Abnormality analysis method combining fault propagation path and topological structure

Abstract

The invention provides an anomaly analysis method combining a fault propagation path and a topological structure, and the method comprises the steps: calculating an overall recovery time estimation value under different topological structures according to the business importance and a recovery time target, and obtaining a candidate topology scheme set meeting the recovery time requirement; for the candidate topology scheme set, key nodes and key links are extracted, a fault scene is simulated based on component fault rate and link fault rate parameters, a fault propagation path is analyzed, and the influence of faults on service quality recovery time is evaluated; obtaining an optimal topology scheme considering the recovery time and the dependency strength constraint through the service recovery time output by the service recovery time model; in the solving process, load balancing capacity and fault-tolerant backup capacity are introduced as constraint conditions.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to an anomaly analysis method that combines fault propagation paths and topology. Background Technology

[0002] When any critical node or link in a network fails, its impact is no longer limited to a local area but can rapidly spread throughout the entire network, triggering a chain reaction and causing widespread system paralysis, thus prolonging the actual recovery period. Even more challenging is that in highly interconnected and interdependent network environments, the difficulty of fault location and root cause analysis increases exponentially. Traditional fault diagnosis methods are inadequate when faced with such complex network structures. Maintenance personnel often struggle to quickly pinpoint the root cause of problems when confronted with intricate fault propagation paths, undoubtedly further delaying recovery time and contradicting the initial optimized design. Should network designers favor high redundancy and complex topologies for rapid recovery, or should they opt for simpler network structures to mitigate the risk of fault propagation? How to ensure high network availability while avoiding chain reactions and achieving accurate fault location and efficient recovery has become a pressing technical problem in network operations and maintenance. Summary of the Invention

[0003] This invention provides an anomaly analysis method that combines fault propagation paths and topology, mainly including: Obtain the business importance and recovery time target, and calculate the overall recovery time estimate under different topologies; Based on the comparison between the estimated overall recovery time and a preset threshold, a set of candidate topology schemes is selected. For the set of candidate topology schemes, analyze the fault propagation path and evaluate the impact of the fault on service quality recovery time; By combining the dependency strength attribute to determine the scope of the fault impact, high-risk solutions are eliminated to form a set of preferred topology solutions; Construct a business dependency graph, identify key dependency paths and calculate component dependency strength, and establish a business recovery time model; The recovery time is output through the business recovery time model, and the optimal topology scheme is determined by combining the constraints. Assess the impact of the fault on the quality of business services based on the optimal topology scheme, and formulate mitigation measures for critical fault propagation paths.

[0004] Furthermore, the process of obtaining business importance and recovery time targets, and calculating the overall recovery time estimate under different topologies, includes: The weights of each time item are determined based on the importance of the business, and the overall recovery time estimate is calculated using a weighted summation method. If the estimated total recovery time is less than the preset recovery time threshold, the corresponding topology will be added to the candidate topology scheme set. For each of the candidate topology schemes in the set of candidate topologies, a preliminary screening is performed based on network bandwidth and latency factors. The selected solution set was simulated and tested. The probability distribution of actual recovery time was obtained through multiple simulations to verify whether it met the expected goals.

[0005] Furthermore, the step of analyzing the fault propagation path and evaluating the impact of the fault on service quality recovery time for the candidate topology scheme set includes: Identify key nodes and key links using node importance calculation methods; Obtain the failure rate parameters of the key nodes and the key links, and construct a fault tree model to generate multiple sets of failure scenarios; For each set of fault scenarios, a search algorithm is used to simulate the fault propagation process and record the propagation path and scope of impact; Based on propagation delay and information about affected nodes, estimate service quality recovery time, calculate average recovery time and distribution characteristics, and assess the impact of the fault on service quality recovery time.

[0006] Furthermore, the step of combining the dependency strength attribute to determine the scope of the fault impact and eliminating high-risk solutions to form a set of preferred topology solutions includes: Obtain data on the coupling degree and dependency degree between components, and calculate the dependency strength attribute value using a weighted summation method; A judgment matrix is ​​constructed based on the dependency strength attribute value, and the scope of the fault impact is calculated. Retrieve a list of key business nodes from a preset database and determine whether the fault propagation path involves the key business nodes; If the scope of the fault exceeds a preset threshold or involves critical business nodes, it is determined to be a high-risk solution and eliminated.

[0007] Furthermore, the construction of the business dependency graph, identification of key dependency paths, and calculation of component dependency strength includes: The business dependency graph is constructed using a graph database, where nodes represent business components and edges represent dependencies. The weights of the edges are calculated based on the frequency of calls and the degree of resource sharing, and the critical path method is used to identify the critical dependency paths. For each of the preferred topology schemes, the component dependency strength is calculated, the comprehensive dependency relationship is evaluated based on multi-dimensional similarity, and the business recovery time model is established.

[0008] Furthermore, the step of determining the optimal topology scheme by outputting the recovery time through the business recovery time model and combining it with constraints includes: Based on the output of the business recovery time model, a multi-objective optimization problem is constructed, with the objectives including minimizing the recovery time and the component dependency strength. Load balancing and fault-tolerant backup capabilities are introduced as constraints, and load difference thresholds and backup redundancy requirements are set. The multi-objective optimization problem is solved using an optimization algorithm, and the quality of candidate solutions is evaluated using a fitness function. A comprehensive analysis of the non-dominated solution set is conducted to select the optimal topology scheme.

[0009] Furthermore, the step of assessing the impact of faults on service quality based on the optimal topology scheme includes: Based on the node connection relationships in the topology, obtain the service continuity parameters of the nodes on the fault path; The number of affected services is calculated by comparing the service quality difference before and after the failure. A risk score is calculated based on business priority and failure probability. If the risk score exceeds a preset threshold, the high-risk business location is marked. By identifying the intersection of high-risk business locations and fault paths, a set of critical paths is determined, the speed of fault propagation and the duration of its impact are analyzed, and a fault isolation boundary is constructed.

[0010] Furthermore, the formulation of mitigation measures targeting critical failure propagation paths includes: Key connection points are identified based on the impact score and the dependency strength attribute value; Determine the priority of interface adjustments by using the ratio of data flow throughput to response time; To address bottleneck interfaces, a modification plan was developed, including adding middleware and asynchronous processing mechanisms, and updating data flow forwarding rules. Adjust load balancing parameters based on load monitoring data, configure fault detection frequency and recovery mechanism, and ensure business continuity through primary / backup switching or service degradation.

[0011] Furthermore, mitigation measures for critical fault propagation paths include adjusting interfaces and data flows between highly dependent components, optimizing load balancing strategies, and improving fault detection and recovery mechanisms.

[0012] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: This invention discloses an anomaly analysis method combining fault propagation paths and topology, addressing the complex requirements of business systems in fault scenarios for recovery time and service quality assurance. It integrates key issues such as recovery time targets, fault propagation path analysis, and dependency optimization. This invention calculates estimated recovery times under different topologies, filters candidate solutions that meet the requirements, and simulates fault scenarios using failure rate parameters to analyze propagation paths and impact ranges, eliminating high-risk solutions to form an optimal set. Simultaneously, it constructs a business dependency graph, identifies critical paths, and determines the optimal topology solution by combining recovery time models with load balancing and fault-tolerant backup constraints. Furthermore, it quantifies the impact of faults on service quality through simulation and historical data, and formulates targeted mitigation measures, such as optimizing interfaces and data flows and improving detection and recovery mechanisms. By dynamically balancing recovery time, dependency strength, and fault impact, it ensures business continuity and system stability, significantly improving the recovery efficiency and service quality assurance capabilities of critical businesses in fault scenarios. Attached Figure Description

[0013] Figure 1 This is a flowchart of an anomaly analysis method combining fault propagation path and topology structure according to the present invention.

[0014] Figure 2 This is a schematic diagram of an anomaly analysis method combining fault propagation path and topology structure according to the present invention. Detailed Implementation

[0015] The technical solutions of the embodiments of the present invention will be clearly and thoroughly described below with reference to the accompanying drawings. The described embodiments are merely some embodiments of the present invention.

[0016] like Figures 1-2 This embodiment of an anomaly analysis method combining fault propagation path and topology may specifically include: Step S101: Based on the business importance and recovery time target, calculate the overall recovery time estimate under different topology structures, and obtain a set of candidate topology schemes that meet the recovery time requirements.

[0017] To determine the business importance and recovery time target, a weighted summation method is used to calculate the estimated overall recovery time for different topologies. The overall recovery time includes fault detection time, location time, isolation time, system restart time, data recovery time, and business recovery time, with the weight of each time item determined based on business importance. The estimated overall recovery time is compared to a preset recovery time threshold. Multiple recovery time thresholds are set, corresponding to different levels of business importance. If the estimated overall recovery time is less than the preset recovery time threshold, the corresponding topology is added to the candidate topology scheme set. For each topology scheme in the candidate topology scheme set, a decision tree algorithm is used to filter based on network bandwidth, latency, and cost factors, resulting in a filtered set of candidate topologies. The filtered candidate topology scheme set is then simulated to verify whether the actual recovery time meets expectations. Multiple simulations are performed using Monte Carlo simulation tools to derive the candidate topology scheme set, and the probability distribution of the actual recovery time for each scheme is calculated.

[0018] The importance of the business and the recovery time target are determined. For example, the platform's order system has an importance score of 9 out of 10, and a recovery time target of 5 minutes. A weighted summation method is used to calculate the overall recovery time estimate, with weights allocated as follows: fault detection time 0.1, location time 0.2, isolation time 0.1, system restart time 0.2, data recovery time 0.3, and business recovery time 0.1. For a star topology, the calculated overall recovery time estimate is 4.5 minutes. Multi-level recovery time thresholds are set: 9-10 points correspond to 5 minutes, 7-8 points to 10 minutes, and 5-6 points to 15 minutes. If the estimated recovery time of 4.5 minutes is less than the threshold of 5 minutes, the star topology is added to the candidate solution set. Candidate solutions are then screened, considering factors such as network bandwidth of 100Mbps, latency of 20ms, and cost of 5000 yuan / month, using the C4.5 decision tree algorithm for evaluation. The decision tree is constructed based on the training dataset of historical topology solutions. Feature engineering converts network bandwidth, latency, and cost into numerical features, and the information gain ratio is used as the splitting criterion. The model input consists of feature vectors for each solution, and the output is a classification label indicating whether to retain the solution. The decision tree uses recovery time as the root node, network bandwidth, latency, and cost as child nodes, and leaf nodes indicating whether to retain the solution. After selection by the decision tree, two candidate solutions are retained: star topology and ring topology. Monte Carlo simulation is used to perform 10,000 simulation tests on the selected solutions to simulate actual recovery times under different fault conditions. The average recovery time for star topology is 4.8 minutes with a standard deviation of 0.5 minutes; the average recovery time for ring topology is 5.2 minutes with a standard deviation of 0.6 minutes. Based on the simulation results, the probability of recovery within 5 minutes is calculated to be 65% for star topology and 55% for ring topology. Star topology is selected as the optimal solution and included in the candidate topology set.

[0019] Step S102: For the candidate topology scheme set, extract key nodes and key links, simulate failure scenarios based on component failure rate and link failure rate parameters, analyze failure propagation paths, and evaluate the impact of failures on service quality recovery time.

[0020] Based on the topology of the candidate topology scheme set, the importance of nodes is calculated using the betweenness centrality algorithm. Nodes with importance indices higher than a preset threshold are selected as critical nodes, and the links connecting these critical nodes are identified as critical links. The component failure rates of the critical nodes and the link failure rates of the critical links are obtained, and a binary tree-structured fault tree model is constructed. Multiple fault scenarios are generated using the inverse transform sampling method, with each scenario containing randomly selected faulty components and links. For each fault scenario, a depth-first search algorithm is used to simulate the fault propagation process in the network, recording the fault propagation path, calculating the number of affected nodes and the impact range, and estimating the propagation delay based on link length and transmission rate. Based on the fault propagation path and affected node information, the M / M / 1 queuing model is used to estimate the service quality recovery time for each fault scenario. The recovery process is simulated using an exponential distribution, and the average recovery time and standard deviation are obtained through multiple iterations. recovery =1 / (μ-λ)+ρ / (1-ρ)*1 / μ,T recoveryLet μ represent the average service recovery time based on the M / M / 1 queuing model, λ represent the fault repair service rate, λ represent the fault arrival rate, and ρ represent the system utilization rate, which is equal to λ / μ. This formula calculates the expected recovery time under fault scenarios using queuing theory. Based on the recovery time distribution, the recovery time is calculated to assess the impact of faults on service quality recovery time. Specifically, for the network topology of a cloud service provider, the betweenness centrality algorithm is used to calculate node importance, setting an importance threshold of 0.5, selecting 5 critical nodes and 6 critical links. The component failure rate of critical nodes is set to 0.01 times / year, and the failure rate of critical links is set to 0.005 times / year. A binary tree fault tree model is constructed, containing 20 basic events and 15 logic gates. 1000 sets of fault scenarios are generated using the inverse transform sampling method, with 1-3 faulty components and 0-2 faulty links randomly selected in each set. For each set of fault scenarios, a depth-first search algorithm is used to simulate fault propagation, and the propagation path is recorded. Assuming an average link length of 100km and a transmission rate of 200km / ms, the estimated average propagation delay is 0.5ms. The number of affected nodes ranges from 2 to 15, with an average impact range of 30% of the total network nodes. Using an M / M / 1 queuing model, assuming an average service rate μ = 10 times / minute and an arrival rate λ = 8 times / minute, the service quality recovery time for each failure scenario is estimated. Simulating 1000 recovery processes using an exponential distribution, the average recovery time is calculated to be 15 minutes with a standard deviation of 3 minutes. Based on the recovery time distribution, a 90% probability recovery time of 19 minutes and a 99% probability recovery time of 23 minutes are calculated, thus assessing the impact of the failure on service quality recovery time.

[0021] Step S103: Combine the coupling degree between components, the degree of data dependence, and the degree of interface dependence to determine the scope of the fault impact. If the scope of the fault impact exceeds the preset threshold or the fault propagation path involves key business nodes, the topology scheme is determined to be a high-risk scheme and is removed from the candidate topology scheme set to form the preferred topology scheme set.

[0022] For each candidate topology scheme, the degree of coupling between components, the degree of data dependency, and the degree of interface dependency are obtained. A weighted summation method is used to calculate the dependency strength attribute value, with weights set to 0.4, 0.3, and 0.3 based on historical data analysis results. The dependency strength attribute value is calculated using the formula: DS = 0.4CC + ​​0.3DD + 0.3*ID, where DS is the dependency strength, CC is the degree of coupling between components, DD is the degree of data dependency, and ID is the degree of interface dependency. Based on the dependency strength attribute value, a judgment matrix is ​​constructed, a weight vector is calculated, and the eigenvector is obtained using the power method. The fault impact range is obtained through matrix iteration. A list of key business nodes is obtained from a pre-established business importance database. Business importance is assessed based on business revenue contribution and user impact range, and the database is updated and maintained using real-time business monitoring data. It is determined whether the fault propagation path involves key business nodes; if so, the degree of impact on those nodes is recorded. If the fault impact range exceeds a preset range for the total number of nodes or if a key business node is affected at a level higher than the preset range, the topology scheme is determined to be a high-risk scheme and removed from the candidate topology scheme set. The remaining schemes form the preferred topology scheme set. The set of preferred topology schemes is sorted, and the optimal scheme is selected using a genetic algorithm, taking into account the scope of the fault impact, the degree of impact on key business nodes, and the complexity of the topology.

[0023] In the network topology optimization process of a cloud service provider, the dependency attribute data of five candidate topology schemes were first obtained. Scheme A has an inter-component coupling degree of 0.7, a data dependency degree of 0.6, and an interface dependency degree of 0.5, resulting in a dependency strength attribute value of 0.62. The dependency strength of other schemes was calculated using a similar method, constructing a 5x5 judgment matrix. ij =D i / D j A ij D represents the element in the i-th row and j-th column of the matrix. i D represents the dependency strength of the i-th solution. jThis represents the dependency strength of the j-th option. This formula is used to construct a 5x5 judgment matrix, establishing relative importance relationships by comparing the dependency strengths of each option pairwise. This vector, based on the dependency strength ratio of the judgment matrix, reflects the relative importance of each option in fault propagation, i.e., the scope of fault impact. The value of the eigenvector represents the weight of each option's impact on the overall system fault; the larger the value, the wider the impact. This result is obtained through the Analytic Hierarchy Processing (AHP), ensuring the scientific and consistent weight allocation. Three key business nodes were obtained from the business importance database: the order processing node had an importance of 0.9, the user authentication node 0.8, and the payment gateway node 0.7. Analysis of the fault propagation path revealed that option A affects the order processing node by 60%, and option B affects the user authentication node by 40%. Since the fault impact scope of option A (30%) exceeds the preset threshold, and the 60% impact on key business nodes is higher than the 50% threshold, it is judged as a high-risk option and eliminated. The remaining four options form the preferred topology set. A genetic algorithm was used to optimize the four schemes using multiple objectives. The population size was set to 100, the number of iterations was 500, the crossover probability was 0.8, and the mutation probability was 0.1. Scheme D, with the smallest fault impact range (0.15), the lowest impact on critical business nodes (20%), and a moderate topology complexity (0.6), was selected as the optimal scheme. The collaboration patterns were analyzed through the interaction patterns between components to identify interfaces that caused chain reactions and business nodes with zero tolerance for faults. High-risk topology schemes were automatically identified through risk assessment. Interaction log data of each component in the system was obtained, and the call frequency and response time series between components were analyzed. A directed graph structure was constructed based on the call relationships within the time window, and the in-degree and out-degree values ​​of each node were calculated. The PageRank algorithm was used to identify core component nodes, determine strong and weak dependencies between components, and obtain a component dependency graph containing weight information. For each edge in the component dependency graph, the probability of fault propagation is analyzed. If the number of downstream components of a certain interface exceeds a preset threshold, it is marked as a high-risk propagation node. A depth-first search algorithm is used to trace the fault propagation path, recording the component sequence and cumulative impact weight on each path. The existence of circular dependencies in the propagation path is determined, resulting in a fault propagation link graph. Based on the critical paths in the fault propagation link graph, business nodes with zero tolerance for faults are identified. Monte Carlo simulation is used to evaluate the business impact range under different fault scenarios, calculating the fault impact score for each business node. The impact score equals the fault probability multiplied by the business loss weight, resulting in a business node risk score matrix. Using the business node risk score matrix, a set of high-risk nodes is selected. For each high-risk node, its topology characteristics are analyzed. If a single point of failure risk or cascading failure risk exists, the topology is automatically marked as a high-risk solution. Specifically, the interaction log data between components contains rich call relationship information.

[0024] Specifically, when a user initiates an order request, the front-end service calls the order service, which in turn calls the inventory service, payment service, and user service. These call relationships are recorded in the interaction log. By parsing this log data, the frequency of each component's call can be extracted. For example, the order service might be called 1000 times per minute, while the inventory service is called 1500 times per minute. This difference in frequency reflects the importance of different components in the system. Constructing a directed graph structure can visually represent the dependencies between components.

[0025] In one possible implementation, each component is represented as a node in the graph, and the call relationships between components are represented as directed edges. The weight of each edge can be set as the call frequency or average response time. By calculating the in-degree and out-degree of nodes, the core components of the system can be identified. Components with high in-degrees indicate that they are depended on by multiple other components; for example, user authentication services typically have high in-degree values. Components with high out-degrees indicate that they depend on multiple downstream services; for example, aggregation services typically need to call multiple data sources. The PageRank algorithm plays a crucial role in identifying core components. This algorithm iteratively calculates the importance score of each node, considering not only the number of direct connections but also the importance of the connected nodes.

[0026] For example, even if a caching service has few direct connections, its PageRank score can still be high if it is depended on by multiple core business services, thus identifying it as a critical component. Analyzing the probability of fault propagation requires considering multiple dimensions.

[0027] It's important to note that when the number of downstream components for a given interface exceeds a threshold, a failure at that interface will cause the impact to spread in a fan-shaped pattern. For example, if the database connection pool service encounters a problem, all services that depend on the database will be affected. Tracing the failure propagation path using a depth-first search algorithm can uncover hidden cascading risks. In e-commerce systems, a payment service failure can lead to order service anomalies, which in turn affect logistics and after-sales services, forming a complete failure propagation chain. Circular dependencies are a crucial issue that requires special attention in system design.

[0028] In one embodiment, service A calls service B, service B calls service C, and service C calls service A in turn. This circular structure can lead to a cascading effect that is difficult to interrupt in the event of a failure. Recording access paths during depth-first traversal can effectively detect such circular dependencies. Monte Carlo simulation methods have unique advantages in evaluating failure scenarios. This method randomly simulates a large number of possible failure scenarios and statistically analyzes the business impact of each scenario.

[0029] For example, simulating 10,000 random failure scenarios, 30% of which result in order service unavailability for more than 5 minutes, and 20% of which affect payment functionality, allows for a quantitative assessment of the business impact of different failures. The construction of a risk scoring matrix requires a comprehensive consideration of both failure probability and business loss.

[0030] Preferably, for components on the core payment chain, even if the probability of failure is low, their risk scores will still be marked as high-risk due to the extremely high weight of business loss. In contrast, some auxiliary functions, such as recommendation services, although frequently invoked, have a lower weight of business loss and therefore relatively lower risk scores. Analysis of topology characteristics can reveal system vulnerabilities. Single point of failure risks typically occur at critical nodes without redundancy, such as authentication services with only one instance. Cascading failure risks are more common in highly coupled systems, where the failure of one component can trigger a domino effect of continuous service degradation. By automatically identifying these high-risk topology schemes, targeted improvement suggestions can be generated, such as adding service instances, introducing circuit breakers, or optimizing service dependencies, thereby improving the overall system stability and reliability.

[0031] Step S104: Construct a business dependency graph, which reflects the dependency relationships between various business components in the system. Identify key dependency paths from the business dependency graph, calculate the component dependency strength of each scheme in the preferred topology scheme set, and construct a business recovery time model.

[0032] A business dependency graph is constructed using the Neo4j graph database. Nodes represent business components, and edges represent dependencies between components. The weight W of an edge is calculated by combining the call frequency F, resource sharing degree S, and concurrent access intensity C, using the formula W = 0.4F. norm +0.3S +0.3C norm , where F norm and C norm The F and C values ​​are normalized, ranging from 0 to 1. The critical path method is used to identify critical dependency paths from the business dependency graph. The total float time for each path is calculated, and the top 5 paths with the smallest total float times are selected as critical dependency paths. These paths are then used in S106 to identify fault propagation paths to avoid potential islanding. For each scheme in the preferred topology set, the component dependency strength, D, is calculated. ij =Σ(k=1tom)w k ×sim k(i，j) D ij w represents the overall dependency strength between component i and component j. k Let Σw be the weight coefficient of the k-th dimension. k =1, sim k(i，j)Let m represent the similarity / dependency between components i and j in the k-th dimension, and m be the total number of dimensions used to evaluate the total number of dimensions. Dimension k includes features such as functional dependency, data interaction, and API call frequency. f il f is the value of the l-th feature of component i in dimension k. jl Let be the value of the l-th feature of component j in dimension k, and n be the number of features in each dimension. Based on component dependency strength and historical business recovery time data, the established business recovery time model can... T r T represents the total business recovery time. base Basic system startup time, P i T represents the probability of component i failing. i D is the individual recovery time for component i. i Let D be the set of dependent components of component i. ij R represents the dependency strength of component i on component j. j Let be the failure propagation coefficient of component j, and N represent the total number of components, services, or modules in the system. The output of the business recovery time model is the business recovery time. The parameters of this analytical model are fitted using an RBF kernel SVR model, optimized through grid search, and the optimal parameter configuration is determined using 5-fold cross-validation. The input of the SVR model is historical business recovery time data and component dependency characteristics, and the output is the predicted T. r .

[0033] Neo4j graph database has a natural advantage in building business dependencies, as its node and edge data structure can intuitively express complex business topologies.

[0034] In one possible implementation, each business component is abstracted as a node in a graph, storing attribute information such as the component's name, type, and resource consumption. Edges represent the call relationships or data dependencies between components, distinguishing between the initiating caller and the callee through directionality. The weight calculation of edges comprehensively considers multiple dimensions: call frequency F reflects the density of interactions between components; an interface called 10,000 times per hour is obviously more critical than an interface called 100 times per day; resource sharing degree S measures the competition of components for shared resource pools, such as database connection pools and cache space; and concurrency access intensity C reflects the pressure transmission capability under high-concurrency scenarios.

[0035] It's important to note that the critical path method plays a crucial role in identifying system bottlenecks. Total float represents the maximum time that activities on a given path can be delayed without affecting the overall completion time. A smaller total float indicates less time flexibility for that path, meaning any delay will directly impact overall performance. In e-commerce systems, the path from order placement to payment completion typically has the smallest total float because users have extremely low tolerance for waiting during this process. By selecting the top N critical paths, we can focus on the dependency chains that have the greatest impact on the business, avoiding excessive resource allocation to secondary paths.

[0036] Specifically, the calculation of component dependency strength employs a multi-dimensional comprehensive evaluation method. Different dimensions reflect different aspects of the dependency relationship. For example, functional dependency measures the degree of coupling between components in business logic, data dependency focuses on the amount of data exchange between components, and temporal dependency considers the time constraints of component calls. Similarity calculation uses the cosine similarity formula, which can effectively handle similarity measurements in high-dimensional feature spaces.

[0037] For example, order services and inventory services are highly similar in terms of functionality because they both involve processing the quantity of goods; however, in terms of data, order services need to frequently query inventory data, forming a strong data dependency.

[0038] In one embodiment, the business recovery time model is constructed by fully considering the cascading effects of failures. Basic system startup time includes necessary steps such as operating system startup and basic service loading, which are unavoidable fixed overheads in any recovery process. Component failure probabilities are derived from historical failure data statistics. Hardware failures typically have a low but stable probability, while software failure probabilities dynamically change with system load and runtime. Individual recovery time reflects the complexity of the component itself and the efficiency of the recovery mechanism. Database services, due to the need for data consistency checks and transaction recovery, typically have much longer recovery times than stateless application services.

[0039] Understandably, the fault impact propagation coefficient R j The impact of upstream component failures on downstream components was quantified. When the caching service fails, although the application layer can directly access the database, performance will significantly degrade. j A value of 0.7 might indicate that 70% of the performance loss propagates downstream. Dependency component set D iIdentifying dependencies requires a depth-first search, considering not only direct dependencies but also indirect ones. The RBF kernel function excels at handling nonlinear relationships, making it particularly suitable for predicting complex interactive effects, such as business recovery time. Grid search systematically traverses the parameter space, including the bandwidth and regularization coefficients of the RBF kernel, to find the optimal parameter combination. Five-fold cross-validation divides the dataset into five parts, using four parts for training and one part for validation in rotation, making full use of limited historical data while avoiding overfitting.

[0040] Step S105: Obtain the optimal topology scheme that balances recovery time and dependency strength constraints by using the business recovery time output by the business recovery time model; in the solution process, load balancing capability and fault tolerance backup capability are introduced as constraints.

[0041] Based on the business recovery time output by the business recovery time model, a multi-objective optimization problem is constructed. The objective functions include minimizing the business recovery time f1(x) and minimizing the component dependency strength f2(x), where x represents the topology scheme and f2(x) is the dependency strength D of all components. ij The average value of D ij This represents the dependence strength of component i on j. Load balancing capability and fault tolerance / redundancy are introduced as constraints. The load balancing threshold is set to a load difference between nodes not exceeding 20%, and the fault tolerance / redundancy requirement is that critical components must have at least one backup node. These constraints are transformed into inequality constraints and added to the optimization problem. The NSGA-II algorithm is used to solve the multi-objective optimization problem. Pareto optimality analysis is performed on the obtained non-dominated solution set, including non-dominated sorting and congestion calculation. The TOPSIS method is used to comprehensively evaluate the performance of each objective and select the optimal topology scheme.

[0042] In optimizing a transaction system, a multi-objective optimization problem was constructed. The objective function f1(x) minimizes the business recovery time, and f2(x) minimizes the component dependency strength. The initial mapping model shows that when the dependency strength is 0.8, the recovery time is 15 minutes. The load balancing threshold is set to a load difference between nodes not exceeding 20%, and the fault tolerance redundancy is set to at least one backup node for critical components. A genetic algorithm is used to solve the problem, with a population size of 100, a crossover rate of 0.8, a mutation rate of 0.1, and 1000 iterations. The fitness function is 1 / (0.6f1+0.4f2), and the weights are determined according to the importance of the business. After non-dominated sorting and congestion calculation, 20 Pareto optimal solutions are obtained. The TOPSIS method is used for evaluation, and the optimal solution is selected: recovery time 12 minutes, dependency strength 0.7, load difference 18%, and critical components all have one backup node. A sensitivity analysis was conducted on this solution. The load balancing threshold was adjusted to 25%, and the number of fault-tolerant backups was increased to 2. The results showed that the recovery time changed by no more than 5%, and the dependency intensity changed by no more than 3%, proving that the solution has good robustness.

[0043] Step S106: Based on the optimal topology scheme and combined with the fault propagation path of key services, assess the impact of the fault on the quality of service, including quantifying the scope of fault propagation, analyzing fault repair time and service recovery time, and determining the specific degree of impact of the fault on service quality through simulation and historical data analysis.

[0044] Based on the connectivity of nodes in the topology and the distribution characteristics of critical nodes, the service continuity parameters and service quality benchmark values ​​of each node on the fault path are obtained. The number of affected services within the propagation range is calculated by comparing the service quality difference before and after the fault. A risk score for each service is obtained by multiplying its priority weight by the fault probability. If the risk score exceeds a preset threshold, the service is identified as high-risk and its position in the topology is marked. Using the intersection of the marked high-risk service location information and the fault path, a set of critical paths for fault propagation is determined. By analyzing the fault frequency and average repair time in the historical records of each node on the critical path, the fault propagation speed and cumulative impact duration for each path are calculated. The expected time series of the fault reaching each service node is obtained based on the ratio of path length to propagation speed. By comparing the expected time series with the recovery time of each service node, it is determined which nodes have recovery times shorter than the fault arrival time during the fault propagation process. Fault isolation boundaries are constructed based on the positions of these nodes in the topology. Simulation data is used to verify the effectiveness of the isolation boundaries, obtaining the probability value that the fault is confined to a specific propagation range. Based on the probability value and the degree of service quality degradation of each service within the propagation range, the impact index of the failure on the overall service is calculated. A comprehensive impact score for the failure event is obtained by weighted summation of repair time, recovery time, and impact index. If the score is lower than the minimum requirement for business continuity, the failure scenario is judged as an unacceptable risk, and the corresponding risk level is output. Optionally, business continuity parameters include redundancy R, the ratio of backup paths to primary paths, service quality baseline Q0, and bandwidth and latency indicators under normal conditions. Risk score S b =w p ×P f ×(Q0-ΔQ), where w p The priority weights are: high 1.0, medium 0.7, and low 0.4, P. f Let ΔQ be the historical failure probability of the node, and ΔQ be the difference in service quality before and after the failure. The path length L is measured in hops, the propagation speed V = 1 / (f × MTTR), f is the historical failure frequency, MTTR is the mean time to repair, and the expected time series T is... i =∑(L j / V j ), j from source to i. Overall impact score C = α × MTTR + β × RTO + γ × I q Where α=0.4, β=0.3, γ=0.3, RTO is the recovery time target, I q This is an indicator of the degree of impact. If C > 8, it is considered an unacceptable risk.

[0045] For example, in fault propagation analysis, the connection relationships between nodes in the topology directly determine the propagation path and scope of the fault.

[0046] In a typical platform architecture, a front-end load balancer connects to multiple application servers, which in turn connect to a database cluster and a caching service. When the load balancer fails, its propagation path spreads downstream along these connections. By analyzing the business continuity parameters of each node, such as node redundancy, backup mechanisms, and failover time, the impact of the failure on service quality can be quantitatively assessed.

[0047] In one possible implementation, service quality benchmarks typically include metrics such as response time, throughput, and availability. When the response time of the order processing service increases from the normal 200 milliseconds to 2000 milliseconds, this tenfold performance drop directly impacts user experience. By calculating the service quality difference before and after the failure, and combining this with the number of affected services, the accurate scope of the impact can be determined. The setting of service priority weights takes into account the importance of different services; payment services typically have a higher weight than browsing services because the former is directly related to transaction completion and revenue.

[0048] It should be noted that the critical path for fault propagation is often not a single path, but rather a set of paths. By analyzing historical fault data, it can be found that the frequency of faults on certain paths is significantly higher than that on other paths.

[0049] For example, the average recovery time for a failure caused by database connection pool exhaustion is 15 minutes, while the average recovery time for a failure caused by network outage is 30 minutes. The calculation of failure propagation speed takes into account the characteristics of different types of failures. Software failures typically propagate faster than hardware failures because software failures can affect all dependent services within milliseconds.

[0050] Specifically, constructing fault isolation boundaries requires identifying nodes with self-healing capabilities. When the recovery time of a caching service is 5 minutes, and it takes 10 minutes for a fault to propagate from its source to that node, this caching service becomes a natural isolation point. The simulation data verification process tests the effectiveness of the isolation boundary under various conditions by simulating different fault scenarios. This method can assess the effectiveness of protective measures before actual faults occur, significantly improving system reliability.

[0051] In one embodiment, the comprehensive impact score calculation fully considers time factors and business loss. A failure affecting 1000 users for 30 minutes and a failure affecting 10000 users for 5 minutes, while having similar total user time, will receive a higher comprehensive score due to its wider impact. By setting a minimum threshold for business continuity requirements, unacceptable risk scenarios can be quickly identified, providing a clear direction for subsequent improvement measures. This quantitative assessment method transforms fault management from reactive response to proactive prevention, significantly improving the overall stability and reliability of the service.

[0052] Step S107: Based on the specific degree of impact, formulate and implement mitigation measures for critical fault propagation paths, including adjusting interfaces and data flows between components with high dependency intensity, optimizing load balancing strategies, and improving fault detection and recovery mechanisms.

[0053] Based on the quantitative score of the impact, the dependency strength coefficient of each node on the propagation path is calculated by multiplying the impact by the dependency coefficient. Key connection points in the component relationship graph whose dependency strength exceeds a preset threshold are identified. The ratio of data flow throughput to response time (R=T / S), where T is the response time and S is the throughput, is calculated to determine the priority sequence for interface adjustments. If the response time latency of an interface exceeds the business tolerance limit, the interface is identified as a bottleneck interface based on the priority sequence, and its location information in the propagation path is recorded. Using the location information and data flow characteristic parameters of the bottleneck interface, including traffic peaks and data types, extracted from monitoring logs, an interface transformation plan is constructed. By adding message queue middleware and asynchronous processing mechanisms to the original interface, a new data flow forwarding rule set is obtained. Based on the routing conditions and resource allocation weights in the rule set, the request distribution ratio of each component under different load conditions is obtained. By combining request distribution ratios with real-time load monitoring data, and calculating the matching degree from service nodes (match degree = sum of absolute values ​​of (actual ratio - expected ratio) divided by the number of nodes), dynamic adjustment parameters for load balancing are determined. Based on these parameters, the weight values ​​and health check intervals of each service node are updated. A weighted round-robin mechanism is used to distribute requests among healthy nodes, resulting in an optimized load distribution and resource utilization for each node. Based on resource utilization and fault detection thresholds (thresholds are utilization exceeding 80% or response timeout exceeding 200 milliseconds), the frequency and timeout parameters for heartbeat detection are configured. By comparing the detection results with a pre-defined fault characteristic database, the health status and fault type of the nodes are determined. If a fault is detected, a recovery mechanism is triggered, employing primary / backup switching or service degradation to ensure business continuity and obtain service availability metrics after fault handling.

[0054] When the order service heavily relies on the inventory service, any anomaly in the inventory service will immediately impact the order processing flow. Dependency strength is typically quantified by dimensions such as call frequency, data transfer volume, and business relevance. An interface called 1000 times per second has a significantly higher dependency strength than an interface called 10 times per minute. By analyzing these highly dependent connection points, vulnerable points in the system can be accurately located.

[0055] In one possible implementation, the ratio of interface response time to data throughput reflects the interface's processing efficiency. When the payment interface's response time increases from 100 milliseconds to 500 milliseconds, while the throughput remains unchanged, a significant increase in this ratio indicates that the interface has become a performance bottleneck. Setting a business tolerance limit needs to consider user experience and business characteristics. In e-commerce scenarios, users' tolerance for waiting in the payment process is typically no more than 3 seconds; exceeding this limit will cause a large number of users to abandon their transactions.

[0056] It's worth noting that the introduction of message queue middleware can effectively decouple strong dependencies between components. By transforming synchronous calls into asynchronous message passing, services that were originally tightly coupled can run and scale independently.

[0057] For example, after an order is created, the inventory deduction interface is no longer called directly. Instead, a message is sent to a queue, and the inventory service consumes the message according to its processing capacity. This change ensures that even if the inventory service is temporarily unavailable, the order service can still receive requests normally, significantly improving the system's fault tolerance.

[0058] Specifically, implementing a weighted round-robin mechanism requires real-time monitoring of the performance metrics of each node. When a node's CPU utilization reaches 80%, its weight value is reduced accordingly, causing new requests to be distributed more frequently to nodes with lighter loads. The health check interval needs to be set to strike a balance between detection accuracy and system overhead. Too frequent checks increase network load, while excessively long intervals may delay fault detection. An adaptive adjustment approach is typically used, increasing the check interval when the system is stable and shortening it when anomalies are detected.

[0059] In one embodiment, the fault signature database contains pattern information for various common faults. Network timeout faults manifest as multiple consecutive unresponsive requests, while service overload faults manifest as gradually increasing response times and rising error rates. Pattern matching enables rapid identification of fault types, thereby triggering corresponding recovery mechanisms. Master-slave failover is suitable for scenarios where nodes completely fail, with the standby node immediately taking over the service upon detecting a master node failure. Service degradation is suitable for situations with excessive system pressure, ensuring the normal operation of core services by disabling some non-core functions.

[0060] Understandably, this multi-layered fault handling mechanism forms a complete protection system. From decoupling and modification at the interface level to dynamic adjustment at the load level, and rapid recovery after a fault occurs, each step ensures system stability. Continuous monitoring and feedback of service availability metrics enable the entire mechanism to be continuously optimized and improved, ultimately achieving the goal of business continuity.

[0061] The above description is only a preferred embodiment of the present invention. It should be noted that those skilled in the art can make several improvements and additions without departing from the principle of the present invention, and these improvements and additions should also be considered within the scope of protection of the present invention.

Claims

1. An anomaly analysis method that combines failure propagation paths with topology, characterized by, The method comprises the following steps: Obtain business importance and recovery time target, calculate overall recovery time estimate under different topology structures; Compare the overall recovery time estimate with a preset threshold value, and screen out a candidate topology scheme set; Analyze the fault propagation path and evaluate the impact of the fault on the service quality recovery time for the candidate topology scheme set; Determine the fault influence range combined with the dependency strength attribute, and eliminate high-risk schemes to form an optimal topology scheme set; Construct a business dependency relationship graph, identify a key dependency path, and calculate component dependency strength to establish a business recovery time model; Output the recovery time through the business recovery time model, and determine the optimal topology scheme combined with the constraint condition; Evaluate the impact of the fault on the business service quality according to the optimal topology scheme, and develop mitigation measures for the key fault propagation path.

2. The method of claim 1, wherein the method further comprises: The method for obtaining business importance and recovery time target, and calculating overall recovery time estimate under different topology structures comprises the following steps: Determine the weight of each time item according to the business importance, and calculate the overall recovery time estimate by using the weighted summation method; If the overall recovery time estimate is less than the preset recovery time threshold value, the corresponding topology structure is added to the candidate topology scheme set; Preliminarily screen each scheme in the candidate topology scheme set combined with network bandwidth and delay factors; Simulate the screened scheme set, obtain the actual recovery time probability distribution through multiple simulations, and verify whether the expected target is met.

3. The method of claim 1, wherein the method further comprises: identifying a fault propagation path of the abnormal event; and identifying a topology of the abnormal event. The method for analyzing the fault propagation path and evaluating the impact of the fault on the service quality recovery time for the candidate topology scheme set comprises the following steps: Identify key nodes and key links by using a node importance calculation method; Obtain the failure rate of the key nodes and the failure rate parameters of the key links, construct a fault tree model, and generate multiple fault scenarios; For each fault scenario, simulate the fault propagation process by using a search algorithm, and record the propagation path and influence range; According to the propagation delay and the information of the affected nodes, estimate the service quality recovery time, calculate the average recovery time and distribution characteristics, and evaluate the impact degree of the fault on the service quality recovery time.

4. The method of claim 1, wherein the method further comprises: identifying a fault propagation path of the fault; and identifying a topology of the fault propagation path. The method for combining the dependency strength attribute to determine the fault influence range, and eliminating high-risk schemes to form an optimal topology scheme set comprises the following steps: Obtain the coupling degree and dependency degree data between components, and calculate the dependency strength attribute value by using the weighted summation method; Construct a judgment matrix according to the dependency strength attribute value, and calculate the fault influence range; Obtain a list of key business nodes from a preset database, and judge whether the fault propagation path involves the key business nodes; If the fault influence range exceeds a preset threshold value or involves the key business nodes, the scheme is determined as a high-risk scheme and is eliminated.

5. The method of claim 1, wherein the method further comprises: identifying a fault propagation path of the abnormal event; and identifying a topology of the abnormal event. The method for constructing a business dependency relationship graph, identifying a key dependency path, and calculating component dependency strength comprises the following steps: Construct the business dependency relationship graph through a graph database, wherein a node represents a business component, and an edge represents a dependency relationship; Calculate the weight of the edge according to the calling frequency and the resource sharing degree, and identify the key dependency path by using the critical path method; For each scheme in the set of optimal topology schemes, the component dependency strength is calculated, the comprehensive dependency relationship is evaluated based on multi-dimensional similarity, and the service recovery time model is established.

6. The method of claim 1, wherein the method further comprises: identifying a fault propagation path of the fault; and identifying a topology of the fault propagation path. The recovery time is output through the service recovery time model, and the optimal topology scheme is determined in combination with the constraint condition, including: According to the output result of the service recovery time model, a multi-objective optimization problem is constructed, and the objectives include minimizing the recovery time and the component dependency strength; The load balancing capability and fault tolerance backup capability are introduced as constraint conditions, and the load difference threshold and backup redundancy requirement are set; An optimization algorithm is used to solve the multi-objective optimization problem, and the quality of the candidate solution is evaluated through the fitness function; The non-dominated solution set is comprehensively analyzed, and the optimal topology scheme is selected.

7. The method of claim 1, wherein the method further comprises: identifying a fault propagation path of the fault; and identifying a topology of the fault propagation path. The influence of the fault on the service quality of the business is evaluated according to the optimal topology scheme, including: According to the connection relationship of the nodes in the topology structure, the service continuity parameters of the nodes on the fault path are obtained; By comparing the service quality difference before and after the fault, the number of affected services is calculated; According to the business priority and the fault probability, the risk score is calculated, and if the risk score exceeds the preset threshold, the high-risk business position is marked; Through the intersection of the high-risk business position and the fault path, the key path set is determined, the fault propagation speed and the influence time are analyzed, and the fault isolation boundary is constructed.

8. The method of claim 1, wherein the method further comprises: identifying a fault propagation path of the fault; and identifying a topology of the fault propagation path. The mitigation measures for the key fault propagation path are formulated, including: According to the influence degree score and the dependency strength attribute value, the key connection point is identified; Through the ratio of data flow throughput and response time, the interface adjustment priority is determined; For the bottleneck interface, a reconstruction scheme is constructed, the middleware and asynchronous processing mechanism are increased, and the data flow forwarding rule is updated; According to the load monitoring data, the load balancing parameters are adjusted, the fault detection frequency and the recovery mechanism are configured, and the service continuity is guaranteed through master-slave switching or service degradation.

9. The method of claim 1, wherein the method further comprises: identifying a fault propagation path of the fault; and identifying a topology of the fault propagation path. The mitigation measures for the key fault propagation path include adjusting the interface and data flow between components with high dependency strength, optimizing the load balancing strategy, and improving the fault detection and recovery mechanism.

Images