Software development deployment system based on micro-service architecture

Abstract

The invention relates to a software development deployment system based on a micro-service architecture. The system comprises a micro-service management module, a load monitoring module, an output evaluation module, a task virtual module and a micro-service optimization module. A real-time resource monitoring system is constructed, and a load monitoring and output evaluation module is combined with a measured data correction mechanism, so that the accuracy of computing resource load and component output data is ensured, and a reliable basis is provided for optimization decision making; a task virtual module is innovatively designed, a virtual task in a complex scene can be simulated through task screening, learning and dynamic generation mechanisms in combination with a replacement ant algorithm and comprehensive reliable vector evaluation, and the adaptability test capability of a micro-service component under different load conditions is improved; and the micro-service management module determines load, value association and substitution logic among the components through component association relationship analysis and substitution feature generation, and provides data support for dynamic deployment and substitution.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of software development, more particularly, to a software development and deployment system based on micro-service architecture. BACKGROUND

[0002] With the wide application of micro-service architecture in the field of software development, software development and deployment systems based on micro-service have gradually become a research hotspot. However, the existing technical solutions still have deficiencies in micro-service management, load monitoring, output evaluation, and task virtualization, which are difficult to meet the efficient deployment requirements in complex scenarios, and also limit the flexibility and scalability of the system.

[0003] A software development platform based on micro-service inheritance is disclosed in CN118626054B. This patent constructs a digital model through a data modeling module and provides general attributes and behaviors using low-level models. High-level models achieve business specialization through inheritance and extension, thereby supporting efficient data management and rapid business expansion. However, this technical solution mainly focuses on data modeling and business application construction, lacking dynamic monitoring and optimization capabilities for resource load and output efficiency during micro-service operation. In addition, it does not involve task virtualization mechanisms, which cannot simulate the running state of micro-services in complex scenarios, thereby limiting the adaptability and reliability of the system. A monolithic and micro-service integrated software development method is disclosed in CN115729526B, published on August 20, 2024. This patent realizes automatic identification of monolithic and micro-service modes through custom annotation classes and aspect classes, and improves user experience through a unified front-end interface. However, this technical solution has obvious deficiencies in the running monitoring and optimization of micro-service after deployment, and does not provide real-time evaluation mechanisms for micro-service component load and output. It also does not involve the design of task triggering conditions and virtual task generation. This leads to the inability to adjust micro-service deployment strategies in time to meet dynamic requirements when dealing with high-concurrency or complex business scenarios.

[0004] The above problems show that the existing software development and deployment systems based on micro-service architecture still have certain deficiencies in micro-service management information generation, load monitoring and correction, output evaluation and optimization, and task virtualization. SUMMARY

[0005] Therefore, the present application aims to provide a software development and deployment system based on micro-service architecture.

[0006] To solve the above technical problems, the technical solution of the present application is: a software development and deployment system based on micro-service architecture, comprising a micro-service management module, a load monitoring module, an output evaluation module, a task virtualization module, and a micro-service optimization module.

[0007] The microservice management module is connected with a microservice component library storing a plurality of different types of microservice components, and is configured to generate microservice management information according to deployment feedback information;

[0008] The load monitoring module is configured to monitor the load of each deployed microservice component on the computing resources to generate microservice load data;

[0009] The output evaluation module is configured to evaluate the output result of each deployed microservice component to generate microservice output data;

[0010] The task virtualization module is configured with a plurality of task triggering conditions, and when a task triggering condition is met, a virtual task is generated and sent to the corresponding microservice component;

[0011] The microservice optimization module is configured with a service optimization strategy, which includes:

[0012] Step A1, obtaining microservice load data and microservice output data to generate optimization target information;

[0013] Step A2, processing the optimization target information according to the currently deployed microservice components to generate optimization clue information;

[0014] Step A3, comparing the optimization clue information with the microservice management information to generate a corresponding optimization scheme. The system connects the component library storing different types of microservice components through the microservice management module, generates management information according to deployment feedback information; the load monitoring module monitors the load of each component on the computing resources in real time to generate load data; the output evaluation module evaluates the output result of each component to generate output data; the task virtualization module generates a virtual task and sends it to the corresponding component when the triggering condition is met; the microservice optimization module obtains load and output data to generate optimization target information, processes the optimization clue in combination with the currently deployed components, and compares it with the management information to generate an optimization scheme. The system realizes the full life cycle management and dynamic optimization of microservice components, and through the collaborative work of multiple modules, it can monitor the resource load and output efficiency in real time, generate optimization schemes based on data, solve the deficiencies in microservice management, load monitoring and other aspects in the prior art, improve the deployment efficiency and flexibility of the system in complex scenarios, and enhance the adaptability and expandability of the system.

[0015] Further, the microservice management module comprises a load management unit, a value management unit and an association management unit, the load management unit is configured with a preset load management algorithm for calculating static load data of the corresponding microservice component, the value management unit is configured with a preset value management algorithm for calculating static output data of the corresponding microservice component, and the association management unit is configured with an association sub-strategy, the association sub-strategy comprises

[0016] Step B1, service characteristics of each microservice component are analyzed to obtain a service characteristic group;

[0017] Step B2, service characteristic connections are generated according to the association relationship between the microservice components, and the association relationship is generated according to the comparison result between the service characteristic groups;

[0018] Step B3, a preset replacement constraint condition is configured, a service characteristic connection type meeting the replacement constraint condition is configured as a replacement characteristic connection, and static load data and static output data are obtained according to the association relationship to generate corresponding replacement characteristics;

[0019] Step B4, the microservice management information is generated according to the replacement characteristics. The microservice management module comprises a load management unit, a value management unit and an association management unit, the load management unit calculates the static load data of the component by using a preset algorithm, the value management unit calculates the static output data, the association management unit obtains the characteristic group by analyzing the service characteristics of the component, generates the association relationship and the service characteristic connection according to the comparison result of the characteristic group, sets the connection meeting the replacement constraint condition as the replacement characteristic connection, generates the replacement characteristics by combining the load and output data of the association relationship, and finally forms the microservice management information. Through the fine management of the load, value and association relationship of the microservice component, the replacement characteristics and the association relationship between the components are clarified, data support is provided for the dynamic replacement and optimization of the microservice, the management ability of the system for the association relationship between the components is improved, and the system has more flexibility and reliability in the deployment and adjustment of the components.

[0020] Further, the load monitoring module is configured with a load calculation unit and a measured load correction unit, the load calculation unit is configured with a preset load calculation strategy algorithm to generate a measured load value according to the running information of the microservice component, the measured load correction unit is configured with a load correction condition and a load correction algorithm, when the measured load value meets the load correction condition, the load correction algorithm is executed to generate a load correction factor to the microservice component library;

[0021] The output evaluation module is configured with an output calculation unit and a measured output correction unit. The output calculation unit is configured with a preset output calculation algorithm to generate a measured output value based on the running information of the micro-service components. The measured output correction unit is configured with an output correction condition and an output correction algorithm. When the measured output value meets the output correction condition, the output correction algorithm is executed to generate an output correction factor for the micro-service component library. The load calculation unit of the load monitoring module generates a measured load value based on the component running information. The measured load correction unit executes an algorithm to generate a load correction factor for the component library when the load value meets the correction condition. The output calculation unit of the output evaluation module generates a measured output value. The measured output correction unit executes an algorithm to generate an output correction factor for the component library when the output value meets the condition. Through real-time monitoring and correction of the measured load value and the output value, the accuracy of the load data and the output data is improved, providing a reliable data foundation for subsequent micro-service optimization, ensuring that the system can make decisions based on real and accurate data, avoiding optimization errors caused by data deviation, and improving the stability and reliability of the system.

[0022] Further, the load monitoring module further includes a quasi-state load unit, the quasi-state load unit is configured with a load analysis sub-strategy, the load analysis sub-strategy is used to generate a load correction factor according to the running information of the micro-service component executing a virtual task;

[0023] The output evaluation module is configured with a quasi-state output unit, the micro-service component processes a virtual task to generate a virtual output, the quasi-state output unit is configured with an output analysis sub-strategy, the output analysis sub-strategy is used to generate the virtual correction factor according to the virtual output. The quasi-state load unit of the load monitoring module generates a load correction factor according to the running information of the component executing a virtual task through the load analysis sub-strategy; the quasi-state output unit of the output evaluation module generates a virtual correction factor through the output analysis sub-strategy according to the virtual output generated by the component processing a virtual task. The virtual task simulates the running state of the component in different scenarios, further perfecting the correction mechanism of the load and output data, enabling the system to evaluate the component performance in non-actual business scenarios, discovering potential problems in advance, enhancing the adaptability of the system to complex scenarios, and improving the stability and reliability of the micro-service component under different task loads.

[0024] Further, the task virtualization module includes a task screening unit and a task learning unit. The task screening unit is configured with a task evaluation algorithm, which is used to evaluate micro-service tasks and generate a task evaluation vector. The task screening unit is configured with a screening constraint. When the task evaluation vector meets the screening constraint, the corresponding micro-service task is output to the task learning unit. The task learning unit is configured with a task learning sub-strategy, which includes:

[0025] Step C1, decompose the current micro-service task to obtain several task sub-items, and identify the editable task sub-items through the pre-constructed editable identification model;

[0026] Step C2, call the editing data sub-item corresponding to the editable task sub-item from the pre-constructed association replacement library, associate the editing data sub-item with the editable task sub-item to establish an editing association network, wherein the editing association network has a task sub-item as a center node and an editing data sub-item as a branch node, and the connection between the branch node and the center node is configured with a dynamic editing factor;

[0027] Step C3, calculate the dynamic editing factor through the preset dynamic editing algorithm, and return to step C2 until the traversal of all editable task sub-items is completed, and the editing association network of the current micro-service task is configured in the preset task generation model. The task screening unit of the task virtual module generates a task evaluation vector using a task evaluation algorithm, and the tasks that meet the screening constraints are output to the task learning unit; the learning unit decomposes the task to obtain sub-items, identifies editable sub-items through an editable identification model, calls corresponding editing data sub-items from an association replacement library to establish an editing association network with a task sub-item as a center node and an editing data sub-item as a branch node, configures a dynamic editing factor, traverses all sub-items after calculation by a dynamic editing algorithm, and configures the editing association network to a task generation model. Through the screening, decomposition and learning of the micro-service task, an editable task association network is constructed, a structured model is provided for the generation of virtual tasks, the system can flexibly generate diversified virtual tasks according to actual needs, and the flexibility and adaptability of the system in task simulation and processing are improved, and a more comprehensive task scenario is provided for the testing and optimization of micro-service components.

[0028] Further, the task virtual module further comprises a task generation unit, and the task generation unit is configured with a task generation sub-strategy, and the task generation sub-strategy comprises:

[0029] Step D1, determine the task type of the current micro-service task, and call the corresponding editing association network from the task generation model according to the task type;

[0030] Step D2, calculate the replacement weight corresponding to each dynamic editing factor through a preset replacement weight algorithm according to the network environment data;

[0031] Step D3, generate a replacement ant at the center node, and determine the replacement ant running path according to the replacement weight until the replacement ant stays at a certain branch node, the replacement ant is generated with repulsive pheromone, and the repulsive pheromone is used to reduce the replacement weight corresponding to the branch node on the replacement ant path;

[0032] Step D4, calculate the comprehensive reliability vector corresponding to the branch node where the replacement ant is located by a preset vector evaluation algorithm, and return to step D2 until the comprehensive reliability vector meets the preset vector constraint, which is generated according to the network environment information at the current moment;

[0033] Step D5, configure the micro-service task benchmark with the task type, and replace the editing data sub-item corresponding to the branch node to the corresponding micro-service task benchmark to generate the virtual task. The task generation unit of the task virtual module determines the task type, retrieves the corresponding editing associated network from the task generation model, calculates the replacement weight of the dynamic editing factor according to the network environment data, generates a replacement ant at the center node, determines the path according to the weight and generates repulsive pheromone to reduce the branch weight, calculates the comprehensive reliability vector of the branch node by the vector evaluation algorithm, until the vector constraint is met, and finally replaces the editing data sub-item to the task benchmark to generate the virtual task. The virtual task is dynamically generated in combination with the network environment, and through the evaluation of the replacement ant algorithm and the comprehensive reliability vector, it is ensured that the generated virtual task has high security, uniqueness and verifiability, improving the quality and effectiveness of the virtual task, so that the system can simulate real task scenarios in complex network environments, providing more accurate task input for performance testing and optimization of micro-service components. Further, the comprehensive reliability vector includes a security component, a uniqueness component, and a verification component, the security component is used to evaluate the content security of the current virtual task, the uniqueness component is used to evaluate the content uniqueness of the current virtual task, and the verification component is used to evaluate the verifiability of the current virtual task. The comprehensive reliability vector includes a security component, a uniqueness component, and a verification component, which are used to evaluate the content security, content uniqueness, and verifiability of the virtual task, respectively. The virtual task is comprehensively evaluated from multiple dimensions to ensure that the virtual task meets the requirements in terms of content security, uniqueness, and verifiability, avoiding the generation of invalid or unsafe tasks, improving the reliability and effectiveness of the virtual task, and further ensuring the stability and accuracy of the micro-service components when processing the virtual task.

[0034] Further, step A1 includes:

[0035] Step A1-1, configure a preset benefit evaluation algorithm to calculate the benefit value of each micro-service component;

[0036] Step A1-2, configure a demand prediction model to generate a theoretical demand task at the next moment;

[0037] Step A1-3, compare the theoretical demand task with the currently deployed micro-service components to generate a missing data set;

[0038] Step A1-4, determining the microservice components to be optimized according to the missing data set and the revenue value to generate the optimization target information. In step A1 of the optimization module, the revenue evaluation algorithm is used to calculate the component revenue value, the demand prediction model is used to generate the theoretical demand task at the next time, the theoretical task is compared with the current deployed components to generate the missing data set, and the missing data set and the revenue value are used to determine the components to be optimized to generate the optimization target information. Based on the revenue evaluation and demand prediction, the microservice components to be optimized are accurately positioned, the optimization target is more targeted, blind optimization is avoided, the optimization efficiency is improved, and the system can be reasonably adjusted according to the future demand and component revenue, improving the resource utilization and business adaptability of the system.

[0039] Further, the step A3 includes:

[0040] Step A3-1, determining the microservice components to be optimized and the optimization task target according to the optimization clue information;

[0041] Step A3-2, generating the combination of optimization sub-schemes of the microservice components according to the microservice management information to generate a plurality of proposed optimization schemes;

[0042] Step A3-3, configuring an optimization evaluation algorithm to evaluate the comprehensive optimization value of each proposed optimization scheme, and determining the proposed optimization scheme with the highest comprehensive optimization value as the output optimization scheme. In step A3 of the optimization module, the components to be optimized and the optimization task target are determined according to the optimization clue information, the optimization sub-scheme combination is generated according to the microservice management information to form the proposed optimization scheme, the comprehensive optimization value of each scheme is calculated by the optimization evaluation algorithm, and the scheme with the highest comprehensive optimization value is determined as the output. By generating a plurality of proposed optimization schemes and performing comprehensive evaluation, the optimal scheme is selected for implementation, ensuring the scientificity and rationality of the optimization decision, avoiding the limitations of a single scheme, improving the effect of microservice deployment optimization, and enabling the system to achieve optimal configuration in terms of resource allocation and component adjustment, further improving the performance and efficiency of the system.

[0043] The technical effects of the present application mainly embody in the following aspects: a real-time resource monitoring system is constructed, a load monitoring and output evaluation module are combined with a measured data correction mechanism to ensure the accuracy of the operation resource load and component output data, and to provide reliable basis for optimization decision; a task virtual module is innovatively designed, through task screening, learning and dynamic generation mechanism, combined with replacement ant algorithm and comprehensive reliable vector evaluation, virtual tasks in complex scenarios can be simulated, and the adaptability test ability of micro service components under different load conditions is improved; the micro service management module generates replacement features through component association relationship analysis, and the load, value association and replacement logic between components are clear, which provides data support for dynamic deployment and replacement; the optimization module realizes accurate positioning and optimal scheme generation of the to-be-optimized components based on revenue evaluation, demand prediction and multi-scheme comprehensive evaluation, improves resource utilization and system performance, solves the problems of insufficient dynamic monitoring, task simulation and rough optimization decision in the prior art, and significantly enhances the deployment efficiency, flexibility and reliability of the system in complex scenarios. BRIEF DESCRIPTION OF DRAWINGS

[0044] Figure 1 The system architecture principle diagram of the present application.

[0045] The micro service management module 100, the load management unit 110, the value management unit 120, the association management unit 130, the load monitoring module 200, the load calculation unit 210, the measured load correction unit 220, the analog load unit 230, the output evaluation module 300, the output calculation unit 310, the measured output correction unit 320, the analog output unit 330, the task virtual module 400, the task screening unit 410, the task learning unit 420, the task generation unit 430, and the micro service optimization module 500. DETAILED DESCRIPTION

[0046] The specific embodiments of the present application are further described in detail below with reference to the accompanying drawings, so that the technical scheme of the present application is easier to understand and master.

[0047] A software development and deployment system based on a micro service architecture, comprising a micro service management module 100, a load monitoring module 200, an output evaluation module 300, a task virtual module 400, and a micro service optimization module;

[0048] The micro service management module 100 is connected with a micro service component library, the micro service component library stores a plurality of different types of micro service components, and the micro service management module 100 is used for generating micro service management information according to deployment feedback information;

[0049] The microservice management module 100 comprises a load management unit 110, a value management unit 120, and an association management unit 130. The load management unit 110 is configured with a preset load management algorithm for calculating static load data of the corresponding microservice component. The value management unit 120 is configured with a preset value management algorithm for calculating static output data of the corresponding microservice component. The association management unit 130 is configured with an association sub-strategy, which comprises

[0050] Step B1, analyzing the service characteristics of each microservice component to obtain a service characteristic group;

[0051] Step B2, generating service characteristic connections according to the association relationship between the microservice components, which is generated according to the comparison result between the service characteristic groups;

[0052] Step B3, configuring a preset substitution constraint condition, configuring the service characteristic connection type satisfying the substitution constraint condition as a substitution characteristic connection, and obtaining static load data and static output data according to the association relationship to generate a corresponding substitution characteristic;

[0053] Step B4, generating the microservice management information according to the substitution characteristic. The microservice management module 100 is connected to a microservice component library storing microservice components in different business fields such as order processing user authentication. The module generates microservice management information based on deployment feedback information, and the core purpose is to mark the functions that can be substituted by the component functions, so that the microservice components can be quickly indexed when they need to be replaced. In addition, it also includes basic evaluation of load output, in order to have data quantization basis when replacing or needing to increase microservice instances. The module comprises a load management unit 110, a value management unit 120, and an association management unit 130.

[0054] The load management unit 110 is configured with a preset load management algorithm, which calculates static load data by collecting running parameters such as CPU occupancy rate, memory throughput, and disk I / O frequency of the microservice component during running. For example, a certain order processing microservice has a CPU occupancy rate of 30%, a memory throughput of 200 MB / s, and a disk I / O frequency of 50 times / s under standard test scenario. The algorithm calculates these parameters according to the preset weight, in which the CPU weight accounts for 40%, the memory accounts for 35%, and the I / O accounts for 25%. The weighted sum of the static load value of the component is 0.3 x 40% + 0.2 x 35% + 50 x 0.001 x 25% = 0.2325, which is the value after dimensionless processing.

[0055] The value management unit 120 calculates the static output data by a value management algorithm, which evaluates the output efficiency based on the throughput, response time, error rate and other indicators of the component processing tasks. Taking the user authentication microservice as an example, it processes 200 authentication requests per second with an average response time of 150 ms and an error rate of 0.1%. The algorithm normalizes the throughput to 0.8, the response time to 0.9, and the error rate to 0.95 through a mapping relationship. After weighted calculation, where the throughput accounts for 40%, the response time accounts for 35%, and the error rate accounts for 25%, the static output value is 0.8 x 40% + 0.9 x 35% + 0.95 x 25% = 0.8775.

[0056] The association management unit 130 constructs the association relationship between components through association sub-strategies. In step B1, the service feature group refers to a set of features formed by analyzing the interface protocol data format and function description of the microservice component, such as the service feature group of the payment microservice, which includes supporting the UnionPay and Alipay interface processing amount precision to cents, returning the payment status code, etc. In step B2, by comparing the service feature groups of different components, if two components have more than 70% of the feature items matching, a service feature connection is generated to represent the association relationship, for example, the order microservice and the inventory microservice both contain the goods ID parameter processing inventory deduction interface feature, with a feature matching degree of 75%, thus establishing an association connection.

[0057] In step B3, the replacement constraint conditions include functional equivalence, such as two user authentication microservices both supporting OAuth2.0 protocol, and performance deviation threshold, such as the response time deviation of the replacement component not exceeding 15% of the original component, etc. The service feature connection that meets the conditions is marked as a replacement feature connection. For example, when the main payment microservice needs to be replaced due to high traffic, if the functional feature matching degree of the standby payment microservice reaches 80% and the response time deviation is within 10%, the connection between the two is set as a replacement feature connection, and the static load data of the two is obtained according to the association relationship, the main component 0.25 and the standby component 0.23, and the static output data of the two is obtained, the main component 0.85 and the standby component 0.83, a replacement feature containing the load difference -0.02 and the output difference -0.02 is generated.

[0058] In step B4, all replacement feature component basic information such as ID version, static load data, and static output data are integrated into microservice management information, stored in a structured data format, for example, a certain management information record is, component A ID: MS001, service feature group [interface type RESTful data format JSON processing order creation], replacement component list [MS003 matching degree 82% load difference -0.03 output difference +0.01], static load 0.22, static output 0.88.

[0059] In the e-commerce platform application scenario, when the order creation microservice MS001 has performance bottlenecks due to promotional activities, the microservice management module 100 quickly indexes to the alternative component MS003 through management information, the service feature matching degree of which is 82%, the load value is 0.19 which is lower than the original component 0.22, and the output value is 0.89 which is slightly higher than the original component. Based on these data, the system quantifies the decision, switches part of the traffic to MS003, realizes dynamic replacement of components, adjusts the number of instances according to the load data to ensure efficient use of resources.

[0060] The load monitoring module 200 is configured to monitor the load of each microservice component deployed on the computing resources to generate microservice load data; the load monitoring module 200 is configured with a load calculation unit 210 and a measured load correction unit 220, the load calculation unit 210 is configured with a preset load calculation algorithm to generate a measured load value according to the running information of the microservice component, the load calculation unit 210 uses a load evaluation model based on time series dynamic weight, and the algorithm formula is as follows:

[0061]

[0062] Wherein, L(t) is the measured load value at time t, n is the resource index dimension which can be CPU, memory, network, α i (t) is the dynamic weight coefficient of the i-th index at time t, which is calculated by exponential moving average (EMA):

[0063]

[0064] In the formula, λ is the forgetting factor (value 0.85), var(x i (t-10:t-1)) represents the variance of the i-th index in the previous 10 time windows; x i (t) is the original value of the i-th index at time t, μ i is the historical mean of the index, σ i is the historical standard deviation, ∈ is a small quantity 10 -6 , to prevent division by zero error, w iis the basic weight of the index. Take an e-commerce recommendation microservice as an example, its resource indicators include CPU occupancy rate x1, unit %, memory bandwidth x2, unit GB / s, network traffic x3, unit Mbps, basic weights w1=0.4, w2=0.3, w3=0.3. The data collected at time t is: x1=75, μ1=60, σ1=10; x2=2.8, μ2=2.0, σ2=0.5; x3=800, μ3=500, σ3=150. The dynamic weights at the previous time are α1(0)=0.45, α2(0)=0.3, α3(0)=0.25, the CPU variance of the previous 10 windows is 25, the memory variance is 0.16, and the network variance is 9000, then:

[0065]

[0066] Similarly, α2(1)≈0.300, α3(1)≈0.317, substitute into the load formula: L(1)=0.383×1.5×0.4+0.300×1.6×0.3+0.317×2.0×0.3=0.564.

[0067] The measured load correction unit is configured with load correction conditions and load correction algorithm, when the measured load value meets the load correction conditions, the load correction algorithm is executed to generate a load correction factor to the microservice component library; the measured load correction unit adopts a double filtering correction mechanism, first, the ARIMA(1,1,1) model is used to predict the load trend: ΔL(t)=φ1ΔL(t-1)+θ1∈(t-1)+∈(t) where ΔL(t) is the load change at time t, φ1 is the autoregressive coefficient, θ1 is the moving average coefficient, and ∈(t) is the white noise sequence. When the measured load value deviates from the predicted value by more than a threshold value δ, for example, more than 20%, Kalman filter correction is started: wherein is the predicted load, K(t) is the Kalman gain, P(t|t-1) is the prediction error covariance, Q is the process noise covariance, and R is the measurement noise covariance. The correction factor is calculated as:

[0068]

[0069] Take an order microservice as an example, the measured load L(10) at time t-1 is 0.65, the predicted value the error covariance P(10|9)=0.02, the process noise Q=0.001, and the measurement noise R=0.01, then: the measured load L(11) at time t is 0.78, the predicted value Deviation over 20% triggers correction: K(11) = (0.0067 + 0.001) / (0.0067 + 0.001 + 0.01) ≈ 0.696 Correction factor F(11) = (0.696 - 0.78) / 0.78 ≈ -0.108, i.e. load reference value is lowered by 10.8%.

[0070] The load monitoring module 200 further comprises a quasi-state load unit 230, which is configured with a load analysis sub-strategy for generating a load correction factor according to running information when the microservice component executes a virtual task; the quasi-state load unit constructs a CNN-LSTM hybrid neural network model to generate a load correction factor through virtual task execution data. The model input is multi-dimensional resource time series data (such as CPU, memory, I / O indicators collected every 50 ms) when the virtual task is executed, and the spatial features are extracted through the CNN layer:

[0071] f l = σ(W l *f l-1 +b l ) In the formula, f l is the l-th layer feature map, W l is the convolution kernel, b l is the bias, and σ is the ReLU activation function. Then through the LSTM layer to capture the time series dependence: f t = σ(W xf x t +W hf h t-1 +W cf c t-1 +b f ), c t = f t ⊙c t-1 +i t ⊙tanh(W xc x t +W hc h t-1 +b c ), o t = σ(W xo x t +W ho h t-1 +W co c t +b o ), h t = o t ⊙tanh(c t ), and finally through the fully connected layer to output the load correction factor prediction value: F v = softmax(W f hT +b f )。

[0072] where, f l : feature map of the i-th layer of CNN network, dimension HxWxC (H is height, W is width, C is channel number. W l : convolution kernel weight matrix of the i-th layer of CNN network, dimension KxKxC i nxC o ut, K is the size of convolution kernel, C i n is the input channel number, C o ut is the output channel number. b l : bias vector of the i-th layer of CNN network, dimension C o ut. σ: ReLU activation function, mathematical expression is σ(x) = max(0, x), used to introduce nonlinear characteristics. t : input gate output of LSTM network at time t, value range [0, 1], control the degree of updating the current input to the cell state. f t : forget gate output of LSTM network at time t, value range [0, 1], control the degree of forgetting the historical cell state. o t : output gate output of LSTM network at time t, value range [0, 1], control the degree of output of cell state to hidden state. W x i, W h i, W c i: weight matrix of LSTM network input gate, corresponding to the weights of input x t , hidden state h t-1 , cell state c t-1 , dimension DxD (D is the dimension of hidden layer), dimensionless. b i ,b f ,b o ,b c : bias vector of each gate of LSTM network, dimension D, dimensionless. c t : cell state of LSTM network at time t, dimension D, store long-term memory information. h t : hidden state of LSTM network at time t, dimension D, store short-term memory information. F v : paratonic load correction factor, value range is usually [-0.5, 0.5], positive number means to increase the load reference value, negative number means to reduce. W f : weight matrix of fully connected layer, dimension Dx1, used to map the LSTM hidden state to the correction factor. b f : bias of fully connected layer, scalar, used to adjust the output offset of fully connected layer.

[0073] In the financial transaction system stress test scenario, the payment microservice is injected with a virtual high-frequency transaction task: 2000 transactions per second, and 1000 groups of 50ms interval resource data are collected: CPU occupancy rate sequence, memory bandwidth sequence, and disk I / O sequence. The CNN-LSTM model is input: the CNN layer contains 64 3x3 convolution kernels, and the LSTM layer contains 128 neurons. After training, the output correction factor of the model for a certain virtual task is 0.23, that is, the load of the microservice in the real high-frequency transaction scenario needs to be increased by 23% based on the standard value. The system adjusts the load benchmark value of this component from 0.45 to 0.55 accordingly, and expands 2 instances in advance to ensure that the component load remains below the safety threshold of 0.8 during the transaction peak period.

[0074] The load monitoring module 200 realizes high-precision monitoring and prediction of microservice resource load through dynamic weight load calculation, double filtering correction, and deep learning isomorphic analysis, provides multi-dimensional load data including time series features, abnormal fluctuation correction, and extreme scenario simulation for the microservice optimization module, and effectively solves the technical problems of load evaluation lag in dynamic scenarios and insufficient adaptability of complex business modes in traditional monitoring solutions.

[0075] The output evaluation module 300 is configured to evaluate the output results of each deployed microservice component to generate microservice output data.

[0076] The output evaluation module 300 is configured with an output calculation unit 310 and a measured output correction unit 320. The output calculation unit 310 is configured with a preset output calculation algorithm to generate a measured output value based on the running information of the microservice component. The output calculation unit 310 uses a multi-index comprehensive evaluation model based on entropy weight-TOPSIS, and the algorithm formula is:

[0077]

[0078] Where O(t) is the measured output value at time t, m is the number of evaluation indexes, ω j is the entropy weight coefficient of the jth index, which is calculated by entropy method:

[0079]

[0080] Where e j is the entropy value of the jth index, p ij is the proportion of the ith index in the jth sample, and n is the sample size; r j (t) is the original value of the jth output index at time t, and respectively. Taking the user authentication microservice as an example, the throughput r1, unit: request / s, the response time r2, unit: ms, and the error rate r3, unit: % are selected as evaluation indexes. The historical data shows that the throughput range is [100, 300], the response time range is [50, 200], and the error rate range is [0, 1]. The data collected at time t is: the throughput is 220 requests / s, the response time is 80 ms, and the error rate is 0.3%. The normalized values of each index are calculated: the throughput is (220-100) / (300-100) = 0.6, the response time is (200-80) / (200-50) = 0.8, and the error rate is (1-0.3) / (1-0) = 0.7. The weights are calculated by the entropy value method, assuming that the sample number n = 100, e1 = 0.92, e2 = 0.88, and e3 = 0.95 are calculated, then ω1 = 0.32, ω2 = 0.48, and ω3 = 0.2, and the output formula is substituted: O(t) = 0.32*0.6 + 0.48*0.8 + 0.2*0.7 = 0.192 + 0.384 + 0.14 = 0.716.

[0081] The measured output correction unit 320 is configured with an output correction condition and an output correction algorithm. When the measured output value meets the output correction condition, the output correction algorithm is executed to generate an output correction factor to the microservice component library. The measured output correction unit 320 adopts an adaptive correction mechanism based on Bayesian estimation. First, the exponential smoothing method is used to predict the output trend: wherein is the predicted output value at time t, and α is the smoothing coefficient (value 0.7). When the deviation between the measured output value and the predicted value exceeds the threshold η, 15%, the Bayesian correction is started: P(O(t)|D)∝P(D|O(t))·P(O(t)) In the formula, P(O(t)) is the prior probability distribution of the output value, P(D|O(t)) is the likelihood function, and D is the observation data. The correction factor is calculated as:

[0082]

[0083] wherein E[O(t)|D] is the posterior expectation based on the observation data. Taking the order processing microservice as an example, the measured output O(10) = 0.82 at time t-1, and the predicted value The measured output O(11) = 0.65 at time t = 11, and the deviation is (0.79-0.65) / 0.79≈17.7% which exceeds 15% to trigger the correction. Assuming that the prior distribution of the output value obeys the normal distribution N(0.75, 0.05 2 ), and the likelihood function is based on the historical error distribution N(0, 0.06 2 ), the posterior distribution is:

[0084]

[0085] The posterior expectation E[O(11)|D] = 0.702, and the correction factor C(11) = (0.702-0.65) / 0.65≈0.08, i.e. the output benchmark value is increased by 8%.

[0086] The output evaluation module 300 is configured with a mimic output unit 330, which processes virtual tasks to generate virtual output, and the mimic output unit is configured with an output analysis sub-strategy for generating the virtual correction factor according to the virtual output. The mimic output unit constructs a Transformer-based generative model to generate a mimic correction factor through the output data of the virtual task. The model input is the multi-dimensional output time series data during the execution of the virtual task, such as the throughput, response time, and error rate sequence collected every 100 ms, which captures long-distance dependencies through a multi-head self-attention mechanism:

[0087]

[0088] where Q, K, V are query, key, and value matrices, and d_k is the key vector dimension. The correction factor is generated by a feedforward neural network: F o =σ(W2·ReLU(W1·Z+b1)+b2) where Z is the output of the attention layer, W1 and W2 are weight matrices, and b1 and b2 are bias vectors. In the financial risk control microservice stress testing scenario, a virtual high-risk transaction task is injected: 5000 transactions per second, and 2000 groups of output data are collected at 100 ms intervals: throughput sequence, response time sequence, and error rate sequence, which are input into the Transformer model: 6-layer encoder, 8-head attention, and hidden layer dimension 512. After training, the output correction factor of the model for a certain virtual task is 0.15, i.e. the output of the microservice in the real high-risk transaction scenario needs to be increased by 15% based on the standard value, and the system adjusts the output benchmark value of this component from 0.7 to 0.805 accordingly, and optimizes the resource allocation strategy to ensure that the risk control rule execution efficiency is improved by more than 15%. The output evaluation module realizes multi-dimensional accurate evaluation of microservice output efficiency through entropy weight-TOPSIS multi-index evaluation, Bayesian adaptive correction, and Transformer mimic analysis, provides the microservice optimization module with multi-dimensional output information including real-time output data, abnormal fluctuation correction, and extreme scenario simulation, and effectively solves the technical problems of fixed index weight, insufficient dynamic adaptability, and missing extreme scenario evaluation in traditional evaluation schemes.

[0089] The task virtual module 400 is configured with a plurality of task trigger conditions, and when the task trigger conditions are met, a virtual task is generated and sent to the corresponding microservice component; the task virtual module 400 includes a task screening unit 410 and a task learning unit 420,

[0090] The task screening unit 410 is configured with a task evaluation algorithm, which is used to evaluate microservice tasks and generate a task evaluation vector. The task screening unit 410 is configured with a task evaluation algorithm, which generates a task evaluation vector through multi-dimensional indexes, and the formula is:

[0091]

[0092] Among them, is the task evaluation vector, e1 is the task importance component, T(t) is the task trigger frequency at time t (unit times / second), T max is the historical maximum trigger frequency; e2 is the resource correlation component, R(t) is the number of task associated resources, R avg is the average number of associated resources; e3 is the error sensitivity component, E(t) is the historical error rate of the task (unit %); e4 is the business impact component, C(t) is the business value of the task, C ref is the reference business value. The screening constraint condition is (τ is the screening threshold, and the value is 0.7). Taking the order creation task in the e-commerce promotion activity as an example, T(t) = 500 times / second, T max = 800 times / second, R(t) = 5 (5 types of resources such as associated users, inventory, payment, etc.), R avg = 3, E(t) = 0.5%, C(t) = 100 yuan, C ref = 50 yuan, then: e1 = 0.625, e2 = 1.667, e3 = 0.995, e4 = 2; the task evaluation vector norm satisfies the screening constraint and is output to the task learning unit.

[0093] The task screening unit 410 is configured with a screening constraint, and when the task evaluation vector satisfies the screening constraint, the corresponding microservice task is output to the task learning unit 420, and the task learning unit 420 is configured with a task learning sub-strategy, and the task learning sub-strategy includes:

[0094] Step C1, decompose the current microservice task to obtain a plurality of task sub-items, and identify the editable task sub-items through a pre-constructed editable recognition model;

[0095] Step C2, retrieve the editing data sub-item corresponding to the editable task sub-item from the pre-constructed association replacement library, associate the editing data sub-item with the editable task sub-item to establish an editing association network, wherein the editing association network has a task sub-item as a center node and an editing data sub-item as a branch node, and the connection between the branch node and the center node is configured with a dynamic editing factor;

[0096] Step C3, calculate the dynamic editing factor through a pre-set dynamic editing algorithm, and return to step C2 until the traversal of all editable task sub-items is completed, and the editing association network of the current micro-service task is configured in the pre-set task generation model. The task learning unit 420 constructs the editing association network through the task learning sub-strategy. In step C1, the editable recognition model adopts a CNN-BiLSTM architecture, the input is a task text feature vector, and the output is an editable probability matrix. In step C2, the editing association network has a task sub-item as a center node and an editing data sub-item as a branch node, and the dynamic editing factor is calculated through an attention mechanism:

[0097]

[0098] Wherein, n i is the i j th task sub-item vector, d T is the j 验证 th editing data sub-item vector, m is the total number of branch nodes, and the attention function is: Attention(a,b)=a j ·W·b, W is an attention weight matrix with a dimension of d×d. Taking the user registration task as an example, it is decomposed into three task sub-items: "mobile phone number verification", "password encryption" and "user information storage", and the corresponding editing data sub-items (such as "international number verification" and "virtual number segment detection") are retrieved from the association replacement library. Assuming that W is a unit matrix, the dynamic editing factor of the "mobile phone number verification" sub-item and the "international number verification" data sub-item is calculated as follows: n 验证 =[0.8,0.2,0.1]; d 国际 =[0.7,0.3,0.0], Attention=0.8×0.7+0.2×0.3+0.1×0.0=0.56+0.06=0.62; If the attention values of other data sub-items are 0.38 and 0.25 respectively, then γ ij =e 0.62 / (e 0.62 +e 0.38 +e 0.25 )≈1.859 / (1.859+1.462+1.284)≈0.42.

[0099] The task virtual module 400 further comprises a task generation unit 430, which is configured with a task generation sub-strategy, the task generation sub-strategy comprising:

[0100] Step D1, determine the task type of the current microservice task, and call the corresponding editing association network from the task generation model according to the task type;

[0101] Step D2, calculate the replacement weight corresponding to each dynamic editing factor according to the network environment data through a preset replacement weight algorithm;

[0102] Step D3, generate a replacement ant at the center node, and determine the replacement ant running path according to the replacement weight until the replacement ant stays at a certain branch node, the replacement ant generates repulsive pheromone, and the repulsive pheromone is used to reduce the replacement weight corresponding to the branch node on the replacement ant path;

[0103] Step D4, calculate the comprehensive reliable vector corresponding to the branch node where the replacement ant is located through a preset vector evaluation algorithm, return to step D2 until the comprehensive reliable vector meets the preset vector constraint, the vector constraint is generated according to the network environment information at the current moment; the comprehensive reliable vector includes a security component, a uniqueness component and a verification component, the security component is used to evaluate the content security of the current virtual task, the uniqueness component is used to evaluate the content uniqueness of the current virtual task, and the verification component is used to evaluate the verifiable nature of the current virtual task.

[0104] Step D5, configure the microservice task benchmark with the task type, and replace the editing data subitem corresponding to the branch node to the corresponding microservice task benchmark to generate the virtual task. The task generation unit 430 dynamically generates a virtual task through a task generation sub-strategy, and the replacement weight algorithm formula in step D2 is: ij (t)=γ ij ·exp(-λ·freq ij )·σ(env(t))wherein, w ij (t) is the replacement weight from the i th center node to the j th branch node at t moment, γ ij is a dynamic editing factor, λ is a forgetting factor, and takes a value of 0.5, freq ij is a historical replacement frequency, and σ(env(t)) is a network environment fitness function, env(t) is a normalized vector of environment indexes such as network delay and bandwidth. The comprehensive reliable vector calculation in step D4 is: wherein, s sec is a security component, F sec is a security feature vector; s uni is a uniqueness component, calculated by a hash similarity; s val is a verification component, and VerAcc(d j ) is the verification accuracy of the editing data subitem.

[0105] In the electric pin promotion scenario, the order creation task type is "high concurrency rush purchase", the corresponding editing associated network is called, the current network environment index is delay 50 ms (normalized 0.6), bandwidth 80% (normalized 0.8), and σ(env(t)) = 0.6x0.4+0.8x0.6=0.72. The historical replacement frequency freq ij of a certain center node "inventory deduction" is 3, the dynamic editing factor γ ij is 0.35, the replacement weight w ij is 0.35x e -0.5×3 x0.72≈0.35x0.223x0.72≈0.056. After generating the replacement ant, the comprehensive reliable vector of a certain branch node "distributed inventory lock" is calculated: the safety feature vector s sec =0.85 after Softmax, the hash similarity s uni =0.92, the verification accuracy s val =0.95, and the comprehensive reliable vector The vector constraint (each component is ≥0.8) is met, the editing data sub-item is replaced to the task benchmark, and a virtual rush purchase task containing a distributed inventory lock mechanism is generated. The task virtual module realizes intelligent generation of virtual tasks through task evaluation screening, editing associated network construction and replacement ant dynamic generation mechanism, provides test tasks covering high concurrency, abnormal input and other scenarios for microservice components, solves the technical problems of lack of dynamic adaptability and insufficient scene coverage in traditional virtual task generation, and ensures the stability and reliability of microservices in real business scenarios.

[0106] The microservice optimization module 500 is configured with a service optimization strategy, and the service optimization strategy includes:

[0107] Step A1, obtaining microservice load data and microservice output data to generate optimization target information; step A1 includes:

[0108] Step A1-1, configured with a preset revenue evaluation algorithm to calculate the revenue value of each microservice component;

[0109] Step A1-2, configured with a demand prediction model to generate a theoretical demand task at the next moment;

[0110] Step A1-3, comparing the theoretical demand task with the currently deployed microservice components to generate a missing data set;

[0111] Step A1-4, determine the microservice components to be optimized according to the missing data set and the benefit value to generate the optimization target information. Step A1 generates optimization target information through benefit evaluation and demand prediction, which is implemented as follows: Benefit evaluation algorithm: the component benefit value is calculated by using a multi-index weighted method, and the formula is:

[0112]

[0113] wherein, R i is the benefit value of the i-th microservice component, p is the number of evaluation indexes, β k is the weight of the k-th index (∑β k = 1), o ik is the measured value of the k-th index of the i-th component, and are the historical extreme values of the index. Taking the order microservice as an example, the indexes include throughput, weight 0.4, response time, weight 0.3, error rate, weight 0.3, measured values are 200 requests / s, 150 ms, 0.2%, and historical extreme values are [100, 300] requests / s, [50, 200] ms, [0, 1]%, then: R i = 0.4 x 0.5 + 0.3 x 0.333 + 0.3 x 0.8 = 0.2 + 0.0999 + 0.24 = 0.5399. The demand prediction model uses an LSTM neural network to predict the theoretical demand task at the next time, and the model input is the historical task sequence (such as the task amount every 10 minutes in the past 24 hours), and the output is the theoretical demand task number in the next 10 minutes. The model structure includes 2 layers of LSTM (128 neurons each), 1 layer of full connection layer, the loss function is mean square error, and the optimizer is Adam. Taking an e-commerce platform as an example, the prediction result of the model after training during the double 11 promotion period is: the theoretical demand task number at the next time is 12000 times / s, the total processing capacity of the currently deployed order microservice is 8000 times / s, and the missing data set is 4000 times / s. The component to be optimized is determined: the optimization target is determined by comparing the missing data set and the benefit value. If the component load rate exceeds the threshold (such as 70%) and the benefit value is lower than the average level, it is included in the optimization list. Assuming that the current load rate of the order microservice is 85% (exceeding the threshold), and the benefit value is 0.54, which is lower than the average 0.6, it is determined as the component to be optimized, and the optimization target information is generated: reduce the load to below 70%, and improve the benefit value to above 0.6.

[0114] Step A2, process the optimization target information according to the currently deployed microservice component to generate optimization clue information; step A2 generates specific optimization clues based on the optimization target information of the currently deployed component, and the implementation is as follows: load gap calculation: calculate the load gap according to the theoretical demand and the current component capacity, and the formula is: wherein, GL Load gap, L req Theoretical demand load (normalized value), L i Unit load processing capacity of the ith component, c i The number of component instances. For example, the theoretical demand load L req = 1.2, and the current 3 order microservice instances, each instance L i = 0.3, then: G L = 1.2 - 3 x 0.3 = 1.2 - 0.9 = 0.3 Output efficiency analysis: calculate the component output gap Where O req is the theoretical demand output, O i is the unit output efficiency of the component. If the order microservice theoretical demand output O req = 1.0, and the current total output of the instances is 0.8, then G O = 0.2. Optimization clue generation: combine the load and output gaps to generate specific optimization directions. For example, the load gap 0.3 can be solved by increasing 1 instance (0.3 / 0.3 = 1), and the output gap 0.2 can be solved by replacing with a more efficient component (such as unit output O i = 0.35), forming the optimization clue set: [expand 1 instance, replace with a high-efficiency component].

[0115] Step A3, compare the optimization clue information with the microservice management information to generate the corresponding optimization scheme. The step A3 includes:

[0116] Step A3-1, determine the microservice component to be optimized and the optimization task target according to the optimization clue information;

[0117] Step A3-2, generate combinations of optimization sub-schemes of the microservice component according to the microservice management information to generate several proposed optimization schemes;

[0118] Step A3-3, configure an optimization evaluation algorithm to evaluate the comprehensive optimization value of each proposed optimization scheme, and determine the proposed optimization scheme with the highest comprehensive optimization value as the output optimization scheme. Step A3 generates the optimal scheme by comparing the microservice management information with the optimization clues, as follows: optimization sub-scheme combination: generate candidate schemes according to the component association and replacement characteristics in the microservice management information. For example, for order microservice optimization, the sub-schemes include: scheme 1: expand 2 original component instances scheme 2: replace 1 original component with a high-efficiency component (load processing capacity 0.35, revenue value 0.65)

[0119] Scheme 3: expand 1 original component and replace 1 with a high-efficiency component Comprehensive evaluation algorithm: use the analytic hierarchy process to calculate the scheme comprehensive optimization value, the formula is:

[0120]

[0121] wherein S j is the comprehensive optimization value of the jth scheme, q is the evaluation dimension (such as cost, efficiency, reliability), ω l is the dimension weight, v jl is the index value of the scheme j in the dimension l, respectively the maximum index value and the minimum index value in the scheme group. The micro-service optimization module realizes intelligent optimization of the micro-service components through the generation of the revenue-demand two-dimensional target, the load-output gap analysis and the multi-scheme comprehensive evaluation, solves the problems of lack of data driving and rough decision-making of the traditional optimization scheme, improves the system resource utilization rate and the business adaptability, and ensures efficient operation of the micro-service architecture in complex scenarios.

[0122] Of course, the above is only a typical example of the present application, in addition to which the present application can have other various specific embodiments, and any technical solution formed by equivalent replacement or equivalent transformation falls within the scope of the present application.

Claims

1. A software development and deployment system based on a microservice architecture, characterized in that: It includes a microservice management module, a load monitoring module, an output evaluation module, a task virtualization module, and a microservice optimization module; The microservice management module is connected to the microservice component library, which stores several different types of microservice components. The microservice management module is used to generate microservice management information based on deployment feedback information. The load monitoring module is used to monitor the load on computing resources of each deployed microservice component to generate microservice load data. The output evaluation module is used to evaluate the output results of each deployed microservice component to generate microservice output data. The task virtual module is configured with several task triggering conditions. When the task triggering conditions are met, a virtual task is generated and sent to the corresponding microservice component. The microservice optimization module is configured with service optimization strategies, which include: Step A1: Obtain microservice load data and microservice output data to generate optimization target information; Step A2: Process the target information based on the currently deployed microservice components to generate optimization clue information; Step A3: Compare the optimization clue information with the microservice management information to generate the corresponding optimization plan.

2. The software development and deployment system based on a microservice architecture as described in claim 1, characterized in that: The microservice management module includes a load management unit, a value management unit, and an association management unit. The load management unit is configured with a preset load management algorithm to calculate the static load data of the corresponding microservice components. The value management unit is configured with a preset value management algorithm to calculate the static output data of the corresponding microservice components. The association management unit is configured with association sub-policies, which include... Step B1: Analyze the service characteristics of each microservice component to obtain the service characteristic group; Step B2: Generate service feature connections based on the relationships between microservice components, wherein the relationships are generated based on the comparison results between service feature groups; Step B3: Configure preset alternative constraints, configure the service feature connection type that meets the alternative constraints as an alternative feature connection, and obtain static load data and static output data according to the correlation to generate the corresponding alternative features. Step B4: Generate the microservice management information based on the substitution features.

3. The software development and deployment system based on a microservice architecture as described in claim 1, characterized in that: The load monitoring module is configured with a load calculation unit and a measured load correction unit. The load calculation unit is configured with a preset load calculation algorithm to generate a measured load value based on the running information of the microservice components. The measured load correction unit is configured with load correction conditions and a load correction algorithm. When the measured load value meets the load correction conditions, the load correction algorithm is executed to generate a load correction factor to the microservice component library. The output evaluation module is configured with an output calculation unit and a measured output correction unit. The output calculation unit is configured with a preset output calculation algorithm to generate measured output values ​​based on the running information of the microservice components. The measured output correction unit is configured with output correction conditions and an output correction algorithm. When the measured output value meets the output correction conditions, the output correction algorithm is executed to generate an output correction factor to the microservice component library.

4. A software development and deployment system based on a microservice architecture as described in claim 3, characterized in that: The load monitoring module also includes a mimic load unit, which is configured with a load parsing sub-policy. The load parsing sub-policy is used to generate a load correction factor based on the running information of the microservice component when it executes virtual tasks. The output evaluation module is configured with a mimicry output unit. The microservice component processes virtual tasks to generate virtual outputs. The mimicry output unit is configured with an output parsing sub-strategy. The output parsing sub-strategy is used to generate the virtual correction factor based on the virtual output.

5. A software development and deployment system based on a microservice architecture as described in claim 1, characterized in that: The task virtual module includes a task filtering unit and a task learning unit. The task filtering unit is configured with a task evaluation algorithm, which evaluates microservice tasks and generates task evaluation vectors. The task filtering unit is configured with filtering constraints. When a task evaluation vector satisfies the filtering constraints, the corresponding microservice task is output to the task learning unit. The task learning unit is configured with a task learning sub-strategy, which includes: Step C1: Decompose the current microservice task to obtain several task sub-items, and identify the editable task sub-items through a pre-built editable recognition model; Step C2: Retrieve the editable task sub-item from the pre-built association replacement library, and associate the editable task sub-item with the editable task sub-item to establish an editing association network. The editing association network has the task sub-item as the central node and the editable data sub-item as the branch node. The connection between the branch node and the central node is configured with a dynamic editing factor. Step C3: Calculate the dynamic editing factor using a preset dynamic editing algorithm, and return to step C2 until all editable task sub-items have been traversed. Configure the editing association network of the current microservice task in the preset task generation model.

6. A software development and deployment system based on a microservice architecture as described in claim 5, characterized in that: The task virtual module further includes a task generation unit, which is configured with a task generation sub-strategy, the task generation sub-strategy including: Step D1: Determine the task type of the current microservice task, and retrieve the corresponding edit association network from the task generation model according to the task type; Step D2: Calculate the replacement weight corresponding to each dynamic editing factor based on the network environment data using a preset replacement weight algorithm; Step D3: Generate replacement ants at the central node and determine the running path of the replacement ants according to the replacement weight until the replacement ants stop at a certain branch node. The replacement ants generate repulsion pheromones, which are used to reduce the replacement weight of the branch nodes that the replacement ants pass through. Step D4: Calculate the comprehensive reliability vector corresponding to the branch node where the replacement ant is located using a preset vector evaluation algorithm, and return to step D2 until the comprehensive reliability vector satisfies the preset vector constraints, which are generated based on the network environment information at the current moment. Step D5: Configure the microservice task baseline for this task type, and replace the edit data sub-items corresponding to the branch nodes with the corresponding microservice task baseline to generate the virtual task.

7. A software development and deployment system based on a microservice architecture as described in claim 6, characterized in that: The comprehensive reliability vector includes a security component, a uniqueness component, and a verifiability component. The security component is used to evaluate the content security of the current virtual task, the uniqueness component is used to evaluate the content uniqueness of the current virtual task, and the verifiability component is used to evaluate the verifiability of the current virtual task.

8. A software development and deployment system based on a microservice architecture as described in claim 1, characterized in that: Step A1 includes: Step A1-1: Configure a preset revenue evaluation algorithm to calculate the revenue value of each microservice component; Step A1-2: Configure the demand forecasting model to generate the theoretical demand task for the next time step; Steps A1-3: Compare the theoretical requirements with the currently deployed microservice components to generate missing data sets; Steps A1-4: Determine the microservice components to be optimized based on the missing data groups and the revenue values ​​to generate the optimization target information.

9. A software development and deployment system based on a microservice architecture as described in claim 1, characterized in that: Step A3 includes: Step A3-1: Determine the microservice components to be optimized and the optimization task objectives based on the optimization clue information; Step A3-2: Generate a combination of optimization sub-solutions for microservice components based on microservice management information to generate several proposed optimization schemes; Step A3-3: An optimization evaluation algorithm is configured to evaluate the comprehensive optimization value of each proposed optimization scheme, and the proposed optimization scheme with the highest comprehensive optimization value is determined as the output optimization scheme.

Images