A system for testing performance of application server middleware

Abstract

The application relates to the technical field of application server middleware performance test, and particularly discloses an application server middleware performance test system. The system comprises a business scene intelligent modeling module, an adaptive load generation and scheduling module, a full-stack performance data acquisition and correlation analysis module and a test environment automatic management module. Through the cooperative work of the above modules, the test script is automatically generated from natural language or logs, the load pressure is dynamically adjusted based on real-time feedback, the full-stack performance data correlation analysis and root cause positioning are realized, and the test environment is automatically constructed and elastically expanded, so that the efficiency, authenticity and automation level of the performance test are improved.

Description

Technical Field

[0001] This invention belongs to the field of computer software testing technology, and specifically relates to an application server middleware performance testing system. Background Technology

[0002] In the fields of software engineering and information technology, application server middleware serves as a core support platform for enterprise applications, and its performance directly impacts the stability, response speed, and user experience of upper-layer business processes. Performance testing is a crucial step in evaluating and ensuring the quality of middleware services, aiming to identify system problems, verify capacity planning, and optimize resource allocation by simulating real-world loads.

[0003] Performance testing of application server middleware aims to build a test environment that accurately simulates complex business scenarios and generates high concurrency pressure, and precisely collects and analyzes various performance indicators of the system under pressure. This typically involves multiple technical steps, such as generating test scripts, applying test loads, and collecting and monitoring performance data.

[0004] Performance testing solutions often face a series of integration issues. The writing and maintenance of test scripts heavily rely on testers' mastery of business logic and programming languages, leading to inefficient test case construction and difficulty in covering complex interaction scenarios. Traditional load generation tools struggle to dynamically adjust stress models to accurately reflect business peaks and troughs when simulating large-scale, high-concurrency user requests, resulting in test results that deviate from actual production environment conditions.

[0005] Performance data collection is typically scattered across multiple layers, including operating systems, middleware logs, and applications, lacking a unified, real-time monitoring view and correlation analysis capabilities. This leads to difficulties in problem localization and slow response times. In distributed or cloud-based deployment environments, rapid setup of test environments, elastic resource scaling, and automated scheduling of test tasks are common shortcomings of existing solutions. Therefore, building an efficient, accurate, and highly automated application server middleware performance testing system has become a pressing technical challenge for improving software quality and operational efficiency. Summary of the Invention

[0006] The purpose of this invention is to provide an application server middleware performance testing system to solve the technical problems in the prior art, such as low efficiency in test script construction, large deviation between load models and real business, scattered performance data collection and difficult analysis, and insufficient automation and flexibility.

[0007] This invention provides an application server middleware performance testing system, comprising: The business scenario intelligent modeling module is used to automatically convert business processes or user operation logs described in natural language into stress test scripts; The adaptive load generation and scheduling module is used to dynamically generate and apply stress loads that conform to the real business model based on the test strategy. The full-stack performance data acquisition and correlation analysis module is used to synchronously collect, correlate, and visualize performance indicator data from multiple layers of the middleware and runtime infrastructure of the application server under test. The automated test environment management module is used to enable rapid construction of test environments, elastic scaling of resources, and automated orchestration of test tasks.

[0008] Preferably, the intelligent modeling module for business scenarios includes a business process parser and a script generation engine; The business process parser is used to segment and semantically label the input natural language text or structured logs to identify business entities, operations, and the temporal and logical relationships between operations. The script generation engine is used to automatically generate test scripts that conform to the protocol specifications and contain dynamic variables, based on the parsing results of the business process parser and by calling the pre-set protocol template library and parameterized rule library.

[0009] Preferably, the adaptive load generation and scheduling module includes a multi-level load controller and a real-time feedback regulator; The multi-level load controller is used to receive the test script set and load model configuration file output by the business scenario intelligent modeling module, and divide the virtual users into multiple user groups with different behavioral characteristics at the user group level, and dynamically inject latency and data variation that conform to the real scenario into each request of each virtual user at the request level. The real-time feedback regulator is used to receive key system performance index data from the full-stack performance data acquisition and correlation analysis module in real time, and dynamically adjust the number of concurrent users or the request sending rate according to the preset adjustment algorithm.

[0010] Preferably, the full-stack performance data acquisition and correlation analysis module includes a lightweight acquisition probe and correlation analysis engine deployed on the server under test; The lightweight acquisition probes are deployed in the operating system kernel layer, middleware container layer and application code layer in a proxy or bypass manner to collect performance index data at each layer. All collected raw data are accompanied by nanosecond-level timestamps and unified transaction identifiers. The correlation analysis engine is used to align and correlate data streams from different levels in real time based on timestamps and transaction identifiers to build a full-link performance view. It also has a built-in multi-dimensional root cause analysis algorithm to automatically analyze the abnormal propagation path and locate the cause of the performance problem when an abnormality is detected.

[0011] Preferably, the automated test environment management module includes an environment blueprint library, an environment orchestrator, a resource elastic controller, and a task scheduler; The environment blueprint library is used to store predefined environment configuration templates for different middleware types and versions; The environment orchestrator is used to select a matching template from the environment blueprint library according to the test task requirements, and drive the underlying platform to automatically create or clone the test environment. The resource elasticity controller is used to continuously monitor the resource utilization rate of the test environment during test execution, and automatically trigger the horizontal expansion or contraction operation of resources when the resource utilization rate is continuously higher or lower than a preset threshold for a specified period of time. The task scheduler is used to manage the queue, priority, and dependencies of test tasks, and automatically performs environment cleanup and test report archiving operations after the tasks are completed.

[0012] Preferably, the business process parser has a built-in parsing unit based on a deep sequence model; The deep sequence model adopts a bidirectional long short-term memory network structure based on the attention mechanism. The input layer receives the text sequence after word embedding processing, the hidden layer captures contextual semantic information through bidirectional long short-term memory units, the attention layer calculates the contribution weight of each word in the input sequence to the current output action, and the output layer generates the corresponding script action element sequence.

[0013] Preferably, the real-time feedback regulator uses a proportional-integral-derivative (PID) control algorithm. The proportional-integral-derivative (PID) control algorithm uses the difference between the target response time and the actual average response time of the system under test as the error input. After calculation and superposition of the proportional, integral, and derivative terms, it outputs the adjustment amount for the number of concurrent virtual users.

[0014] Preferably, the multidimensional root cause analysis algorithm is constructed based on a causal graph model; The multidimensional root cause analysis algorithm first constructs the collected performance indicators as graph nodes, and the edges between nodes represent the potential causal relationships between the indicators. The weights of the edges are obtained through training with historical data. When a certain indicator becomes abnormal, the multidimensional root cause analysis algorithm traces back and performs probabilistic reasoning along the cause-effect graph, calculates the contribution of each upstream node that caused the abnormality, and determines the node with the highest contribution and its path as the most likely root cause.

[0015] Preferably, the environmental blueprint library is stored and managed using versioning. Each environment configuration template is strictly bound to the middleware version, operating system version, and dependency library version, and records the template creator, creation time, and modification history; The environment orchestrator instantiates environments strictly according to the specified version of the template.

[0016] Preferably, the protocol template library covers multiple application layer communication protocols, including Hypertext Transfer Protocol, Remote Method Invocation Protocol, and Database Connection Protocol. The parameterized rule base defines strategies for dynamically extracting or constructing request parameters from associated data sources.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention, through its intelligent business scenario modeling module, automates the conversion from natural language or operation logs into executable test scripts, reducing the technical threshold and time cost of test case construction. Its built-in deep sequence model accurately understands the temporal and semantic aspects of complex business logic, and the generated scripts highly replicate real user interactions, solving the problems of low efficiency and limited scenario coverage in traditional manual script writing, thus improving the breadth and depth of testing.

[0018] 2. This invention introduces a dynamic pressure adjustment mechanism based on real-time performance feedback through an adaptive load generation and scheduling module. This module not only simulates a business load model that conforms to statistical laws, but also intelligently adjusts the pressure based on the actual system performance during testing, realistically reproducing the dynamic characteristics of user behavior changing with system performance in a production environment. This closed-loop feedback control makes performance testing closer to real-world extreme scenarios, making the test results more valuable and helping to more accurately assess the system's resilience and stability boundaries.

[0019] 3. This invention constructs a unified performance monitoring system from infrastructure to application code through a full-stack performance data acquisition and correlation analysis module. By deploying multi-layer probes and correlating data based on high-precision timestamps and transaction identifiers, it achieves end-to-end visualization and rapid root cause localization of performance issues. Its multi-dimensional root cause analysis algorithm can automatically analyze anomaly propagation paths, freeing operations and maintenance personnel from tedious multi-log comparison work, shortening the fault diagnosis and performance optimization cycle, and improving operations and maintenance response efficiency.

[0020] 4. This invention achieves on-demand and resource optimization of the test environment through an automated test environment governance module. This module fully automates environment building, resource scaling, and task scheduling, supporting rapid replication and consistency assurance of the test environment. This enables frequent and efficient performance testing, seamlessly integrating into continuous integration and continuous delivery processes. Elastic resource control avoids idle and wasted test resources, reduces testing costs, and provides strong infrastructure support for agile development and DevOps practices. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the overall technical solution architecture of the present invention; Figure 2 This is a schematic diagram of the core principle framework of the intelligent modeling module for business scenarios in this invention; Figure 3 This is a schematic diagram of the multi-level interaction relationship and data flow of the adaptive load generation and scheduling module in this invention; Figure 4 This is a logical flow diagram of the full-stack performance data acquisition and correlation analysis module in this invention; Figure 5 This is a flowchart of the main stages of the automated test environment management module in this invention. Detailed Implementation

[0022] Example 1: The overall architecture of the application server middleware performance testing system provided by the present invention is shown in the attached figure. Figures 1 to 5 As shown, the system comprises four main functional units: a business scenario intelligent modeling module, an adaptive load generation and scheduling module, a full-stack performance data acquisition and correlation analysis module, and a test environment automated governance module. These modules collaborate efficiently through a unified data bus and predefined interface protocols, forming a complete closed-loop system covering test design, execution, monitoring, and environment management. This complete closed-loop system can be deployed on an independent test control node or distributed across multiple computing nodes to support large-scale concurrent test tasks. The following will be combined with the attached... Figure 1 To be continued Figure 5 The specific implementation methods of each module are explained in detail.

[0023] The business scenario intelligent modeling module automatically converts business processes described in natural language or structured user operation logs into executable stress test scripts. The internal structure of this module is shown in the attached figure. Figure 2 As shown, it comprises five sub-components: an input parser, a deep sequence model engine, a protocol template library, a parameterized rule library, and a script generator. When a user submits a piece of natural language text or uploads a structured log file, the input parser first preprocesses the raw data. For natural language text, a dictionary-based and context-aware word segmentation algorithm is used for segmentation, and a pre-trained semantic role labeling model is used to identify grammatical components such as subject, predicate, object, and time adverbs. For structured logs, fields such as request path, method type, parameter key-value pairs, and response status code are directly extracted. All parsing results are standardized into a unified intermediate representation format, including metadata such as operation actions, target resources, parameter constraints, and temporal dependencies.

[0024] The deep sequence model engine receives the aforementioned intermediate representation as input and initiates semantic modeling using a bidirectional long short-term memory (LSTM) network based on an attention mechanism. The input layer of this bidirectional LSM network maps each action to a fixed-dimensional word vector, the hidden layer captures historical and future contextual information through forward and backward LSM units respectively, and the attention layer dynamically calculates the contribution weight of each input action to the current output script element. Finally, the output layer generates an ordered sequence of actions that precisely corresponds to the request steps in the test script to be generated. During the training phase, the model uses a corpus consisting of over 100,000 historical test scripts and their corresponding business description texts. Unsupervised pre-training is performed first to learn general semantic patterns, followed by supervised fine-tuning for specific middleware (such as Tomcat and WebLogic) business scenarios to ensure the protocol compliance and logical integrity of the generated scripts.

[0025] After obtaining the action sequence, the script generator sequentially calls the protocol template library and the parameterization rule library to construct the final script. The protocol template library pre-constructs standard request templates for various communication protocols, such as Hypertext Transfer Protocol (HTTP), Remote Method Invocation Protocol (RSIP), and database connection protocol. Each template defines the necessary header fields, method type, path format, and authentication mechanism. For example, the GET request template for HTTP requires standard headers such as Host, User-Agent, and Accept, and supports dynamic injection of the Token authentication header. The parameterization rule library defines parameter sourcing strategies, including reading from a CSV file, querying from a database, extracting from the previous response using regular expressions, or constructing parameters using a random number generator. The generated test script is output in standard JMeter or Gatling script format, containing a complete transaction controller, assertion rules, timers, and variable declarations, and can be directly executed by subsequent modules.

[0026] The adaptive load generation and scheduling module is responsible for dynamically generating and applying stress loads that conform to the real business model based on the testing strategy. The interaction logic of this module is shown in the attached figure. Figure 3 As shown, the adaptive load generation and scheduling module includes a multi-level load controller, which comprises three functional units: a user group scheduler, a request-level mutation injector, and a real-time feedback regulator. The user group scheduler first parses the test script set output by the business scenario intelligent modeling module and, in conjunction with the load model configuration file provided by the user, divides the virtual users into several user groups. Each user group corresponds to a typical user behavior pattern, such as "high-frequency trading user," "low-frequency browsing user," or "mixed operation user." The load model configuration file explicitly defines the proportion of each user group, the maximum number of concurrent users, the ramp-up time, the stabilization duration, and the think time distribution.

[0027] Within a user group, a request-level mutation injector dynamically injects latency and data mutations into each request for each virtual user. Latency injection generates random wait times based on a configured distribution model to simulate pauses between operations by a real user. Data mutation is implemented using dynamic functions in a parameterized rule base. All mutation operations are completed immediately before the request is sent, ensuring the uniqueness and authenticity of each request.

[0028] The real-time feedback regulator continuously receives key system performance indicators, including average response time, error rate, and throughput, from the full-stack performance data acquisition and correlation analysis module via the data bus. When the average response time of the middleware under test exceeds a preset threshold for five consecutive sampling periods, or the error rate exceeds 5%, the regulator immediately initiates a load adjustment mechanism. This load adjustment mechanism employs a proportional-integral-derivative (PID) control algorithm, the mathematical expression of which is as follows: ; For at any time The increase in the number of concurrent virtual users that needs to be adjusted. The difference between the set target response time and the actual average response time. , , These are the proportional, integral, and differential coefficients, respectively, with typical values ​​of 0.8, 0.05, and 0.2. It is a moment The proportional term is used to quickly respond to the current error, the integral term eliminates long-term accumulated steady-state deviation, and the derivative term predicts the error change trend to prevent overshoot. The controller will calculate the error signal. The command is sent to the user group scheduler, which dynamically increases or decreases the number of active virtual users accordingly. For example, if the system response slows down, the scheduler will gradually reduce the number of concurrent users, simulating user churn due to poor experience in real-world scenarios; conversely, if the system performance is stable, the load can be appropriately increased to explore performance limits. This closed-loop control mechanism ensures the safety and realism of the testing process.

[0029] The full-stack performance data acquisition and correlation analysis module is used to synchronously collect, correlate, and visualize performance indicator data from multiple layers of the middleware and runtime infrastructure of the application server under test. The logical flow of this full-stack performance data acquisition and correlation analysis module is shown in the attached figure. Figure 4 As shown, the system comprises a three-layer probe deployment architecture: operating system layer probe, middleware layer probe, and application layer probe. The operating system layer probe runs as a lightweight daemon on the server under test. It periodically collects metrics such as CPU utilization (accuracy 0.1%), memory usage (in MB), disk I / O operation count (IOPS), and network transmission and reception bytes (in KB / s) by calling system call interfaces. The sampling frequency is once per second.

[0030] Middleware layer probes are integrated into middleware containers (such as Tomcat's Valve component or WebLogic's Diagnostic Framework) via embedded proxies. They obtain runtime metrics such as the number of active threads, idle threads, and maximum threads in the thread pool; the number of current connections and waiting connections in the database connection pool; the total number of garbage collection cycles; the time taken for each generation of garbage collection (in milliseconds); and the current depth of the request queue. The probes trigger data collection once per request processing cycle to ensure synchronization with business traffic.

[0031] Application-layer probes employ bytecode enhancement technology, dynamically modifying the bytecode of critical business methods during application startup via a Java Agent mechanism, and implanting monitoring stubs at method entry and exit points. These stubs record the method's start time, end time, call parameters, return value, and exception information, generating a unique transaction ID. This transaction ID is transparently transmitted via HTTP headers or message queue attributes during cross-service calls, ensuring end-to-end traceability. All raw data collected by the probes is appended with nanosecond-precision timestamps and a unified transaction ID, and sent to the correlation analysis engine via UDP or gRPC protocols.

[0032] After receiving data streams from the three-layer probes, the correlation analysis engine first aligns the sliding windows based on timestamps, and then aggregates metrics belonging to the same business transaction into a single end-to-end record based on transaction identifiers. For example, a record for a "user order placement" transaction will include CPU peak values ​​at the operating system layer, thread pool utilization at the middleware layer, order placement method time at the application layer, and database query time. Based on this, the engine's built-in multi-dimensional root cause analysis algorithm initiates anomaly detection. This algorithm is built on a causal graph model, first treating performance metrics as graph nodes, with edges between nodes representing statistical causal relationships in historical data, and edge weights obtained through Bayesian network training. When a metric exceeds the dynamic baseline, the algorithm traces back along the causal graph, calculating the conditional probability contribution of each upstream node. For example, if the response time is abnormal, and CPU utilization spikes simultaneously while the thread pool is saturated, the algorithm will determine "thread pool resource exhaustion" as the most likely root cause and generate an analysis report containing a time series graph, call chain screenshots, and a metric correlation matrix for testers to make quick decisions.

[0033] The automated test environment management module enables rapid test environment construction, elastic resource scaling, and automated orchestration of test tasks. The main stages of this module's workflow are shown in the appendix. Figure 5As shown, it comprises four components: an environment blueprint library, an environment orchestrator, a resource elastic controller, and a task scheduler. The environment blueprint library employs a versioned storage mechanism; each environment configuration template is defined in YAML format, strictly binding middleware type, operating system version, virtual machine specifications, container image address, network security group rules, and connection information for dependent services. Template version numbers follow semantic markup and record the creator, creation time, change logs, and checksum hash value to ensure traceability and consistency.

[0034] When a new test task is received, the environment orchestrator first parses the task metadata and matches the most suitable template version from the environment blueprint library. Then, the orchestrator calls the underlying Infrastructure as a Service (IaaS) platform or container orchestration platform via an application programming interface (API) to automatically create virtual machine instances, pull container images, configure network policies, and deploy middleware and dependent services. The entire build process typically completes within 5 minutes and verifies service availability via a health check interface.

[0035] During test execution, the resource elasticity controller continuously monitors the resource utilization of the test environment. Its monitoring metrics include average CPU utilization, remaining memory, and disk space utilization, with a sampling period of 30 seconds. When CPU utilization exceeds 85% for three consecutive periods, or remaining memory falls below 512MB, the controller triggers a horizontal scaling operation, requesting additional middleware Pod replicas from the Kubernetes cluster. Conversely, when CPU utilization falls below 30% for ten consecutive periods, a scaling operation is triggered, removing replicas. All scaling operations are executed using a rolling update strategy to ensure uninterrupted service.

[0036] The task scheduler manages the entire lifecycle of test tasks. It maintains a priority queue, supporting sorting by project, urgency, resource requirements, and other dimensions. Tasks can be triggered by timers, events, or manual intervention. The scheduler reserves necessary resources before a task begins, monitors its status during execution, and automatically executes a cleanup script and pushes the test report to a designated storage location upon completion. The entire process requires no manual intervention, enabling unattended operation of test tasks.

[0037] In summary, this embodiment constructs a highly automated, intelligent, and closed-loop application server middleware performance testing system through the deep collaboration of four major modules. The system not only solves the pain points of traditional testing, such as low script building efficiency, distorted load models, scattered data, and complex environment management, but also improves the realism, accuracy, and efficiency of performance testing by introducing advanced technologies such as deep learning, closed-loop feedback control, full-stack correlation analysis, and elastic resource governance. This provides solid support for quality assurance of modern high-concurrency, high-availability middleware systems.

[0038] Example 2: Building upon Example 1, this example further refines the enhancement mechanism of the intelligent business scenario modeling module when handling complex business processes, and expands the collaborative strategy of the adaptive load generation and scheduling module in multi-regional distributed testing scenarios. This example is applicable to large-scale financial or e-commerce systems whose business processes involve multi-service orchestration, asynchronous callbacks, and strong transaction consistency requirements, and whose users are distributed across multiple geographical regions globally.

[0039] In the intelligent modeling module for business scenarios, when the input natural language description contains asynchronous interaction logic such as "after the user submits a payment request, the system calls a third-party payment gateway, and updates the order status after receiving an asynchronous notification," the deep sequence model engine will additionally activate the state machine inference submodule. When generating the script action sequence, this submodule not only outputs synchronous request steps but also automatically generates a virtual service stub (Mock ServiceStub) for listening to asynchronous callbacks. This virtual service stub is configured to listen on a specific port, receive callback requests from the system under test, and return corresponding responses according to preset rules. When building the final script, the script generator embeds the start and stop instructions of this virtual service stub into the setup and teardown phases of the test lifecycle and inserts wait assertions in the main flow to ensure that a valid callback has been received before continuing to execute subsequent steps. Furthermore, for scenarios involving distributed transactions, the model engine will recognize "atomicity" semantics and add transaction rollback verification logic to the generated script, i.e., after simulating a network interruption, it checks whether the inventory and order status have both recovered to their initial values.

[0040] In the adaptive load generation and scheduling module, to support multi-region testing, the multi-level load controller has been extended to a geographically distributed architecture. When dividing user groups, the user group scheduler adds a "geographical location" dimension, such as "Eastern North America users," "Central Europe users," and "Southeast Asia Pacific users." Each geographical user group is assigned to a load generation node deployed in the corresponding cloud region. These nodes are synchronized through a global coordinator to ensure consistent load loading across regions. The request-level mutation injector, when injecting latency, considers not only local think time but also a network transmission latency model. The model is based on historical real user data, pre-setting the average round-trip time from each region to the tested data center, and dynamically adding latency values ​​before each request to more realistically simulate the intercontinental access experience.

[0041] In this scenario, the real-time feedback regulator employs a tiered adjustment strategy. The local regulator monitors the impact of load generated in its local region on the system under test, adjusting the number of concurrent users in that region only when the error rate exceeds the limit. The global regulator aggregates performance metrics from all regions and calculates the global average response time. If global metrics are abnormal, but local metrics in each region are normal, it is determined that the backend service of the system under test has a regional problem. In this case, the global regulator issues a unified load reduction command to all regions to prevent a cascading failure caused by local overload. This tiered mechanism ensures both the independence of regional testing and the consistency of global testing.

[0042] In this embodiment, the full-stack performance data acquisition and correlation analysis module enhances cross-regional transaction tracing capabilities. When generating transaction identifiers, the application-layer probe attaches a regional label. The correlation analysis engine uses the regional label as a key dimension when constructing the full-link view, supporting performance data filtering by region. The multi-dimensional root cause analysis algorithm has also been extended to identify performance anomalies caused by network partitioning. For example, when only the Asia-Pacific region experiences abnormal response times while other regions are normal, and the network packet loss rate from that region to the database is elevated, the algorithm identifies the root cause as "Asia-Pacific region network link failure," rather than an application code issue.

[0043] The automated governance module for the test environment supports the construction of multi-active data center environments. A new "Multi-Active Deployment" template type has been added to the environment blueprint library, defining the middleware configuration, data synchronization strategies, and traffic switching rules for the primary and backup data centers. When creating an environment, the environment orchestrator deploys service instances in both data centers simultaneously and configures a global load balancer. The resource elasticity controller monitors the independent resource pools of each data center, enabling on-demand scaling. The task scheduler supports the "Disaster Recovery Test" task type, which can automatically trigger primary data center outage drills and verify the backup data center's takeover capabilities and data consistency.

[0044] Through the above enhancements, this embodiment extends the system capabilities from single-point testing to global, highly complex business scenarios, further improving the real-world coverage and fault simulation capabilities of the tests. It is particularly suitable for large enterprise-level applications with stringent requirements for system resilience and global user experience.

[0045] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0046] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An application server middleware performance testing system, characterized in that, include: The business scenario intelligent modeling module is used to automatically convert business processes or user operation logs described in natural language into stress test scripts; The adaptive load generation and scheduling module is used to dynamically generate and apply stress loads that conform to the real business model based on the test strategy. The full-stack performance data acquisition and correlation analysis module is used to synchronously collect, correlate, and visualize performance indicator data from multiple layers of the middleware and runtime infrastructure of the application server under test. The automated test environment management module is used to enable rapid construction of test environments, elastic scaling of resources, and automated orchestration of test tasks.

2. The application server middleware performance testing system according to claim 1, characterized in that, The intelligent modeling module for business scenarios includes a business process parser and a script generation engine. The business process parser is used to segment and semantically label the input natural language text or structured logs to identify business entities, operations, and the temporal and logical relationships between operations. The script generation engine is used to automatically generate test scripts that conform to the protocol specifications and contain dynamic variables, based on the parsing results of the business process parser and by calling the pre-set protocol template library and parameterized rule library.

3. The application server middleware performance testing system according to claim 2, characterized in that, The adaptive load generation and scheduling module includes a multi-level load controller and a real-time feedback regulator. The multi-level load controller is used to receive the test script set and load model configuration file output by the business scenario intelligent modeling module, and divide the virtual users into multiple user groups with different behavioral characteristics at the user group level, and dynamically inject latency and data variation that conform to the real scenario into each request of each virtual user at the request level. The real-time feedback regulator is used to receive key system performance index data from the full-stack performance data acquisition and correlation analysis module in real time, and dynamically adjust the number of concurrent users or the request sending rate according to the preset adjustment algorithm.

4. The application server middleware performance testing system according to claim 3, characterized in that, The full-stack performance data acquisition and correlation analysis module includes a lightweight acquisition probe and correlation analysis engine deployed on the server under test. The lightweight acquisition probes are deployed in the operating system kernel layer, middleware container layer and application code layer in a proxy or bypass manner to collect performance index data at each layer. All collected raw data are accompanied by nanosecond-level timestamps and unified transaction identifiers. The correlation analysis engine is used to align and correlate data streams from different levels in real time based on timestamps and transaction identifiers to build a full-link performance view. It also has a built-in multi-dimensional root cause analysis algorithm to automatically analyze the abnormal propagation path and locate the cause of the performance problem when an abnormality is detected.

5. The application server middleware performance testing system according to claim 4, characterized in that, The automated test environment management module includes an environment blueprint library, an environment orchestrator, a resource elastic controller, and a task scheduler. The environment blueprint library is used to store predefined environment configuration templates for different middleware types and versions; The environment orchestrator is used to select a matching template from the environment blueprint library according to the test task requirements, and drive the underlying platform to automatically create or clone the test environment. The resource elasticity controller is used to continuously monitor the resource utilization rate of the test environment during test execution, and automatically trigger the horizontal expansion or contraction operation of resources when the resource utilization rate is continuously higher or lower than a preset threshold for a specified period of time. The task scheduler is used to manage the queue, priority, and dependencies of test tasks, and automatically performs environment cleanup and test report archiving operations after the tasks are completed.

6. The application server middleware performance testing system according to claim 5, characterized in that, The business process parser has a built-in parsing unit based on a deep sequence model; The deep sequence model adopts a bidirectional long short-term memory network structure based on the attention mechanism. The input layer receives the text sequence after word embedding processing, the hidden layer captures contextual semantic information through bidirectional long short-term memory units, the attention layer calculates the contribution weight of each word in the input sequence to the current output action, and the output layer generates the corresponding script action element sequence.

7. The application server middleware performance testing system according to claim 6, characterized in that, The real-time feedback regulator uses a proportional-integral-derivative (PID) control algorithm. The proportional-integral-derivative (PID) control algorithm uses the difference between the target response time and the actual average response time of the system under test as the error input. After calculation and superposition of the proportional, integral, and derivative terms, it outputs the adjustment amount for the number of concurrent virtual users.

8. The application server middleware performance testing system according to claim 7, characterized in that, The multidimensional root cause analysis algorithm is based on a causal graph model. The multidimensional root cause analysis algorithm first constructs the collected performance indicators as graph nodes, and the edges between nodes represent the potential causal relationships between the indicators. The weights of the edges are obtained through training with historical data. When a certain indicator becomes abnormal, the multidimensional root cause analysis algorithm traces back and performs probabilistic reasoning along the cause-effect graph, calculates the contribution of each upstream node that caused the abnormality, and determines the node with the highest contribution and its path as the most likely root cause.

9. The application server middleware performance testing system according to claim 8, characterized in that, The environmental blueprint library is stored and managed using versioning. Each environment configuration template is strictly bound to the middleware version, operating system version, and dependency library version, and records the template creator, creation time, and modification history; The environment orchestrator instantiates environments strictly according to the specified version of the template.

10. The application server middleware performance testing system according to claim 9, characterized in that, The protocol template library covers multiple application-layer communication protocols, including Hypertext Transfer Protocol, Remote Method Invocation Protocol, and Database Connection Protocol. The parameterized rule base defines strategies for dynamically extracting or constructing request parameters from associated data sources.

Images