strategies.

[0078] The workflow of the intelligent operation and maintenance management system in this embodiment is as follows: The first step is to continuously collect resource and business indicator data from the server cluster, and store them in the time-series database after preprocessing. The second step, the business intent modeling module, extracts key information from the business calendar through the business event parsing unit, establishes a mapping relationship between business events and resource pools through the association graph construction unit, forms a business resource association graph, and then calculates the correlation coefficient matrix through the business indicator association analysis unit to optimize the edge weights of the association graph. The third step, the dynamic baseline generation module, automatically switches to the corresponding scenario baseline based on the current business event type in the business resource association graph, and generates a dynamic threshold range with business context labels. The fourth step, the fault prediction module, takes in preprocessed resource indicator data and business indicator data, and outputs a sequence of predicted resource indicator values ​​and a fault probability distribution for future time periods through a trained long short-term memory network model. When the fault probability exceeds a set threshold, it is determined that there is a potential fault mode. The root cause localization unit of the fifth intelligent decision-making module combines resource indicator data, business indicator data, and business resource correlation graph information to determine the root cause of the anomaly; the strategy generation unit generates multiple candidate self-healing strategies based on the fault probability distribution and business impact assessment, calculates the cost function value of each strategy, and selects the optimal self-healing strategy. The sixth step of the security execution module's operation orchestration unit decomposes the optimal self-healing strategy into a sequence of atomic operations and calls the cloud platform interface to execute them sequentially; the circuit breaker protection unit monitors the execution status of the atomic operations in real time and takes appropriate measures based on the severity of the anomaly. The seventh step, the feedback learning module, collects the execution result data of the self-healing strategy, uses reinforcement learning algorithms to update the strategy generation rules of the intelligent decision-making module, and periodically retrains the long short-term memory network model of the fault prediction module to complete the system's self-optimization.

[0079] The above process is executed cyclically to achieve closed-loop operation and maintenance management from resource monitoring to predictive self-healing.

[0080] The intelligent operation and maintenance management system for server clusters in the cloud computing environment of this embodiment realizes the association mapping between business and resources through the business intent modeling module, enabling the system to understand business intent, effectively distinguish between normal business load fluctuations and abnormal precursors, and reduce invalid alarms; through the long short-term memory network model of the fault prediction module, it can identify potential faults in advance, realizing the transformation from passive response to proactive intervention; through the intelligent decision-making module to generate the optimal self-healing strategy, combined with the non-destructive operation mechanism of the safe execution module, the safety and effectiveness of self-healing operations are ensured; through the continuous optimization of the feedback learning module, the adaptability and stability of the system are improved.

[0081] This system can improve the intelligence level of server cluster operation and maintenance management in a cloud computing environment, reduce operation and maintenance costs, and improve service reliability.

[0082] Example 2 like Figure 3 As shown in Example 1, this example details the specific steps of a server cluster intelligent operation and maintenance management system in a cloud computing environment when performing operation and maintenance management tasks. It can be understood that the workflow described in this example is based on the system architecture and specific implementation methods of each module described in Example 1. Its core lies in achieving closed-loop operation and maintenance management through business intent perception and predictive self-healing mechanisms. Those skilled in the art can fully reproduce this workflow based on the following description.

[0083] Specifically, the system workflow includes the following steps: The first step involves the data acquisition module continuously collecting resource and business metrics data from the server cluster. Resource metrics include computing resource metrics such as CPU utilization and memory usage, storage resource metrics such as disk utilization and disk I / O rate, and network resource metrics such as network bandwidth utilization and network latency. Business metrics include transaction success rate, request response time, concurrent users, and business throughput. The data collection method combines agent-based collection with API calls. Agents are deployed on each node in the cluster to collect local resource metrics, while business metrics are obtained by calling business system monitoring interfaces or parsing business logs. The collected raw data undergoes data cleaning, normalization, and alignment preprocessing before being stored in a time-series database. This step provides high-quality, real-time, and consistent foundational data support for subsequent modules.

[0084] The second step involves the business intent modeling module parsing business event information and constructing a business resource association graph. The business event parsing unit extracts event types, event times, and expected load characteristics from the business calendar using natural language processing. The association graph construction unit establishes a mapping relationship between business event nodes and resource nodes based on the extraction results. This mapping relationship is represented by a graph structure, where vertices include both business event nodes and resource nodes, and edges are directed with weights representing the impact factors of business events on resources. The business indicator association analysis unit calculates the Pearson correlation coefficient matrix between business indicators and resource indicators and optimizes the edge weights in the graph based on the correlation coefficient. This step realizes the association mapping from business intent to system resources, providing business context for subsequent analysis.

[0085] The third step involves the dynamic baseline generation module generating dynamic threshold ranges with business context labels based on the business resource association graph. This module employs a multi-scenario baseline management approach, covering weekday scenarios, weekend scenarios, peak business scenarios, and system maintenance scenarios. For each scenario, an independent threshold calculation model is established based on time-series analysis of historical data to obtain the dynamic threshold range for resource indicators. The module automatically switches to the corresponding scenario baseline based on the current business event type; for example, when the business event is a promotional activity, it switches to the peak business scenario baseline. This step provides a dynamic standard for fault diagnosis, effectively distinguishing between planned business fluctuations and genuine anomalies.

[0086] The fourth step involves the fault prediction module, which predicts potential fault modes based on resource and business indicator data. This module employs a Long Short-Term Memory (LSTM) network model as its time-series prediction algorithm. It takes a pre-processed sequence of indicator data as input and outputs a sequence of predicted resource indicator values ​​and a fault probability distribution for future time periods through forward computation. When the fault probability exceeds a preset threshold, a potential fault is identified. This step enables proactive fault identification, providing a basis for active intervention.

[0087] The fifth step involves the intelligent decision-making module generating a self-healing strategy based on the prediction results and the business resource correlation graph. The root cause localization unit identifies the root cause of the anomaly through multi-dimensional data fusion analysis, including resource indicator data, business indicator data, and business resource correlation graph information. The strategy generation unit generates candidate self-healing strategies based on the fault probability distribution and business impact assessment. Strategy types include instance migration strategies, resource adjustment strategies, and service restart strategies. This unit calculates the cost function value of each strategy, which is a weighted sum of resource utilization, service level agreement compliance, and operational risk coefficients, and selects the strategy with the lowest cost as the optimal self-healing strategy. This step ensures the accuracy and business adaptability of the self-healing strategy.

[0088] Step six involves the security execution module invoking the cloud platform interface to execute the self-healing strategy and achieve lossless operation of service instances. The operation orchestration unit decomposes the self-healing strategy into an ordered sequence of atomic operations, including instance creation, instance destruction, service registration, and service discovery. The circuit breaker protection unit monitors the execution status of the atomic operations. When it detects abnormal patterns such as consecutive operation failures, resource allocation timeouts, or service response delays, it triggers the circuit breaker mechanism and selects to continue execution, suspend execution, or switch to manual intervention based on the severity of the anomaly. This step ensures the security and reliability of the self-healing operation.

[0089] Step seven involves the feedback learning module collecting data on the execution results of the self-healing strategy and optimizing system performance. This data includes strategy effectiveness metrics and system stability metrics. The module uses reinforcement learning algorithms to update the strategy generation rules of the intelligent decision-making module. The reinforcement learning state space includes system resource states and business event states, the action space includes a set of self-healing strategies, and the reward function is calculated based on the system performance improvement after strategy execution. Simultaneously, the module periodically retrains the time-series prediction algorithm model of the fault prediction module. This step enables continuous self-optimization of the system, improving prediction accuracy and strategy effectiveness.

[0090] The above steps are executed cyclically to form a complete closed loop from data collection to feedback learning, thereby realizing the intelligent and automated operation and maintenance management of server clusters in a cloud computing environment.

[0091] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A server cluster intelligent operation and maintenance management system in a cloud computing environment, characterized in that, It includes a data acquisition module, a business intent modeling module, a dynamic baseline generation module, a fault prediction module, an intelligent decision-making module, and a security execution module: The data acquisition module is used to collect resource indicator data and business indicator data of the server cluster; The business intent modeling module is used to parse business event information and construct a business resource association graph. The dynamic baseline generation module is used to generate dynamic threshold ranges with business context labels based on the business resource association graph; The fault prediction module is used to predict potential fault modes based on resource indicator data and business indicator data. The intelligent decision-making module is used to generate a self-healing strategy based on the prediction results and the business resource association graph. The security execution module is used to call the cloud platform interface to execute the self-healing strategy and achieve lossless operation of service instances.

2. The intelligent operation and maintenance management system for server clusters in a cloud computing environment according to claim 1, characterized in that, The business intent modeling module includes a business event parsing unit and an association graph construction unit; The business event parsing unit extracts event type, event time, and expected load characteristics from the business calendar using natural language processing technology; The association graph construction unit establishes a mapping relationship between business events and resource pools, and the mapping relationship is represented as a graph structure. , where the vertex set Includes business event nodes and resource nodes edge set Includes directed edges connecting business event nodes and resource nodes. ; The business event node The attributes include event type. Start time End time and load level ; The resource node The attributes include resource type Resource Identifier and resource capacity ; The directed edge weight This represents the impact factor of business events on resources, calculated through historical data analysis.

3. The intelligent operation and maintenance management system for server clusters in a cloud computing environment according to claim 1, characterized in that, The fault prediction module uses a time-series prediction algorithm to analyze resource indicator sequences. This time-series prediction algorithm includes a Long Short-Term Memory (LSTM) network model, and the forward computation process of the LSTM network model is expressed as follows: in Indicates time step The input feature vector contains resource indicator data and business indicator data. Indicates time step The hidden state vector, Indicates time step The cell state vector, This represents the output vector of the forget gate. This represents the input gate output vector. This represents the output vector of the output gate. Represents the candidate cell state vector. , , , Represents the weight matrix. , , , This represents the bias vector. This represents the sigmoid activation function. Represents the hyperbolic tangent activation function; The Long Short-Term Memory (LSTM) network model outputs a sequence of predicted resource indicators and a fault probability distribution for future time periods.

4. The intelligent operation and maintenance management system for server clusters in a cloud computing environment according to claim 1, characterized in that, The intelligent decision-making module includes a root cause localization unit and a strategy generation unit; The root cause localization unit determines the root cause of the anomaly through multi-dimensional data fusion analysis, and the multi-dimensional data includes resource indicator data, business indicator data, and business resource correlation map information. The strategy generation unit generates self-healing strategies based on fault probability distribution and business impact assessment. The self-healing strategies include instance migration strategies, resource adjustment strategies, and service restart strategies. The strategy generation unit calculates the cost function value of each strategy. The cost function is a weighted summation function of resource utilization, service level agreement compliance and operational risk coefficient. The strategy generation unit selects the strategy with the minimum cost function value as the optimal self-healing strategy.

5. The intelligent operation and maintenance management system for server clusters in a cloud computing environment according to claim 1, characterized in that, The safety execution module includes an operation orchestration unit and a fuse protection unit; The operation orchestration unit decomposes the self-healing strategy into an ordered sequence of atomic operations, including instance creation, instance destruction, service registration, and service discovery. The circuit breaker protection unit monitors the execution status of atomic operations and triggers the circuit breaker mechanism when an abnormal execution mode is detected. The abnormal execution modes include continuous operation failures, resource allocation timeouts, and service response delays. The circuit breaker protection unit selects to continue execution, suspend execution, or switch to manual intervention mode based on the severity of the abnormality.

6. The intelligent operation and maintenance management system for server clusters in a cloud computing environment according to claim 1, characterized in that, The dynamic baseline generation module adopts a multi-scenario baseline management method, which includes weekday scenarios, weekend scenarios, peak business scenarios, and system maintenance scenarios. The dynamic baseline generation module establishes an independent threshold calculation model for each scenario, and the threshold calculation model obtains the normal fluctuation range based on time series analysis of historical data. The dynamic baseline generation module automatically switches the corresponding scenario baseline based on the current business event. The scenario switching is achieved by matching the event type in the business resource association graph.

7. The intelligent operation and maintenance management system for server clusters in a cloud computing environment according to claim 1, characterized in that, The system also includes a feedback learning module, which collects execution result data of the self-healing strategy, including strategy effectiveness indicators and system stability indicators. The feedback learning module uses a reinforcement learning algorithm to update the strategy generation rules of the intelligent decision-making module. The state space of the reinforcement learning algorithm includes the system resource state and the business event state, the action space includes a set of self-healing strategies, and the reward function is calculated based on the improvement of system performance after the strategy is executed. The feedback learning module periodically retrains the time-series prediction algorithm model of the fault prediction module.

8. The intelligent operation and maintenance management system for server clusters in a cloud computing environment according to claim 1, characterized in that, The business intent modeling module also includes a business indicator correlation analysis unit. This unit calculates the correlation coefficient matrix between business indicators and resource indicators. The correlation coefficient matrix is ​​used to optimize the edge weights of the business resource correlation graph. The correlation coefficient is calculated using the Pearson correlation coefficient formula. in Represents a sequence of business metrics. Represents a sequence of resource indicators. Indicates the number of data points. This represents the mean of the business indicator series. This represents the mean of the resource indicator series. Indicates the first Individual business metrics values, Indicates the first Individual resource indicator values; The business indicator correlation analysis unit adjusts the weight of the impact factor of business events on resources based on the correlation coefficient value.

Images