A method and system for end-to-end stress testing and monitoring

Abstract

This invention discloses a full-link stress testing monitoring method and system, belonging to the field of software testing technology. Based on the priority of event nodes, different data acquisition methods are used to obtain stress testing data when anomalies occur. An anomaly detection model is constructed using the stress testing data when anomalies occur. The priority of the links is calculated, and when an event node anomalies, different acquisition methods are matched to obtain link stress testing data for links of different priorities. When a high-priority event node on a high-priority link anomalies, environmental information surrounding the high-priority event node is simultaneously collected. The characteristics of the collected link stress testing data and surrounding environmental information are compared with the anomaly characteristics of the anomaly detection model to obtain link anomaly information. This method can provide timely feedback on the status of high-priority nodes; and when anomalies occur, the abnormal state of the node itself and the state of the surrounding environment are fully collected, providing sufficient auxiliary information for problem localization.

Description

Technical Field

[0001] This invention relates to the field of software testing technology, and specifically to a full-link stress testing and monitoring method and system. Background Technology

[0002] As the internet industry continues to develop, system architectures are becoming increasingly complex, and business scenarios are becoming more diversified, leading to higher and higher requirements for performance testing. Traditional load testing methods can no longer meet the needs of business and technology development. Compared to traditional load testing methods, end-to-end load testing has become the best practice in the new stage of performance testing.

[0003] End-to-end load testing is a technical engineering project that simulates massive user requests and data based on near-production business scenarios and system environments. It tests and verifies the entire business chain under various scenarios, continuously identifies and optimizes bottlenecks, and ensures system stability. Currently, scenario-based end-to-end testing involves too many business nodes and node characteristics. Existing end-to-end load testing technologies often lack differentiation in the priority of different link nodes and metrics during the load testing process. Data collection mostly relies on conventional timed acquisition mechanisms, resulting in untimely acquisition of anomaly information and insufficient data collection when anomalies occur. This makes it difficult to quickly locate abnormal nodes within massive load testing data. Summary of the Invention

[0004] This invention addresses the shortcomings of existing technologies by providing a full-link stress testing and monitoring method and system. This method enables timely feedback on the status of high-priority nodes; when an anomaly occurs, the abnormal status of the node itself and the status of the surrounding environment are fully collected, providing sufficient auxiliary information for problem localization; during anomaly detection, anomaly detection models are established or model feature analysis is in place, and the feature data is relatively targeted.

[0005] To achieve the above objectives, the technical solution adopted by this invention is as follows: a full-link stress testing and monitoring method, which acquires stress test data when an anomaly occurs using different data acquisition methods based on the priority of event nodes; constructs an anomaly detection model by analyzing the stress test data when an anomaly occurs; calculates the priority of the links, and when an event node anomalies, matches different acquisition methods to acquire link stress test data for links with different priorities; when a high-priority event node of a high-priority link anomalies, simultaneously acquires environmental information surrounding the high-priority event node; compares the features of the acquired link stress test data and surrounding environmental information with the anomaly features of the anomaly detection model to obtain link anomaly information.

[0006] Furthermore, the priority of the event node is calculated as follows:

[0007] ;

[0008] This represents the weighted score of the i-th event node. These are the weighting coefficients of the influence factors. This is the weighting coefficient for the in-degree of the event node. The weight coefficient for the out-degree of the event node. , The value ranges from 0.6 to 0.7. The value ranges from 0.2 to 0.3. Let be the in-degree of the i-th event node. Let M be the out-degree of the i-th event node, and M be the number of influencing factors;

[0009] When the calculation result When the i-th event node has a higher priority, then the i-th event node has a higher priority.

[0010] When the calculation result When the i-th event node has a lower priority, then the i-th event node has a lower priority.

[0011] The threshold is set in advance.

[0012] Furthermore, high-priority event nodes use a push method to collect load test data when an anomaly occurs; low-priority event nodes use a periodic polling method to collect load test data when an anomaly occurs.

[0013] Furthermore, the anomaly detection model matrix for any event node is constructed as follows:

[0014] ;

[0015] in Indicates the i-th ’ The j-th exception type ’ Data information of each feature factor, N is the number of exception types. K is the number of characteristic factors;

[0016] The data information for each feature factor includes the feature factor name, the threshold of the feature factor, and the similarity comparison function of the feature factor.

[0017] Furthermore, the link priority is calculated as follows:

[0018] ;

[0019] in, This indicates the percentage change in the difference between the node values ​​and the mean in the link. This represents the average weight of all nodes in the link. Represents a node The weighted score, , Indicates the number of event nodes contained in the link;

[0020] When the calculation result When this happens, the link has a higher priority;

[0021] When the calculation result When this happens, the link priority is low;

[0022] The threshold is set in advance.

[0023] Furthermore, the surrounding environment information includes server information, business system information, and database components at that point in time.

[0024] Furthermore, the stress test data and surrounding environmental information are filtered, structurally extracted, and clustered.

[0025] Furthermore, it also includes visualizing the results of anomaly analysis, displaying information such as node location, cause of node anomaly, impact, and estimated recovery time.

[0026] Furthermore, the collected end-to-end stress test data and surrounding environmental information are compared with the anomaly features of the anomaly detection model, and the similarity score simValue is calculated.

[0027] ;

[0028] Where eventFea(i) represents the features of the full-link stress test data and surrounding environment information of the i-th event node, and modelFea(i) represents the features of the i-th event node. ’ ) represents the i-th node in the anomaly detection model. ’ Characteristics of each anomaly type P represents the number of features for a certain anomaly type. The function func will return eventFea(i) and modelFea(i). ’ The similarity is calculated using the comparison function, where compare is the similarity comparison function for each feature factor in the anomaly detection model.

[0029] An anomaly is defined as the type of anomaly that scores the highest when it exceeds the similarity threshold.

[0030] This invention also provides a full-link stress testing and monitoring system, including a rule base management module for defining priority rules for event nodes and data collection rules;

[0031] The anomaly detection model management module performs feature analysis on the data information of event nodes and builds an anomaly detection model for each event node.

[0032] The load test data acquisition module is used to calculate the link priority according to the priority division rules of event nodes, and call different data acquisition methods to collect load test data for abnormal links of different priorities.

[0033] The load test data analysis module filters, extracts structured data, and performs clustering processing on the collected load test data.

[0034] The anomaly detection and comparison module takes the event node data from the stress test data analysis module and the anomaly features of the event nodes from the anomaly detection model management module as input for comparison and analysis to obtain link anomaly information;

[0035] The test results visualization module displays the results of the anomaly detection comparison module in a visual format.

[0036] Compared with existing technologies, the advantages of this invention are as follows: This invention distinguishes the priorities of link nodes during the stress testing process using existing end-to-end stress testing technologies, preventing high-priority nodes from failing to provide timely feedback when there are too many event nodes; and it establishes an anomaly detection model based on priority-based collection of stress measurement data when anomalies occur. Secondly, it establishes rules for dividing link priorities. When an anomaly occurs, it collects stress testing data on one hand, and on the other hand, it uses a "snapshot collection" method to fully collect information about the surrounding environment to provide sufficient auxiliary information for facilitating problem localization and avoiding inaccurate anomaly analysis results. Attached Figure Description

[0037] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this application. For those skilled in the art, other drawings can be obtained based on these drawings.

[0038] Figure 1 This is a flowchart of the end-to-end stress testing and monitoring method provided in Embodiment 1 of the present invention;

[0039] Figure 2 This is a flowchart of the surrounding environment information of a high-priority event node provided in Embodiment 1 of the present invention;

[0040] Figure 3 This is a structural diagram of the end-to-end stress testing and monitoring system provided in Embodiment 2 of the present invention. Detailed Implementation

[0041] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0042] Example 1

[0043] This application provides a method for end-to-end load testing and monitoring, such as... Figure 1 As shown, it includes the following steps:

[0044] S1. Based on the priority of event nodes, different data acquisition methods are used to obtain stress test data when an anomaly occurs.

[0045] First, a rule base is constructed, including rules for prioritizing event nodes and data collection rules. By analyzing the execution process of the business workflow, several event nodes are identified. Taking the issuance of an instruction as an example, the analysis shows that a certain link consists of six event nodes: writing, reviewing, issuing, signing, executing, and ending. The impact factors of events are scored using expert scoring and similar project experience methods to determine the weight of each impact factor and calculate the score for each event node. When the score exceeds a threshold, it is defined as a high-priority event node and treated as an important event node, with load testing data collected from the node via push notifications. When the score is less than the threshold, it is defined as a low-priority event node and treated as a general event node, with load testing data collected from the node via periodic polling.

[0046] After the rule base is set up, code tracking is performed based on the priority division results of the rule base.

[0047] Specifically, the method for calculating the score for each event node is as follows:

[0048] By analyzing the entire business scenario, event nodes within the business scenario are identified, and a directed graph of the chain is constructed. Let V represent the set of nodes, E represent the set of edges, and an edge exist between any two nodes. We also retrieve the in-degree and out-degree information of all event nodes. Let the set of all event nodes be denoted as . , This represents the i-th event node. , No. All influence factors of each node are , Let j be the influencing factor of the i-th event node. The weight score for each influencing factor is determined using expert scoring or similar empirical methods. The basic score of each event node is calculated. Then, combining the in-degree and out-degree of each node, the overall score of the node is calculated as follows:

[0049] ;

[0050] This represents the weighted score of the i-th event node. These are the weighting coefficients of the influence factors. This is the weighting coefficient for the in-degree of the event node. The weight coefficient for the out-degree of the event node. , The value ranges from 0.6 to 0.7. The value ranges from 0.2 to 0.3. Let be the in-degree of the i-th event node. Let M be the out-degree of the i-th event node, and M be the number of influencing factors;

[0051] When the calculation result When the i-th event node has a higher priority, then the i-th event node has a higher priority.

[0052] When the calculation result When the i-th event node has a lower priority, then the i-th event node has a lower priority.

[0053] Furthermore, The pre-set threshold can be assigned a value based on expert experience. The value range is 0.6-0.9, with 0.8 being preferred. The out-degree and in-degree of each event node are calculated by constructing a directed graph, and the event node influence factor is superimposed to achieve a quantitative assessment of event node priority.

[0054] High-priority event nodes collect load test data when an anomaly occurs via push notifications; low-priority event nodes collect load test data when an anomaly occurs via periodic polling.

[0055] S2. By analyzing the stress test data during abnormal situations, an anomaly detection model is constructed.

[0056] The specific construction method of the event node anomaly detection model is as follows: In practical implementation, the anomaly classification can be specifically divided according to the actual business characteristics. For example, it can be classified according to the components used by the system, such as operating system anomalies, database anomalies, web component anomalies, cache anomalies, load balancing anomalies, etc., or according to user operations, such as query anomalies, export anomalies, write anomalies, etc. Taking the query anomaly of a certain event node as an example, a series of features are set, such as web connection pool utilization exceeding 90% and an increase in the number of slow database queries. The anomaly detection model matrix for a certain event node is constructed as follows:

[0057] ;

[0058] in Indicates the i-th ’ The j-th exception type ’ Data information of each feature factor. N is the number of exception types. K represents the number of feature factors; the data information for each feature factor includes the feature factor name, the threshold of the feature factor, and the comparison function of the feature factor.

[0059] Understandably, analyzing stress test data during anomalies and constructing an anomaly detection model can employ algorithms such as random forests, expert analysis of the anomaly data, or a combination of both to obtain information such as threshold values ​​for feature factors in the anomaly detection model matrix. The stress test data during anomalies can be static or dynamically updated; preferably, dynamically updated anomaly test data is used to continuously train the model and improve its accuracy.

[0060] S3. Calculate the priority of the link. When an event node fails, use different acquisition methods to obtain link stress test data for links with different priorities.

[0061] The priority of a link is calculated based on the importance of the nodes that make up the link. The link priority is obtained by calculating the link's fluctuation ratio. The link priority calculation method is as follows:

[0062] ;

[0063] in, This represents the percentage change in the difference between the node values ​​and the mean value in the link. A higher percentage change indicates a higher priority for the node in the link, while a lower percentage change indicates a lower priority for the node in the link. This represents the average weight of all nodes in the link. , Indicates the number of event nodes contained in the link;

[0064] When the calculation result When the calculation result is [not specified], the link priority is higher; when the calculation result is [not specified], the link priority is higher When this happens, the link priority is low; The threshold is set in advance.

[0065] When a normal link event node malfunctions (i.e., a low-priority link), load test data is collected by periodic polling. When an important event node malfunctions (i.e., a high-priority link), proactive push is used to address the issue of the inability to provide real-time feedback on the test status of important business nodes.

[0066] S4. When a high-priority event node on a high-priority link experiences an anomaly, environmental information surrounding the high-priority event node is collected simultaneously.

[0067] Furthermore, when an event node on a high-priority link experiences an anomaly, in addition to reporting the node's own status information, an integrated data collection plugin will also perform a "snapshot collection" of the surrounding environment information at the time of the anomaly. The collected information includes server information at that time (CPU usage, memory usage, disk I / O, etc.), business system information (number of processes, number of database connections, number of cached connections, etc.), database components (number of connections, presence of deadlocks, slow queries, remaining disk space, etc.), and the status information of various related components. This solves the problem that testers cannot fully understand the status of the entire system when an anomaly occurs, providing more data support for testers and developers to locate and resolve problems.

[0068] See Figure 2 When critical event nodes experience anomalies, proactive data push and "snapshot collection" are both handled by the collection plugin integrated into the Web component. The specific implementation steps are as follows: When a high-priority critical event node experiences anomalies, the collection plugin first proactively calls the push interface of the load test data collection module to push the load test data of the abnormal event node; then, according to the configuration strategy defined by the rule base, the collection plugin collects status information such as CPU utilization, memory utilization, and network usage of other components such as the operating system, database, cache, and load balancer; finally, the collection plugin calls the push interface of the load test data collection module again to send the collected snapshot data.

[0069] The stress test data and the surrounding environmental information of high-priority event nodes are filtered, structured, and clustered to provide clear and standardized data for the anomaly detection model comparison module.

[0070] S5. Compare the features of the collected link stress test data and surrounding environment information with the abnormal features of the anomaly detection model to obtain link anomaly information.

[0071] The collected stress test data of link event nodes and the features of surrounding environmental information are compared and analyzed with the abnormal features of event nodes in the anomaly detection model. The features in the anomaly detection model include a similarity comparison function for feature factors. When the similarity between the data features of a node and a certain abnormal feature of a node in the model reaches a threshold, the anomaly of the event node is confirmed based on the information in the anomaly detection model, and information such as the scope of impact and recovery time caused by the anomaly is analyzed. When the similarity is below the threshold, it is used as input to the anomaly detection model management module, which expands and refines the anomaly detection model through machine learning and other methods.

[0072] The collected end-to-end stress test data and surrounding environmental information are compared with the abnormal features of the anomaly detection model, and the similarity simValue is calculated.

[0073] ;

[0074] Where eventFea(i) represents the features of the link stress test data and surrounding environment information of the i-th event node, and modelFea(i) represents the features of the link stress test data and surrounding environment information of the i-th event node. ’ ) represents the i-th node in the anomaly detection model. ’ Characteristics of each anomaly type P represents the number of features for a certain anomaly type. The function func will return eventFea(i) and modelFea(i). ’ The similarity is calculated using the `compare` function, which compares the similarity of each feature factor in the anomaly detection model. After comparing with all anomaly types of the event node, the anomaly type with the highest score is defined as such if it exceeds the similarity threshold. Otherwise, the features of the abnormal event node analyzed by the load testing data processing module are used as input to the anomaly detection model management module. Through machine learning methods, the model is continuously optimized and expanded.

[0075] It should be noted that the model can be optimized and expanded using existing algorithms to augment the data. The similarity comparison function can employ the Euclidean distance algorithm.

[0076] Understandable and The threshold is determined based on the complexity of the network topology and the applicable scenarios.

[0077] Preferably, the end-to-end stress testing and monitoring method provided in this embodiment of the invention may further include visualizing the results of anomaly analysis, and displaying information including node location, cause of node anomaly, impact, and estimated recovery time.

[0078] Example 2

[0079] This application provides a full-link stress testing and monitoring system, such as Figure 3 As shown, it includes:

[0080] The rule base management module is used to define priority rules for event nodes and data collection rules. The priority rules categorize event nodes into important and general event nodes based on factors such as the event's impact factor and the node's in-degree and out-degree. Different data collection rules are then matched to event nodes of different priorities, and different data collection methods are invoked through collection plugins.

[0081] The rule base includes priority rules for event nodes, priority rules for links, and environmental information surrounding high-priority event nodes.

[0082] Priority rules for event nodes. Specifically, through end-to-end business scenario analysis, event nodes in the business scenario are identified, and a directed graph of the chain is constructed. Let V represent the set of nodes, E represent the set of edges, and an edge exist between any two nodes. We also retrieve the in-degree and out-degree information of all event nodes. Let the set of all event nodes be denoted as . , This represents the i-th event node. , No. All influence factors of each node are , Let j be the influencing factor of the i-th event node. The weight score for each influencing factor is determined using expert scoring or similar empirical methods. The basic score of each event node is calculated. Then, combining the in-degree and out-degree of each node, the overall score of the node is calculated as follows:

[0083] ;

[0084] This represents the weighted score of the i-th event node. These are the weighting coefficients of the influence factors. This is the weighting coefficient for the in-degree of the event node. The weight coefficient for the out-degree of the event node. , The value ranges from 0.6 to 0.7. The value ranges from 0.2 to 0.3. Let be the in-degree of the i-th event node. Let M be the out-degree of the i-th event node, and M be the number of influencing factors;

[0085] When the calculation result When the i-th event node has a higher priority, then the i-th event node has a higher priority.

[0086] When the calculation result When the i-th event node has a lower priority, then the i-th event node has a lower priority.

[0087] Furthermore, The pre-set threshold can be assigned a value based on expert experience. The value range is 0.6-0.9, with 0.8 being preferred. The out-degree and in-degree of each event node are calculated by constructing a directed graph, and the event node influence factor is superimposed to achieve a quantitative assessment of event node priority.

[0088] High-priority event nodes collect load test data when an anomaly occurs via push notifications; low-priority event nodes collect load test data when an anomaly occurs via periodic polling.

[0089] Rules for link priority. Link priority is calculated based on the importance of the nodes that make up the link. Link priority is obtained by calculating the link's fluctuation ratio. The link priority calculation method is as follows:

[0090] ;

[0091] in, This represents the percentage change in the difference between the node values ​​and the mean value in the link. A higher percentage change indicates a higher priority for the node in the link, while a lower percentage change indicates a lower priority for the node in the link. This represents the average weight of all nodes in the link. Represents a node The weighted score, , Indicates the number of event nodes contained in the link;

[0092] When the calculation result When the calculation result is [not specified], the link priority is higher; when the calculation result is [not specified], the link priority is higher When this happens, the link priority is low; The threshold is set in advance.

[0093] When a normal link event node malfunctions (i.e., a low-priority link), load test data is collected by periodic polling. When an important event node malfunctions (i.e., a high-priority link), proactive push is used to address the issue of the inability to provide real-time feedback on the test status of important business nodes.

[0094] Regarding the environmental information surrounding high-priority event nodes, when an event node on a high-priority link experiences an anomaly, in addition to reporting the node's own status information, an integrated data collection plugin will also perform a "snapshot collection" of the surrounding environmental information at the time of the event. The collected information includes server information at that time (CPU usage, memory usage, disk I / O, etc.), business system information (number of processes, number of database connections, number of cached connections, etc.), database components (number of connections, presence of deadlocks, slow queries, remaining disk space, etc.), and the status information of various related components. This solves the problem that testers cannot fully understand the status of the entire system when an anomaly occurs, providing more data support for testers and developers to locate and resolve problems.

[0095] The anomaly detection model management module performs feature analysis on the data information of event nodes, constructs anomaly detection models for each event node, and provides support for the anomaly detection comparison module. The specific construction method of the event node anomaly detection model is as follows: In practical implementation, anomalies can be classified according to actual business characteristics. For example, anomalies of a certain event node can be classified according to the components used by the system, such as operating system anomalies, database anomalies, web component anomalies, cache anomalies, load balancing anomalies, etc., or according to user operations, such as query anomalies, export anomalies, write anomalies, etc. Taking the query anomaly of a certain event node as an example, a series of features are set, such as web connection pool utilization exceeding 90% and an increase in the number of slow database queries. The anomaly detection model matrix for a certain event node is constructed as follows:

[0096] ;

[0097] in Indicates the i-th ’ The j-th exception type ’ Data information of each feature factor. N is the number of exception types. K represents the number of feature factors; the data information for each feature factor includes the feature factor name, the threshold of the feature factor, and the comparison function of the feature factor.

[0098] Understandably, analyzing stress test data during anomalies and constructing an anomaly detection model can employ algorithms such as random forests, expert analysis of the anomaly data, or a combination of both to obtain information such as threshold values ​​for feature factors in the anomaly detection model matrix. The stress test data during anomalies can be static or dynamically updated; preferably, dynamically updated anomaly test data is used to continuously train the model and improve its accuracy.

[0099] The load testing data acquisition module calculates link priorities based on the priority rules of event nodes and uses different data acquisition methods to collect load testing data for links with different priorities. Specifically, when a link event node fails, i.e., a low-priority link, load testing data is collected through periodic polling; when an important event node fails, i.e., a high-priority link, proactive push is used to address the issue of real-time feedback on the test status of important business nodes. Furthermore, when a high-priority link event node fails, in addition to reporting the node's own status information, an integrated acquisition plugin will perform a "snapshot collection" of the surrounding environment information at the time of the node's occurrence. This collection includes server information (CPU usage, memory usage, disk I / O, etc.), business system information (number of processes, number of database connections, number of cached connections, etc.), database components (number of connections, presence of deadlocks, slow queries, remaining disk space, etc.), and the status information of various related components. This solves the problem that testers cannot fully understand the overall system status at the time of the failure, providing more data support for testers and developers to locate and resolve problems.

[0100] The load test data analysis module filters, extracts structured data, and performs clustering processing on the collected load test data.

[0101] The anomaly detection and comparison module takes the event node data from the stress test data analysis module and the anomaly features of the event nodes from the anomaly detection model management module as input for comparative analysis to obtain link anomaly information.

[0102] The collected stress test data of link event nodes and the features of surrounding environmental information are compared and analyzed with the abnormal features of event nodes in the anomaly detection model. The features in the anomaly detection model include a similarity comparison function for feature factors. When the similarity between the data features of a node and a certain abnormal feature of a node in the model reaches a threshold, the anomaly of the event node is confirmed based on the information in the anomaly detection model, and information such as the scope of impact and recovery time caused by the anomaly is analyzed. When the similarity is below the threshold, it is used as input to the anomaly detection model management module, which expands and refines the anomaly detection model through machine learning and other methods.

[0103] The collected end-to-end stress test data and surrounding environmental information are compared with the abnormal features of the anomaly detection model, and the similarity simValue is calculated.

[0104] ;

[0105] Where eventFea(i) represents the features of the full-link stress test data and surrounding environment information of the i-th event node, and modelFea(i) represents the features of the i-th event node. ’) represents the i-th node in the anomaly detection model. ’ Characteristics of each anomaly type P represents the number of features for a certain anomaly type. The func function will return eventFea(i) and modelFea(i). ’ The similarity is calculated using the `compare` function, which compares the similarity of each feature factor in the anomaly detection model. After comparing with all anomaly types of the event node, the anomaly type with the highest score is defined as such when it exceeds the similarity threshold. Otherwise, the features of the abnormal event node analyzed by the load testing data processing module are used as input to the anomaly detection model management module. Through machine learning methods, the model is continuously optimized and expanded.

[0106] It should be noted that the model can be optimized and expanded using existing algorithms to augment the data. The similarity comparison function can employ the Euclidean distance algorithm.

[0107] The test results visualization module visualizes the results of the anomaly detection comparison module, displaying information such as node location, cause of node anomaly, impact, and estimated recovery time. Thanks to the previously introduced mechanism for real-time push of important event nodes, it ensures that when anomalies occur at important business process event nodes, anomaly information can be obtained immediately, allowing for timely feedback and improving the time testers can spend locating problems.

[0108] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A full-link stress testing monitoring method, characterized in that, different data collection methods are used to obtain stress testing data when an abnormality occurs based on the priority of the event node; an abnormality detection model is constructed by analyzing the stress testing data when an abnormality occurs; the priority of the link is calculated, and different collection methods are matched for different priority links to obtain link stress testing data when an abnormality occurs in the event node; when an abnormality occurs in a high-priority event node of a high-priority link, the surrounding environment information of the high-priority event node is collected simultaneously; the collected link stress testing data and surrounding environment information are compared with the abnormal features of the abnormality detection model to obtain link abnormality information; the priority of the link is calculated as follows: ； wherein, represents the floating proportion of the difference of the node in the link and the mean value, represents the mean value of the weight of all nodes in the link, represents the weighted score of the node , , represents the number of event nodes contained in the link; When the result of the calculation is then the link priority is high; When the calculation result is low, the link priority is low; is a predetermined threshold value; the collected link stress testing data and surrounding environment information are compared with the abnormal features of the abnormality detection model to calculate the similarity simValue, ； wherein eventFea(i) is the feature of the link stress testing data and the surrounding environment information of the ith event node, modelFea(i ’ ) is the feature of the ith ’ abnormal type of the node in the abnormal detection model, P is the feature quantity of the ith ’ abnormal type, the func function returns the similarity of eventFea(i) and modelFea(i ’ ), and compare is the similarity comparison function of each feature factor in the abnormal detection model; when the highest scoring abnormal type exceeds the similarity threshold, it is defined as the abnormal type.

2. The full link pressure measurement monitoring method according to claim 1, characterized in that, The priority of the event node is calculated as follows: ； This represents the weighted score of the i-th event node. These are the weighting coefficients of the influence factors. This is the weighting coefficient for the in-degree of the event node. The weight coefficient for the out-degree of the event node. , The value ranges from 0.6 to 0.

7. The value ranges from 0.2 to 0.

3. Let be the in-degree of the i-th event node. Let M be the out-degree of the i-th event node, and M be the number of influencing factors. Let j be the influencing factor of the i-th event node. for The weighted score; When the result of the calculation is then the i-th event node has a high priority; When the result of the calculation is then the i-th event node has a low priority; is a predetermined threshold value. 3.The full link pressure measurement monitoring method according to claim 2, characterized in that, high-priority event nodes use a push method to collect stress testing data when an abnormality occurs; low-priority event nodes use a regular polling method to collect stress testing data when an abnormality occurs.

4. The full link pressure measurement monitoring method according to claim 1, characterized in that, The abnormality detection model matrix of any event node is constructed as follows: ； in Indicates the i-th ’ The j-th exception type ’ Data information of each feature factor, K is the number of characteristic factors; The data information of each feature factor includes the feature factor name, the threshold value of the feature factor, and the similarity comparison function of the feature factor.

5. The full link pressure measurement monitoring method according to claim 2, wherein, The surrounding environment information includes server information, business system information, and database components corresponding to the time point.

6. The full link pressure measurement monitoring method according to claim 1, wherein, The stress testing data and surrounding environment information are filtered, structured, and clustered.

7. The full link pressure measurement monitoring method according to claim 1, characterized in that, It also includes visualizing the results of abnormal situation analysis, including node location, reason for node abnormality, impact, and time for recovery.

8. A full-link stress testing monitoring system, the full-link stress testing monitoring system employing the full-link stress testing monitoring method of any one of claims 1-7, characterized in that, It includes: a rule base management module for formulating event node priority division rules and data collection rules; an abnormality detection model management module for feature analysis of event node data information and construction of an abnormality detection model for each event node; a stress testing data collection module for calculating link priority based on event node priority division rules and calling different data collection methods for stress testing data collection for different priority link abnormalities; a stress testing data analysis module for filtering, structured extraction, and clustering processing of collected stress testing data; an abnormality detection comparison module for comparing and analyzing the event node data of the stress testing data analysis module and the abnormal features of the event node of the abnormality detection model management module to obtain link abnormality information; a test result visualization display module for visualizing the results of the abnormality detection comparison module.

Images