Dynamic decision-making method and device for service routing, medium and product

Abstract

The invention discloses a dynamic decision-making method and device for service routing, a medium and a product, and the method comprises the steps: collecting multiple pieces of node information of each node in a distributed cluster in real time, and carrying out the real-time updating according to the node information to obtain the current cluster information; in response to a dynamic routing decision request sent by the target node for the target service request, extracting current target node information of the target node and a receiver node corresponding to the target service request from the dynamic routing decision request; and jointly inputting the current cluster information, the current target node information and the receiver node into a pre-trained routing decision model to obtain routing decision information, and feeding back the obtained routing decision information to the target node, so that the target node sends the target service request to a next node in the distributed cluster according to the routing decision information, according to the technical scheme of the embodiment of the invention, through a dynamic intelligent decision-making mechanism, the real-time performance and reliability of service routing are remarkably improved, and the influence of network jitter and node faults is effectively reduced.

Description

Technical Field

[0001] The present invention relates to the field of communication technology, and in particular to a dynamic decision method, device, medium and product for service routing. Background Art

[0002] With the popularization of microservice architecture and the continuous expansion of distributed systems, dynamic optimization and intelligent decision-making of service routing have become particularly important.

[0003] Existing methods for optimizing service calls include static routing strategies, load balancers, and service meshes. Static routing forwards traffic based on preconfigured routing tables, while load balancers distribute requests using algorithms like round-robin and minimum connections. Service meshes implement fine-grained traffic control using sidecar proxies.

[0004] In the process of implementing the present invention, the inventors found that although static strategies are simple, they lack flexibility in dynamic environments; although load balancing can dynamically allocate requests, it often ignores the dependencies between services; and service grids are complex to configure and have high overhead. Summary of the Invention

[0005] The embodiments of the present invention provide a dynamic decision-making method, device, medium and product for service routing, which can significantly improve the response speed and reliability of service calls in a distributed system.

[0006] According to one aspect of an embodiment of the present invention, a dynamic decision method for service routing is provided, which is executed by a microservice deployed on a routing decision node in a distributed cluster. The method includes:

[0007] Collect multiple node information of each node in the distributed cluster in real time, and update the current cluster information in real time based on the node information; wherein the cluster information includes the current network topology information of the distributed cluster and the current node information of each topology node in the current network topology information;

[0008] In response to a dynamic routing decision request sent by a target node for a target service request, extracting current target node information of the target node and a receiver node corresponding to the target service request from the dynamic routing decision request;

[0009] The current cluster information, current target node information and receiving node are input into the pre-trained routing decision model, and the routing decision information is fed back to the target node so that the target node can send the target service request to the next node in the distributed cluster according to the routing decision information.

[0010] According to another aspect of an embodiment of the present invention, a dynamic decision-making device for service routing is provided, which is configured in a microservice deployed on a routing decision node in a distributed cluster, and includes:

[0011] An information collection module is used to collect multiple node information of each node in the distributed cluster in real time, and update the current cluster information in real time based on the node information; wherein the cluster information includes the current network topology information of the distributed cluster and the current node information of each topology node in the current network topology information;

[0012] a request processing module, configured to respond to a dynamic routing decision request sent by a target node for a target service request, and extract the current target node information of the target node and a receiver node corresponding to the target service request from the dynamic routing decision request;

[0013] The intelligent decision-making module is used to input the current cluster information, the current target node information and the receiving node into the pre-trained routing decision model, and feed back the routing decision information to the target node so that the target node can send the target service request to the next node in the distributed cluster according to the routing decision information.

[0014] According to another aspect of an embodiment of the present invention, an electronic device is provided, the electronic device comprising:

[0015] at least one processor; and

[0016] a memory communicatively connected to the at least one processor; wherein,

[0017] The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the dynamic decision method for service routing described in any embodiment of the present invention.

[0018] According to another aspect of an embodiment of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement a dynamic decision method for service routing according to any embodiment of the present invention when executed.

[0019] According to another aspect of an embodiment of the present invention, a computer program product is provided, comprising computer instructions, which implement the steps of the method according to any embodiment of the present invention when executed by a processor.

[0020] The technical solution of the embodiment of the present invention continuously updates the network topology structure and status information of each node in the current cluster by collecting the performance indicators and network status data of each node in the distributed cluster in real time. When the target node initiates a routing decision request for a specific service request, it can automatically extract the real-time status data and recipient node information of the node, and then input the cluster topology, target node status and recipient node information into a pre-trained intelligent routing decision model to generate the optimal routing solution and return it to the target node, guiding it to forward the service request to the most appropriate next-hop node in the cluster. This new dynamic decision-making method for service routing can significantly improve the real-time and accuracy of routing decisions, effectively reduce latency and resource consumption, and enhance cluster throughput and operational stability through an adaptive load change mechanism.

[0021] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present invention, nor is it intended to limit the scope of the present invention. Other features of the present invention will become readily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0023] Figure 1 This is a flow chart of a dynamic decision method for service routing provided according to the first embodiment of the present invention;

[0024] Figure 2 is a flow chart of another dynamic decision method for service routing provided according to embodiment 2 of the present invention;

[0025] Figure 3 is a flowchart of another dynamic decision method for service routing provided according to embodiment three of the present invention;

[0026] Figure 4 1 is a schematic structural diagram of a dynamic decision-making device for service routing according to a fourth embodiment of the present invention;

[0027] Figure 5 The present invention is a schematic structural diagram of an electronic device for implementing a dynamic decision method for service routing according to an embodiment of the present invention. DETAILED DESCRIPTION

[0028] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0029] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0030] Example 1

[0031] Figure 1 This is a flowchart of a dynamic decision-making method for service routing provided in Example 1 of the present invention. This embodiment is applicable to cross-node service call scenarios under a microservice architecture. The method can be executed by a dynamic decision-making device for service routing, which can be implemented in the form of hardware and / or software and can generally be configured in microservices deployed on routing decision nodes in a distributed cluster.

[0032] Microservices can be understood as independently deployed, standardized business components, each implementing specific business functions and communicating through well-defined interfaces. These service instances are deployed in a containerized manner on cluster nodes, with their operating environments preconfigured and isolated to ensure compatibility with the host environment and other service instances.

[0033] Correspondingly, such as Figure 1 As shown, the method includes:

[0034] S110. Collect multiple items of node information of each node in the distributed cluster in real time, and obtain current cluster information based on the node information in real time; wherein the cluster information includes the current network topology information of the distributed cluster and the current node information of each topology node in the current network topology information.

[0035] Among them, a distributed cluster can be understood as a service deployment environment composed of multiple interconnected service nodes (hereinafter referred to as nodes). These nodes may be distributed in different physical locations or data centers, but they work together through network communication to complete the service request processing task. Each node can serve as both a service provider and a routing transit node. Multiple node information can be understood as a variety of operating status data obtained in real time from each node in the cluster, including hardware resources (such as computing core utilization, memory occupancy, and disk I / O throughput), service performance (such as call latency between microservices and interface response time), and network status (such as inter-node communication latency, transport layer retransmission probability, and packet loss rate).

[0036] Network topology information can be understood as structured data about the connectivity and operational status of nodes in a distributed cluster. This includes physical network connectivity, node location attributes, and service dependencies. This information dynamically reflects the cluster's network architecture and service interaction patterns, providing a context-aware foundation for routing decisions.

[0037] Optionally, each active node in the distributed cluster for receiving or forwarding service requests may periodically report node information to the routing decision node in the distributed cluster by sending heartbeat packets.

[0038] In this embodiment, a decentralized proxy deployment architecture is employed, with lightweight data agents independently running on each node. These agents are directly embedded in the service runtime environment (e.g., a container instance or container group), collecting real-time performance metrics such as the node's core count, memory, network latency, and inter-node communication status as node information. The agents support dynamic adjustment of data sampling frequency during runtime and implement custom trigger conditions through a built-in rules engine, completely eliminating reliance on centralized gateways or third-party monitoring components.

[0039] Specifically, each node sends collected node information to a microservice deployed on the decision node. The microservice aggregates the information reported by all nodes to provide real-time insights into the detailed operational status of every node in the entire cluster. Based on this comprehensive and accurate status information, a complete global view can be constructed, providing a reliable basis for subsequent intelligent decision-making.

[0040] S120 , in response to the dynamic routing decision request sent by the target node for the target service request, extract the current target node information of the target node and the receiver node corresponding to the target service request from the dynamic routing decision request.

[0041] Among them, the target node can be a node in the distributed cluster that directly receives the target service request sent by the user, or it can be a relay node in the transmission path of the target service request. The current target node information can be understood as the operating status information of the target node. The receiving node specifically refers to the destination node to which the target service request needs to be sent. In this embodiment, when processing the request message of the dynamic routing decision request of the target node, the request message will first be structured and parsed, and the key data fields will be identified through the predefined message format. The parsing process adopts a streaming processing method, parsing while receiving to ensure low latency. The target node information extracted from the request message reflects the current real-time operating status of the node. In a specific example, the operating status snapshot of the target node can include basic computing core load and memory usage, and can also cover fine-grained indicators such as network throughput, current number of concurrent connections, and service response delay.

[0042] Furthermore, by analyzing the metadata and service identifier fields in the dynamic routing decision request message, the complete address information and service group of the receiving node can be accurately identified. The receiving node identifier is then verified against the cluster's pre-registered service directory to ensure the legitimacy and reachability of the target direction.

[0043] S130, input the current cluster information, current target node information and receiving node into the pre-trained routing decision model, obtain routing decision information and feed it back to the target node, so that the target node can send the target service request to the next node in the distributed cluster according to the routing decision information.

[0044] In this embodiment, the routing decision model can be understood as an intelligent routing path selector. Current cluster information, current target node information, and receiving node information are input into the model through the input layer. After data fusion in the model's feature fusion layer, a feature vector is generated. This feature vector is then input into the model's collaboration layer. The collaboration layer contains three types of decision sub-models, each with its own applicable operating environment. These sub-models process the feature vectors in parallel and generate recommended path solutions. These results are then used by a dynamic arbitrator to make a final decision on the optimal path based on real-time system load. Finally, the target node sends the target service request to the next node in the distributed cluster according to the optimal path provided by the routing decision model.

[0045] The technical solution of the embodiment of the present invention continuously updates the network topology structure and status information of each node in the current cluster by collecting the performance indicators and network status data of each node in the distributed cluster in real time. When the target node initiates a routing decision request for a specific service request, it can automatically extract the real-time status data and recipient node information of the node, and then input the cluster topology, target node status and recipient node information into a pre-trained intelligent routing decision model to generate the optimal routing solution and return it to the target node, guiding it to forward the service request to the most appropriate next-hop node in the cluster. This new dynamic decision-making method for service routing can significantly improve the real-time and accuracy of routing decisions, effectively reduce latency and resource consumption, and enhance cluster throughput and operational stability through an adaptive load change mechanism.

[0046] Example 2

[0047] Figure 2 This is a flowchart of another dynamic decision-making method for service routing provided in the second embodiment of the present invention. This embodiment is optimized based on the above embodiments. Under the condition that the routing decision model includes an input layer, a feature fusion layer, a model collaboration layer, a dynamic arbitrator and an output layer, this embodiment refines the operation of "inputting the current cluster information, the current target node information and the receiving node into the pre-trained routing decision model to obtain routing decision information fed back to the target node" in the above embodiment. Specifically, it can be: inputting the current cluster information, the current target node information and the service provider node into the pre-trained routing decision model; processing the current cluster information, the current target node information and the service provider node through the input layer to obtain time series data features, cluster topology data features and model features. The method adopts the spatiotemporal feature root cross encoder to jointly encode the time series data features and the cluster topology data features through the feature fusion layer to generate a fusion feature vector; the multiple decision sub-models in the model collaboration layer generate a set of routing decision information according to the input fusion feature vector, wherein each set of routing decision information contains multiple next nodes and the model weights matching each next node; the dynamic arbitration layer generates the inter-model weights corresponding to each decision sub-model according to the model weight arbitration feature; the output layer generates and outputs the routing decision information according to multiple sets of routing decision information and the weights between models.

[0048] Correspondingly, such as Figure 2 As shown, the method includes:

[0049] S210. Collect multiple items of node information of each node in the distributed cluster in real time, and obtain current cluster information based on the node information in real time; wherein the cluster information includes the current network topology information of the distributed cluster and the current node information of each topology node in the current network topology information.

[0050] S220 , in response to the dynamic routing decision request sent by the target node for the target service request, extract the current target node information of the target node and the receiver node corresponding to the target service request from the dynamic routing decision request.

[0051] S230: Input the current cluster information, the current target node information, and the service provider node into a pre-trained routing decision model.

[0052] S240 , processing the current cluster information, the current target node information, and the service provider node through the input layer to obtain time series data features, cluster topology data features, and model weight arbitration features.

[0053] Whenever multiple items of node information of each node in the distributed cluster are collected, the above-mentioned items of node information can be input into the routing decision model to assist the routing decision model in making real-time routing decisions.

[0054] Accordingly, the input layer can organize the time series data features of the target node by combining the current target node information and the historical node information of the target node at multiple historical time points. For example, the time series data sequence may include information such as the target node's memory usage, column operation core occupancy rate, and request queue length at multiple consecutive time points. Furthermore, the input layer can also organize the time series data features of other nodes in the distributed cluster based on the historical node information of other nodes at multiple historical time points.

[0055] Cluster topology data features can be understood as the topological features of the network topology composed of all currently available nodes in the distributed cluster. For example, it can be a service dependency graph that includes the connection relationship between nodes and the location information of each node in physical space.

[0056] In this embodiment, the routing decision for the target service request is determined by using multiple decision sub-models. Accordingly, the model weight arbitration feature can be understood as a feature used to arbitrate the model weights of each decision sub-model. This model weight arbitration feature can be dynamically determined based on real-time load information of the distributed cluster, such as the fluctuation rate of the computing cores of each node and the request queue length.

[0057] Specifically, in the process of processing the data received by the input layer, three parallel processing channels are used for feature extraction: the first channel uses a long short-term memory network to perform sliding window analysis and feature encoding on the time series data of the node's current information and historical information, and extracts time series data features with time characteristics; the second channel uses a graph convolutional neural network to perform adjacency matrix operations and node embedding on the cluster topology data and microservice dependencies, and generates cluster topology data features that characterize network connection characteristics; the third channel uses an adaptive normalization algorithm to standardize and dimensionally transform the dynamic control parameters, and outputs quantifiable and comparable model weight arbitration features.

[0058] S250, using a spatiotemporal feature root cross encoder through a feature fusion layer to jointly encode the time series data features and the cluster topology data features to generate a fused feature vector.

[0059] In this embodiment, the special fusion layer receives the output results from the first channel and the second channel, first standardizes the time series data features and the cluster topology data features to eliminate the dimensional differences between different features, and then uses the cross-attention mechanism to establish the correlation between the features. The importance weights of different feature dimensions are calculated through the multi-head attention layer, and then the weighted feature vectors are spliced ​​and input into the fully connected network for nonlinear transformation and dimensionality reduction. The features finally generated retain both the time series data features and the cluster topology data features, providing a unified feature expression for the subsequent model collaboration layer.

[0060] S260. Generate a set of routing decision information based on the input fusion feature vector through multiple decision sub-models in the model collaboration layer, wherein each set of routing decision information includes multiple next nodes and model-internal weights that match each next node.

[0061] The model collaboration layer can be thought of as an intelligent framework for multi-model collaborative decision-making. Unlike traditional single-model decision-making, this layer simultaneously runs multiple decision-making sub-models optimized for different scenarios. Each sub-model independently analyzes the current service request based on its own characteristics and outputs corresponding routing recommendations. Intra-model weights can be thought of as confidence scores assigned by each decision-making sub-model to its recommended candidate nodes. These values ​​range from 0 to 1, with higher values ​​indicating a stronger recommendation for that node. These weights are calculated by each sub-model based on its specialized evaluation dimensions.

[0062] In this embodiment, during the model collaboration layer's operation, multiple decision sub-models simultaneously process the input fused feature vector. Each sub-model independently analyzes and generates its own routing decision recommendations. These recommendations are presented as a set of possible next node options, each with a weight that reflects the sub-model's confidence in the node's selection. Different sub-models, due to varying design characteristics and training objectives, may provide different node recommendations and weight assignments. In a specific example, when processing a routing request for a video conferencing service, the low-latency sub-model recommends Path A, Path B, and Path C based on network latency, assigning weights of 0.7, 0.2, and 0.1, respectively. The bandwidth sub-model recommends Path B, Path C, and Path D based on available bandwidth, assigning weights of 0.6, 0.3, and 0.1, respectively. The long-distance model sub-model recommends Path A, Path B, and Path D based on long-distance service stability, assigning weights of 0.6, 0.2, and 0.2, respectively. This diverse decision information is retained, providing a multi-faceted reference for subsequent comprehensive decision-making.

[0063] S270 , generating inter-model weights corresponding to each decision sub-model respectively according to the model weight arbitration characteristics through the dynamic arbitration layer.

[0064] The dynamic arbitration layer can be understood as an intelligent decision coordinator. It receives different routing solutions from multiple decision sub-models in real time and dynamically evaluates the applicability of each solution by analyzing current network status, service requirements, and historical performance data. This layer uses an adaptive algorithm to comprehensively consider real-time indicators such as latency, bandwidth, and load, as well as the confidence weights of each sub-model, to calculate the optimal routing path within milliseconds. Inter-model weights can be understood as confidence coefficients assigned to each decision sub-model. They are used to balance the decision-making influence of different specialized models. These weights are adjusted based on real-time scenario data to achieve the optimal choice.

[0065] In this embodiment, the routing decision information output by each decision sub-model and its corresponding weight within the model are received. Then, based on the preset evaluation indicators, the confidence score of each sub-model in the current network state is calculated. The score is then converted into an inter-model weight through normalization to ensure that the relative importance of each sub-model is reasonably reflected. In a specific example, the dynamic arbitration layer will first calculate the confidence score (i.e., weight) of each sub-model: the low-latency model (the current network jitter is obvious and the prediction accuracy is high) is 0.45, the bandwidth model (stable performance during peak load) is 0.35, and the long-distance model (lack of regional advantage) is 0.20. The final path score is obtained by weighted calculation, and the one with the highest score is the optimal path.

[0066] S280 , generating and outputting routing decision information through the output layer based on multiple sets of routing decision information and weights between models.

[0067] In this embodiment, the output layer first receives multiple sets of routing decision information and their corresponding inter-model weights from the dynamic arbitration layer. It then merges and removes duplicate candidate nodes from each set of decision information, retaining a unique set of nodes. The scores of identical nodes are then weighted and summed based on the inter-model weights to calculate a comprehensive score for each node. All nodes are then sorted in descending order by comprehensive score, and the candidate nodes with the highest scores are selected to generate decision information.

[0068] S290: Feedback the routing decision information to the target node, so that the target node can send the target service request to the next node in the distributed cluster according to the routing decision information.

[0069] Specifically, after receiving the routing decision information, the target node can extract the multiple next nodes that can serve as the final routing node, which are jointly determined by multiple decision sub-models and contained in the routing decision information, as well as the weight values ​​corresponding to each next node. Furthermore, the next node with the highest weight value can be selected, and the target service request can be transmitted to the next node. The next node will continue to execute the methods of the various embodiments of the present invention to determine a new next node until it is sent to the receiving node. This operation can dynamically determine each routing decision of the target service request, so that the target service request can always select the optimal service route that is most suitable for the current network environment for forwarding.

[0070] It is understandable that if a network abnormality is detected in the next node before the target node sends the target service request to the next node with the largest weight value, the target service request can be sent again to the next node with the second largest weight value obtained in the routing decision information.

[0071] The technical solution of the embodiment of the present invention obtains the cluster state, target node, and service provider node data of the target service request and inputs this information into a pre-trained routing decision model. The input layer processes the data to obtain time series features, cluster topology features, and model weight arbitration features. The feature fusion layer then generates a fused feature vector through spatiotemporal coding. Multiple decision sub-models in the model collaboration layer then generate routing decision information containing multiple next nodes and corresponding model weights based on the fused feature vector. The dynamic arbitration layer generates inter-model weights corresponding to each decision sub-model based on the model weight arbitration features. The output layer generates and outputs the final routing decision information based on multiple sets of routing decision information and inter-model weights. Finally, the target node sends the target service request to the next node in the distributed cluster based on this routing decision information, completing the routing forwarding process of the entire service request. This new dynamic decision-making method for service routing achieves adaptive routing selection through a multi-model collaborative decision-making mechanism and intelligent feature fusion. The dynamic arbitration layer evaluates the performance of each model in real time to ensure that the decision is both adapted to the current state and takes into account the global optimality, significantly reducing latency and improving throughput.

[0072] Optionally, based on the above embodiments, generating and outputting routing decision information through an output layer according to multiple sets of routing decision information and weights between models may include:

[0073] The output layer extracts each next node and its model weight from each set of routing decision information, and deduplicates the extracted results to obtain multiple candidate next nodes, as well as the decision sub-model and model weight corresponding to each candidate next node.

[0074] The output layer obtains the comprehensive weight of each candidate next node based on the inter-model weights of the decision sub-model of each candidate next node and the intra-model weight of each candidate next node;

[0075] Through the output layer, the target number of target next nodes are selected from each candidate next node in descending order of comprehensive weight;

[0076] The output layer generates and outputs routing decision information based on the target next node and the comprehensive weight corresponding to the target next node.

[0077] Generally speaking, when processing routing decision information, the output layer first performs a structured extraction of the recommendations provided by each decision sub-model. Specifically, each set of decision information is parsed to extract the recommended next node and its corresponding in-model weight. These weights reflect the strength of the individual model's recommendation for that node. During this process, since different sub-models may recommend the same node, the output layer performs a deduplication operation, merging duplicate nodes while retaining the information about which sub-models recommended each node and its original weight, thereby forming a non-repeating set of candidate nodes.

[0078] Generally speaking, after obtaining a set of candidate nodes, the output layer initiates a weighted fusion calculation. This not only considers each node's internal weight within each sub-model, but also incorporates the inter-model weights determined by the dynamic arbitration layer, which represent the relative reliability of different sub-models in the current decision-making scenario. By combining these intra-model weights with the inter-model weights, the output layer generates a comprehensive score for each candidate node. This score reflects both the node's inherent recommendation strength and the authoritativeness of the recommendation model, resulting in a more comprehensive evaluation.

[0079] Generally speaking, after the comprehensive score calculation is completed, the output layer enters the decision-making stage, sorting all candidate nodes by comprehensive score. Then, based on a preset number of requirements, the highest-scoring nodes are selected as the final recommended target nodes. When outputting the final routing decision, not only are these target nodes listed, but their standardized comprehensive weights are also included. This information helps the execution layer better understand the relative strengths and weaknesses of each recommended node, providing a more comprehensive basis for subsequent routing execution.

[0080] Optionally, based on the above embodiments, the extraction results are deduplicated through the output layer to obtain multiple candidate next nodes, and the decision sub-models and model weights corresponding to each candidate next node;

[0081] If it is determined that the extracted current next node belongs to multiple sets of current routing decision information at the same time, then obtaining multiple current model weights of the current next node in each set of current routing decision information;

[0082] The current next node is taken as an alternative next node, and after the maximum current model internal weight is determined as the model internal weight of the current next node, the decision sub-model to which the maximum current model internal weight belongs is determined as the decision sub-model corresponding to the current next node.

[0083] Generally speaking, when processing routing decision information, the output layer first removes duplicate nodes and merges them. If a node is recommended by multiple decision sub-models simultaneously, the node's multiple weights in different sub-models are recorded to ensure that no recommendation information is lost.

[0084] Specifically, for nodes recommended by multiple sub-models, the output layer adopts a "maximum value" strategy. It compares the node's weights across the sub-models, selects the largest weight as the node's representative weight, and records the decision sub-model information corresponding to this largest weight. This approach ensures that each candidate node has a representative weight and a corresponding recommended sub-model. This simplifies subsequent computational complexity while retaining the strongest basis for recommendation, ensuring that the final decision reflects the strongest recommendations from each sub-model.

[0085] Optionally, based on the above embodiments, the decision sub-model includes a first decision sub-model, a second decision sub-model and a third decision sub-model, wherein:

[0086] The first decision sub-model is a decision tree model or a random forest model, the second decision sub-model is a gradient boosting machine model, and the third decision sub-model is a long short-term memory model or a sequence model based on the self-attention mechanism.

[0087] Generally speaking, the decision sub-model combines three different types of machine learning models to form a decision-making system with complementary advantages. The first decision sub-model uses tree-based algorithms such as decision trees or random forests. These algorithms can quickly generate initial routing strategies and are particularly well-suited for low-latency scenarios. These models have inherent rule interpretability and can clearly demonstrate the basis for prioritizing service dependency paths, providing intuitive decision analysis for operations personnel. When processing regular traffic, these models can complete routing decisions within milliseconds, ensuring timely service responses.

[0088] Typically, the second decision sub-model uses a gradient boosting machine, which specializes in handling unbalanced datasets through iterative optimization and excels in bursty traffic scenarios. This model effectively optimizes long-tail features, such as predicting the response time of abnormal nodes, and gradually corrects deviations from the initial routing strategy through residual fitting. This model can provide more refined routing adjustments when faced with anomalies in the service call chain.

[0089] Generally, the third decision sub-model uses a long-short-term memory model or a sequence model based on the self-attention mechanism. These models excel at capturing temporal dependencies in network states, such as the periodic characteristics of network jitter. Furthermore, these models can effectively model complex topological relationships across data centers and analyze the spatiotemporal correlations in service call paths. When handling complex routing scenarios that require comprehensive consideration of historical states and spatial topology, these models can provide a more comprehensive basis for decision making.

[0090] Example 3

[0091] Figure 3 This is a flowchart of another dynamic decision-making method for service routing, provided in Example 3 of the present invention. This example is optimized based on the above examples. This example specifically refines the operation of "real-time collection of multiple node information items from each node in a distributed cluster, and real-time updating of current cluster information based on the node information."

[0092] Correspondingly, such as Figure 3 As shown, the method includes:

[0093] S310: Using a processing strategy that matches the information type of the node information, perform outlier filtering on the node information of each node.

[0094] Among them, anomalies can be understood as abnormal data in node information that obviously deviates from the normal range, mainly including three types of situations: one is the sudden abnormality of hardware indicators (such as the computing core utilization rate suddenly soars to 100% and then quickly falls back), the second is abnormal fluctuations in network communication (such as the delay suddenly increases by more than 500 milliseconds), and the third is abnormal values ​​of service performance (such as the response time of an interface exceeds 3 times the standard deviation of the set threshold).

[0095] In this embodiment, outlier filtering is performed on each node's information using a processing strategy tailored to the node information type. Detection methods are selected based on the information type to identify anomalous data points that significantly deviate from the normal range. Filtering is performed on these identified outliers to ensure the quality of the node information used in subsequent processing. After outlier filtering is complete, the remaining data proceeds to the next processing step.

[0096] S320. Clean the node information after the outlier filtering process, and update the current cluster information in real time based on the cleaned data, wherein the cluster information includes the current network topology information of the distributed cluster and the current node information of each topology node in the current network topology information.

[0097] In this embodiment, the node information that has been filtered out of outliers is first standardized and cleaned, including operations such as data format unification, missing value filling, and unit conversion. The real-time status indicators of each node are then recalculated based on the cleaned data, and key parameters such as computing core utilization, memory usage, and network latency are updated. These updated node data are integrated into the cluster topology structure, and the network connection relationship and service call dependency graph between nodes are reconstructed. Finally, a cluster information snapshot containing the latest node status and complete topology relationship is generated, providing an accurate data foundation for subsequent processing links. The entire process adopts an incremental update method, and only the changed data is recalculated to ensure the timeliness of information updates.

[0098] S330 , in response to the dynamic routing decision request sent by the target node for the target service request, extract the current target node information of the target node and the receiver node corresponding to the target service request from the dynamic routing decision request.

[0099] S340, input the current cluster information, current target node information and receiving node into the pre-trained routing decision model, obtain routing decision information and feed it back to the target node, so that the target node can send the target service request to the next node in the distributed cluster according to the routing decision information.

[0100] The technical solution of the embodiment of the present invention first uses a strategy that matches the information type to detect and filter outliers on the operating data of each node, and then performs real-time calculation and update on the cleaned standardized data to generate current cluster information containing network topology and node status. When the target node initiates a routing request, the real-time status of the node and the receiver identification information are automatically extracted, and these data are input into the pre-trained routing decision model. After model analysis and processing, the optimal routing decision result is generated. The final target node accurately forwards the service request to the next hop node based on the returned routing information to complete the entire service call process. This new type of dynamic decision-making method for service routing significantly improves the quality and efficiency of data preprocessing. Through intelligent anomaly detection and real-time data cleaning mechanisms, it effectively eliminates noise interference and ensures that the cluster status information on which routing decisions rely is accurate and reliable. At the same time, the dynamically updated topology and node status information provide a real-time and complete environmental perception basis for subsequent routing analysis, enabling the entire decision-making process to quickly adapt to changes in the network environment and greatly improve the timeliness and accuracy of routing solutions.

[0101] Optionally, based on the above embodiments, a processing strategy that matches the information type of the node information is adopted to filter out abnormal points from the node information of each node, which may include at least one of the following:

[0102] For the first type of node information whose information type is service response time or network delay, after filtering out the outliers of the first type of node information sent by each node according to the preset first threshold condition, fill the historical window mean or remove the outlier data points in the first type of node information set continuously sent by each node, and use the box value plot method to identify and filter out the outliers in each first type of information set;

[0103] For the second type of node information whose information type is processor occupancy or memory occupancy, outliers are filtered out of the second type of information sent by each node according to a preset second threshold condition.

[0104] Generally speaking, when processing first-class node information such as service response time or network latency, preliminary screening is performed based on preset threshold conditions. When an abnormal data point exceeding the threshold is detected, two processing methods are adopted: for transient anomalies, the data is filled using the average value of the historical window; for persistent anomalies, the data point is directly eliminated. Boxplot analysis is then used to identify and filter out statistical outliers in the data set by calculating the interquartile range, ensuring the continuity and stability of this type of time series data.

[0105] Generally speaking, processing the second type of node information, such as processor utilization or memory utilization, is relatively straightforward. Anomaly detection is performed based on pre-set thresholds, which differ from those used for the first type of information. Any abnormal values ​​exceeding the threshold are immediately filtered out. These resource metrics often fluctuate significantly, so threshold settings take into account hardware characteristics and business load characteristics, employing relatively loose but more targeted filtering criteria.

[0106] Generally speaking, exception handling strategies for these two types of information are designed differently based on their data characteristics. Time series information (such as latency) prioritizes data continuity and trend preservation, thus employing mean padding and statistical outlier detection. Resource metrics (such as computing cores) focus more on transient anomalies and employ direct threshold filtering. This categorized processing approach maximizes the preservation of the characteristic value of each type of information while ensuring data quality.

[0107] Optionally, an intelligent failover and multi-path redundancy architecture monitors the node operation status through a continuous heartbeat detection mechanism. Among them, active health checks are performed regularly at preset time intervals (such as 10 seconds). When a node is detected to have timed out for three consecutive responses, it is determined to be unavailable, and an alternative routing path is automatically enabled. It should be noted that these alternative paths are also generated and output by the routing decision model in the aforementioned embodiment, but have not yet been selected as the optimal path under the current environmental assessment.

[0108] Furthermore, a progressive traffic scheduling strategy was designed for node recovery scenarios. Initially, 10% of the baseline traffic is allocated to verify stability, and then the load distribution is dynamically adjusted based on real-time performance indicators. Using an intelligent weight adjustment algorithm, traffic is gradually increased to normal levels, ensuring the stability of the restored node while avoiding impact on the entire cluster.

[0109] Optional, visual real-time monitoring interface, can display core operating indicators such as service response delay, node resource utilization, etc. in real time. The monitoring panel adopts a dynamic refresh mechanism, and the data update frequency can be configured to seconds or minutes, supporting multi-dimensional data perspective and trend analysis. Through the preset threshold alarm function, when an indicator abnormality is detected (such as the computing core continuously exceeds 80% or the delay exceeds 500 milliseconds), a graded alarm notification is automatically triggered, including interface highlight prompts, email push and instant message reminders. At the same time, it provides historical data backtracking function, which can query the operating status curve of any time period to assist operation and maintenance personnel to quickly locate the root cause of the problem.

[0110] Optionally, the model continuously optimizes the closed-loop mechanism, regularly analyzes the effectiveness of routing decisions and collects user feedback, and iteratively upgrades the machine learning model in a data-driven manner to ensure that the algorithm always maintains optimal performance.

[0111] Example 4

[0112] Figure 4 This is a structural diagram of a dynamic decision-making device for service routing provided by the fourth embodiment of the present invention. Figure 4 As shown, the device includes:

[0113] The information collection module 410 is used to collect multiple node information of each node in the distributed cluster in real time, and update the current cluster information in real time based on the node information; wherein the cluster information includes the current network topology information of the distributed cluster and the current node information of each topology node in the current network topology information;

[0114] The request processing module 420 is configured to respond to a dynamic routing decision request sent by a target node for a target service request, and extract the current target node information of the target node and a receiving node corresponding to the target service request from the dynamic routing decision request;

[0115] The intelligent decision module 430 is used to input the current cluster information, the current target node information and the receiving node into a pre-trained routing decision model, obtain the routing decision information and feed it back to the target node, so that the target node can send the target service request to the next node in the distributed cluster according to the routing decision information.

[0116] The technical solution of the embodiment of the present invention continuously updates the network topology structure and status information of each node in the current cluster by collecting the performance indicators and network status data of each node in the distributed cluster in real time. When the target node initiates a routing decision request for a specific service request, it can automatically extract the real-time status data and recipient node information of the node, and then input the cluster topology, target node status and recipient node information into a pre-trained intelligent routing decision model to generate the optimal routing solution and return it to the target node, guiding it to forward the service request to the most appropriate next-hop node in the cluster. This new dynamic decision-making method for service routing can significantly improve the real-time and accuracy of routing decisions, effectively reduce latency and resource consumption, and enhance cluster throughput and operational stability through an adaptive load change mechanism.

[0117] Based on the above embodiments, the routing decision model includes an input layer, a feature fusion layer, a model collaboration layer, a dynamic arbitrator, and an output layer;

[0118] Accordingly, the intelligent decision module 430 may further include:

[0119] A data input unit, used to input current cluster information, current target node information and service provider node into a pre-trained routing decision model;

[0120] A feature extraction unit is used to process the current cluster information, the current target node information, and the service provider node through the input layer to obtain time series data features, cluster topology data features, and model weight arbitration features;

[0121] A special fusion unit is used to jointly encode the time series data features and the cluster topology data features using a spatiotemporal feature root cross encoder through the feature fusion layer to generate a fused feature vector;

[0122] A dynamic weight arbitration unit is used to generate a set of routing decision information based on the input fusion feature vector through multiple decision sub-models in the model collaboration layer, where each set of routing decision information includes multiple next nodes and the model weights that match each next node;

[0123] A weight generation unit, configured to generate inter-model weights corresponding to each decision sub-model according to the model weight arbitration characteristics through a dynamic arbitration layer;

[0124] The decision generation unit is used to generate and output routing decision information through the output layer based on multiple sets of routing decision information and the weights between each model.

[0125] Based on the above embodiment, the decision generation unit is specifically configured to:

[0126] The output layer extracts each next node and its model weight from each set of routing decision information, and deduplicates the extracted results to obtain multiple candidate next nodes, as well as the decision sub-model and model weight corresponding to each candidate next node.

[0127] The output layer obtains the comprehensive weight of each candidate next node based on the inter-model weights of the decision sub-model of each candidate next node and the intra-model weight of each candidate next node;

[0128] Through the output layer, the target number of target next nodes are selected from each candidate next node in descending order of comprehensive weight;

[0129] The output layer generates and outputs routing decision information based on the target next node and the comprehensive weight corresponding to the target next node.

[0130] Based on the above embodiment, the extraction results are deduplicated through the output layer to obtain multiple candidate next nodes, as well as decision sub-models and model weights corresponding to each candidate next node;

[0131] If it is determined that the extracted current next node belongs to multiple sets of current routing decision information at the same time, then obtaining multiple current model weights of the current next node in each set of current routing decision information;

[0132] The current next node is taken as an alternative next node, and after the maximum current model internal weight is determined as the model internal weight of the current next node, the decision sub-model to which the maximum current model internal weight belongs is determined as the decision sub-model corresponding to the current next node.

[0133] Based on the above embodiment, the decision sub-model includes a first decision sub-model, a second decision sub-model and a third decision sub-model, wherein:

[0134] The first decision sub-model is a decision tree model or a random forest model, the second decision sub-model is a gradient boosting machine model, and the third decision sub-model is a long short-term memory model or a sequence model based on the self-attention mechanism.

[0135] Based on the above embodiment, the information collection module 410 may further include:

[0136] An exception processing unit, configured to filter out abnormal points from the node information of each node using a processing strategy that matches the information type of the node information;

[0137] The data cleaning unit is used to clean the node information after the abnormal point filtering process is completed, and update the current cluster information in real time based on the cleaned data.

[0138] Based on the above embodiment, the exception handling unit includes at least one of the following:

[0139] For the first type of node information whose information type is service response time or network delay, after filtering out the outliers of the first type of node information sent by each node according to the preset first threshold condition, fill the historical window mean or remove the outlier data points in the first type of node information set continuously sent by each node, and use the box value plot method to identify and filter out the outliers in each first type of information set;

[0140] For the second type of node information whose information type is processor occupancy or memory occupancy, outliers are filtered out of the second type of information sent by each node according to a preset second threshold condition.

[0141] The dynamic decision-making device for service routing provided by the embodiment of the present invention can execute the dynamic decision-making method for service routing provided by any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0142] In the technical solutions disclosed herein, the collection, storage, use, processing, transmission, provision and disclosure of user personal information involved comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0143] Example 5

[0144] Figure 5 A schematic diagram of the structure of an electronic device 10 that can be used to implement an embodiment of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or claimed herein.

[0145] like Figure 5As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., which is communicatively connected to the at least one processor 11. The memory stores a computer program that can be executed by the at least one processor. The processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 into the random access memory (RAM) 13. Various programs and data required for the operation of the electronic device 10 can also be stored in the RAM 13. The processor 11, ROM 12, and RAM 13 are connected to each other via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0146] Multiple components in the electronic device 10 are connected to the I / O interface 15, including an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a magnetic disk, an optical disk, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0147] The processor 11 may be any general-purpose and / or specialized processing component with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various specialized artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 11 executes the various methods and processes described above, such as a dynamic decision-making method for service routing, namely:

[0148] Collect multiple node information of each node in the distributed cluster in real time, and update the current cluster information in real time based on the node information; wherein the cluster information includes the current network topology information of the distributed cluster and the current node information of each topology node in the current network topology information;

[0149] In response to a dynamic routing decision request sent by a target node for a target service request, extracting current target node information of the target node and a receiver node corresponding to the target service request from the dynamic routing decision request;

[0150] The current cluster information, current target node information and receiving node are input into the pre-trained routing decision model, and the routing decision information is fed back to the target node so that the target node can send the target service request to the next node in the distributed cluster according to the routing decision information.

[0151] In some embodiments, a method for dynamically determining a service route may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the method for dynamically determining a service route described above may be performed. Alternatively, in other embodiments, processor 11 may be configured to perform a method for dynamically determining a service route by any other appropriate means (e.g., by means of firmware).

[0152] Various embodiments of the systems and techniques described herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), system-on-chip systems (SOCs), programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include being implemented in one or more computer programs that are executable and / or interpreted on a programmable system comprising at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0153] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when the computer program is executed by the processor, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The computer program may be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0154] In the context of the present invention, computer-readable storage media can be tangible media that can contain or store a computer program for use with an instruction execution system, device or equipment or used in combination with an instruction execution system, device or equipment. Computer-readable storage media can include but are not limited to electronic, magnetic, optical, electromagnetic, infrared or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, computer-readable storage media can be machine-readable signal media. More specific examples of machine-readable storage media can include electrical connections based on one or more lines, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0155] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0156] The systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0157] A computing system may include clients and servers. The clients and servers are typically remote from each other and typically interact via a communication network. This client-server relationship arises through computer programs running on the respective computers, creating a client-server relationship. The server may be a cloud server, also known as a cloud computing server or cloud host. This server is a hosting product within the cloud computing service ecosystem that addresses the management difficulties and limited scalability of traditional physical hosting and VPS services.

[0158] It should be understood that the various forms of the processes shown above can be used to reorder, add, or delete steps. For example, the steps described in the present invention can be performed in parallel, sequentially, or in a different order, as long as the desired results of the technical solution of the present invention can be achieved. This is not limited herein.

[0159] The above specific embodiments do not limit the scope of protection of the present invention. Those skilled in the art will appreciate that various modifications, combinations, sub-combinations, and substitutions may be made based on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention are intended to be included within the scope of protection of the present invention.

Claims

1. A dynamic decision-making method for service routing, characterized in that: Executed by a microservice deployed on a routing decision node in a distributed cluster, the method includes: Collect multiple node information of each node in the distributed cluster in real time, and update the current cluster information in real time based on the node information; wherein the cluster information includes the current network topology information of the distributed cluster and the current node information of each topology node in the current network topology information; In response to a dynamic routing decision request sent by a target node for a target service request, extracting current target node information of the target node and a receiver node corresponding to the target service request from the dynamic routing decision request; The current cluster information, current target node information and receiving node are input into the pre-trained routing decision model, and the routing decision information is fed back to the target node so that the target node can send the target service request to the next node in the distributed cluster according to the routing decision information.

2. The method according to claim 1, characterized in that The routing decision model includes an input layer, a feature fusion layer, a model collaboration layer, a dynamic arbitrator, and an output layer; Accordingly, the current cluster information, the current target node information, and the receiving node are input into the pre-trained routing decision model, and the routing decision information is fed back to the target node, including: Input the current cluster information, current target node information and service provider node into the pre-trained routing decision model; The input layer processes the current cluster information, current target node information, and service provider nodes to obtain time series data features, cluster topology data features, and model weight arbitration features; The feature fusion layer uses the spatiotemporal feature root cross encoder to jointly encode the time series data features and cluster topology data features to generate a fused feature vector; Multiple decision sub-models in the model collaboration layer generate a set of routing decision information based on the input fusion feature vector. Each set of routing decision information contains multiple next nodes and the model weights that match each next node. Generate inter-model weights corresponding to each decision sub-model through the dynamic arbitration layer according to the model weight arbitration characteristics; The output layer generates and outputs routing decision information based on multiple sets of routing decision information and the weights between each model.

3. The method according to claim 2, characterized in that The output layer generates and outputs routing decision information based on multiple sets of routing decision information and the weights between each model, including: The output layer extracts each next node and its model weight from each set of routing decision information, and deduplicates the extracted results to obtain multiple candidate next nodes, as well as the decision sub-model and model weight corresponding to each candidate next node. The output layer obtains the comprehensive weight of each candidate next node based on the inter-model weights of the decision sub-model of each candidate next node and the intra-model weight of each candidate next node; Through the output layer, the target number of target next nodes are selected from each candidate next node in descending order of comprehensive weight; The output layer generates and outputs routing decision information based on the target next node and the comprehensive weight corresponding to the target next node.

4. The method according to claim 3, characterized in that The extracted results are deduplicated through the output layer to obtain multiple candidate next nodes, as well as the decision sub-models and model weights corresponding to each candidate next node; If it is determined that the extracted current next node belongs to multiple sets of current routing decision information at the same time, then obtaining multiple current model weights of the current next node in each set of current routing decision information; The current next node is taken as an alternative next node, and after the maximum current model internal weight is determined as the model internal weight of the current next node, the decision sub-model to which the maximum current model internal weight belongs is determined as the decision sub-model corresponding to the current next node.

5. The method according to claim 2, characterized in that The decision sub-model includes a first decision sub-model, a second decision sub-model and a third decision sub-model, wherein: The first decision sub-model is a decision tree model or a random forest model, the second decision sub-model is a gradient boosting machine model, and the third decision sub-model is a long short-term memory model or a sequence model based on the self-attention mechanism.

6. The method according to any one of claims 1 to 5, characterized in that Collect multiple node information of each node in the distributed cluster in real time, and update the current cluster information in real time based on the node information, including: Adopting a processing strategy that matches the information type of the node information, the node information of each node is filtered out of abnormal points; The node information that has completed the outlier filtering process is cleaned, and the current cluster information is updated in real time based on the cleaned data.

7. The method according to claim 6, characterized in that A processing strategy that matches the information type of the node information is used to filter out abnormal points from the node information of each node, including at least one of the following: For the first type of node information whose information type is service response time or network delay, after filtering out the outliers of the first type of node information sent by each node according to the preset first threshold condition, fill the historical window mean or remove the outlier data points in the first type of node information set continuously sent by each node, and use the box value plot method to identify and filter out the outliers in each first type of information set; For the second type of node information whose information type is processor occupancy or memory occupancy, outliers are filtered out of the second type of information sent by each node according to a preset second threshold condition.

8. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor. The computer program is executed by the at least one processor to enable the at least one processor to perform the dynamic decision method for service routing according to any one of claims 1 to 7.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the dynamic decision method for service routing according to any one of claims 1 to 7 when executed.

10. A computer program product, characterized in that The computer program product comprises a computer program, which, when executed by a processor, implements the dynamic decision method for service routing according to any one of claims 1 to 7.

Images