A business processing method and related equipment
By constructing a state detection model and combining detection rules from the resource layer and the business layer, resource nodes can be flexibly selected, solving the problem of inflexible allocation of computing resources in model optimization business and improving resource utilization efficiency and business execution efficiency.
Patent Information
- Application Number
- CN202510591635.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-08
- Publication Date
- 2026-11-10
AI Technical Summary
In existing technologies, the allocation of computing resources for model optimization is too fixed and lacks flexibility, resulting in suboptimal resource utilization.
By constructing a state detection model and combining the detection rules of the resource layer and the business layer, the node marking state of resource nodes is determined, and suitable resource nodes are flexibly selected to execute the model to optimize business and utilize their computing resources.
It enables flexible allocation of computing resources for model optimization, improves resource utilization efficiency, reduces operation and maintenance costs, and enhances business execution efficiency and computing service quality.
Smart Images

Figure CN123068053A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of computer equipment technology, specifically to a business processing method and related equipment. Background Technology
[0002] With the development of artificial intelligence technology, more models are being trained and used. However, model training requires a large amount of computing resources. Currently, when a model optimization task is received, it is directly assigned to a device with sufficient computing resources. This device then uses its own computing resources to execute the model optimization task. However, this method of allocating computing resources for model optimization tasks is too fixed and inflexible. Summary of the Invention
[0003] This application provides a business processing method and related equipment, which can flexibly allocate computing resources for model optimization business.
[0004] On one hand, embodiments of this application provide a business processing method, which is applied to a business processing system including multiple resource nodes. The method includes:
[0005] Upon receiving a model optimization request, obtain the service parameters of the model optimization request;
[0006] Based on business parameters, determine the node requirement status of resource nodes that support model optimization business;
[0007] Based on the node demand status and the node marking status of each resource node, multiple target resource nodes are determined from multiple resource nodes to perform model optimization business; the node marking status includes at least a first node status and a second node status; the computing power resources of resource nodes in the first node status and resource nodes in the second node status are different;
[0008] The model optimization service is distributed to the service processing containers corresponding to multiple target resource nodes. The service processing containers are used to execute the model optimization service by utilizing the computing resources of the corresponding target resource nodes.
[0009] The node status of each resource node includes the node marking status of each resource node under different business types; based on the node demand status and the node marking status of each resource node, multiple target resource nodes for performing model optimization business are determined from multiple resource nodes, including:
[0010] Based on the node marking status of each resource node under different business types, determine the node marking status of each resource node under the business type to which the model optimization business belongs;
[0011] Based on the node demand status and the node marking status of each resource node under the business type to which the model optimization business belongs, multiple target resource nodes are determined from multiple resource nodes to perform the model optimization business.
[0012] The business parameters include node status identifiers; based on these business parameters, the node requirement status of resource nodes supporting model optimization services is determined, including:
[0013] If the node status identifier is the identifier of the first node status, then the node requirement status of the resource node supporting the model optimization business is determined to be the first node status.
[0014] If the node status identifier is the identifier of the second node status, then the node requirement status of the resource node supporting the model optimization business is determined to be the second node status.
[0015] The business parameters include the priority of model optimization services. Based on these parameters, the node requirement status of resource nodes supporting model optimization services is determined, including:
[0016] If the priority of the model optimization service is lower than the preset priority, then the node requirement status of the resource node supporting the model optimization service is determined to be the first node status.
[0017] If the priority of the model optimization service is higher than the preset priority, then the node requirement status of the resource node supporting the model optimization service is determined to be the second node status.
[0018] Among them, the business parameters include the number of computing cards required for model optimization business, and the number of computing cards included in different resource nodes is different;
[0019] Based on business parameters, determine the node requirement status of resource nodes that support model optimization services, including:
[0020] If the demand for computing power cards required for model optimization services is lower than the preset number, then the node demand status of the resource nodes supporting model optimization services is determined to be the first node status.
[0021] If the demand for computing power cards required for model optimization services is higher than or equal to the preset number, then the node demand status of the resource nodes supporting model optimization services is determined to be the second node status.
[0022] Among them, the business parameters include node status indication information, which is configured according to the interface parameters provided by the business processing system.
[0023] Based on business parameters, determine the node requirement status of resource nodes that support model optimization services, including:
[0024] If the status indication information indicates that a resource node in the first node state is used, then the node requirement status of the resource node supporting the model optimization business is the first node state.
[0025] If the status indication information indicates that a resource node in the second node state is used, then the node requirement status of the resource node supporting the model optimization business is the second node state.
[0026] The methods also include:
[0027] Obtain resource detection rules and business detection rules; resource detection rules are used to detect resources in resource nodes from the resource layer, and business detection rules are used to detect resource nodes from the business layer based on business execution.
[0028] Based on resource detection rules and business detection rules, the status of each resource node in the business processing system is detected to obtain the node marking status of each resource node.
[0029] Store the node tag status of each resource node.
[0030] Specifically, based on resource detection rules and business detection rules, the status of each resource node in the business processing system is detected to obtain the node marker status of each resource node, including:
[0031] If a first resource node hits the resource detection rule during the execution of its business, then the node marking status of the first resource node under different business types is determined to be the first node status.
[0032] If a second resource node fails to meet the resource detection rule but meets the business detection rule during the execution of a business, then the second resource node is identified. The node marking status under the business type to which the business belongs is the first node status, but the node marking status under other business types is the second node status.
[0033] If a third resource node fails to meet either the resource detection rule or the business detection rule during business execution, then the node marking status of the third resource node under different business types is determined to be the second node status.
[0034] Among them, the first resource node, the second resource node, and the third resource node are resource nodes in the business processing system.
[0035] The resource acquisition detection rules and business detection rules include:
[0036] Obtain load test data and business execution data of resource nodes when they are performing business operations; load test data refers to the data obtained when a resource node experiences more errors than a preset number within a target time period.
[0037] Based on load test data and business execution data, determine at least one detection rule for detecting the node status of resource nodes;
[0038] At least one detection rule is separated to obtain resource detection rules and business detection rules.
[0039] Specifically, at least one detection rule is separated to obtain resource detection rules and business detection rules, including:
[0040] Among at least one detection rule, the detection rule determined from the stress test data is identified as the resource detection rule;
[0041] Determine the number of services that cause anomalies when resource nodes execute different services under each detection rule;
[0042] The detection rules corresponding to the number of abnormal services being greater than or equal to a preset number are determined as resource detection rules, and the detection rules corresponding to the number of abnormal services being less than a preset number are determined as service detection rules.
[0043] On one hand, embodiments of this application provide a service processing apparatus applied to a service processing system, the service processing system including multiple resource nodes; the apparatus includes:
[0044] The acquisition unit is used to acquire the service parameters of the model optimization service when the model optimization service is received.
[0045] The processing unit is used to determine the node requirement status of resource nodes that support model optimization services based on business parameters.
[0046] The processing unit is also used to determine multiple target resource nodes for performing model optimization business from multiple resource nodes based on the node demand status and the node marking status of each resource node; the node marking status includes at least a first node status and a second node status; the computing power resources of resource nodes in the first node status and resource nodes in the second node status are different.
[0047] The processing unit is also used to distribute model optimization services to business processing containers corresponding to multiple target resource nodes. The business processing containers are used to execute model optimization services by utilizing the computing resources of the corresponding target resource nodes.
[0048] The node status of each resource node includes the node marking status of each resource node under different business types; when the processing unit determines multiple target resource nodes for performing model optimization business from multiple resource nodes based on the node demand status and the node marking status of each resource node, it can be specifically used for:
[0049] Based on the node marking status of each resource node under different business types, determine the node marking status of each resource node under the business type to which the model optimization business belongs;
[0050] Based on the node demand status and the node marking status of each resource node under the business type to which the model optimization business belongs, multiple target resource nodes are determined from multiple resource nodes to perform the model optimization business.
[0051] The business parameters include node status identifiers; when the processing unit determines the node requirement status of resource nodes supporting model optimization services based on the business parameters, it can specifically use them for:
[0052] If the node status identifier is the identifier of the first node status, then the node requirement status of the resource node supporting the model optimization business is determined to be the first node status.
[0053] If the node status identifier is the identifier of the second node status, then the node requirement status of the resource node supporting the model optimization business is determined to be the second node status.
[0054] Among these, the business parameters include the priority of the model optimization business. When the processing unit determines the node requirement status of the resource nodes supporting the model optimization business based on the business parameters, it can specifically use them for:
[0055] If the priority of the model optimization service is lower than the preset priority, then the node requirement status of the resource node supporting the model optimization service is determined to be the first node status.
[0056] If the priority of the model optimization service is higher than the preset priority, then the node requirement status of the resource node supporting the model optimization service is determined to be the second node status.
[0057] Among them, the business parameters include the number of computing cards required for model optimization business, and the number of computing cards included in different resource nodes is different;
[0058] When determining the node requirement status of resource nodes that support model optimization services based on business parameters, the processing unit can specifically be used for:
[0059] If the demand for computing power cards required for model optimization services is lower than the preset number, then the node demand status of the resource nodes supporting model optimization services is determined to be the first node status.
[0060] If the demand for computing power cards required for model optimization services is higher than or equal to the preset number, then the node demand status of the resource nodes supporting model optimization services is determined to be the second node status.
[0061] Among them, the business parameters include node status indication information, which is configured according to the interface parameters provided by the business processing system.
[0062] When determining the node requirement status of resource nodes that support model optimization services based on business parameters, the processing unit can specifically be used for:
[0063] If the status indication information indicates that a resource node in the first node state is used, then the node requirement status of the resource node supporting the model optimization business is the first node state.
[0064] If the status indication information indicates that a resource node in the second node state is used, then the node requirement status of the resource node supporting the model optimization business is the second node state.
[0065] The acquisition unit is also used to: acquire resource detection rules and business detection rules; the resource detection rules are used to detect resources in resource nodes from the resource layer, and the business detection rules are used to detect resource nodes from the business layer using business execution status;
[0066] The processing unit is also used to perform status detection on each resource node in the business processing system according to resource detection rules and business detection rules, and obtain the node marking status of each resource node.
[0067] The processing unit is also used to store the node tag status of each resource node.
[0068] Specifically, when the processing unit performs status detection on each resource node in the business processing system according to resource detection rules and business detection rules to obtain the node tag status of each resource node, it can be used for:
[0069] If a first resource node hits the resource detection rule during the execution of its business, then the node marking status of the first resource node under different business types is determined to be the first node status.
[0070] If a second resource node fails to meet the resource detection rule but meets the business detection rule during the execution of a business, then the second resource node is identified. The node marking status under the business type to which the business belongs is the first node status, but the node marking status under other business types is the second node status.
[0071] If a third resource node fails to meet either the resource detection rule or the business detection rule during business execution, then the node marking status of the third resource node under different business types is determined to be the second node status.
[0072] Among them, the first resource node, the second resource node, and the third resource node are resource nodes in the business processing system.
[0073] Specifically, when acquiring resource detection rules and business detection rules, the acquisition unit can be used for:
[0074] Obtain load test data and business execution data of resource nodes when they are performing business operations; load test data refers to the data obtained when a resource node experiences more errors than a preset number within a target time period.
[0075] Based on load test data and business execution data, determine at least one detection rule for detecting the node status of resource nodes;
[0076] At least one detection rule is separated to obtain resource detection rules and business detection rules.
[0077] Specifically, when the acquisition unit separates at least one detection rule to obtain resource detection rules and business detection rules, it can be used for:
[0078] Among at least one detection rule, the detection rule determined from the stress test data is identified as the resource detection rule;
[0079] Determine the number of services that cause anomalies when resource nodes execute different services under each detection rule;
[0080] The detection rules corresponding to the number of abnormal services being greater than or equal to a preset number are determined as resource detection rules, and the detection rules corresponding to the number of abnormal services being less than a preset number are determined as service detection rules.
[0081] On one hand, embodiments of this application provide a computer device, the computer device comprising:
[0082] A processor is used to execute computer programs;
[0083] A computer-readable storage medium storing a computer program, which, when executed by a processor, implements the business processing method described above.
[0084] On one hand, embodiments of this application provide a computer-readable storage medium storing a computer program, which is loaded by a processor and executed as described above in the business processing method.
[0085] On the one hand, embodiments of this application provide a computer program product, which includes a computer program or computer instructions, and the computer program or computer instructions implement the above-mentioned business processing method when executed by a processor.
[0086] In this embodiment, upon receiving a model optimization service, the service parameters of the model optimization service are obtained. Based on the service parameters, the node demand status of the resource nodes supporting the model optimization service is determined. Thus, based on the node demand status and the node marking status of each resource node, multiple target resource nodes for executing the model optimization service are determined from multiple resource nodes. The node marking status includes at least a first node status and a second node status. Resource nodes in the first node status and resource nodes in the second node status have different computing resources. The model optimization service is distributed to the service processing containers corresponding to the multiple target resource nodes. The service processing containers are used to execute the model optimization service using the computing resources of the corresponding target resource nodes. Therefore, this embodiment utilizes service parameters and the node marking status of each resource node to achieve flexible selection of the resource nodes required for the model optimization service. Furthermore, resource nodes in different node states have different computing resources, which allows for the reasonable utilization of the computing resources of resource nodes in different node marking states to execute the model optimization service, achieving flexible allocation of computing resources for the model optimization service. Attached Figure Description
[0087] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0088] Figure 1 A schematic diagram of a state detection model provided in this application embodiment;
[0089] Figure 2 An architecture diagram of a business processing system provided in this application embodiment;
[0090] Figure 3 A schematic diagram of a detection rule provided in an embodiment of this application;
[0091] Figure 4 An architecture diagram of another business processing system provided in this application embodiment;
[0092] Figure 5 A schematic diagram of a business processing flow provided in an embodiment of this application;
[0093] Figure 6 A flowchart illustrating a business processing method provided in an embodiment of this application;
[0094] Figure 7 A flowchart illustrating a business processing method provided in an embodiment of this application;
[0095] Figure 8 A schematic diagram of a resource node stress test provided in an embodiment of this application;
[0096] Figure 9 This is a schematic diagram illustrating the execution status of resource nodes under different business conditions, provided as an embodiment of this application.
[0097] Figure 10 This is a schematic diagram of the structure of a service processing device provided in an embodiment of this application;
[0098] Figure 11 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation
[0099] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.
[0100] First, the technical terms involved in the embodiments of this application will be explained.
[0101] I. Resource Nodes
[0102] A resource node refers to a device that can provide computing power resources. This resource node can also be called a computing power device. This resource node may include computing power cards, such as graphics cards, CPU cards, etc. Computing power cards can provide powerful computing capabilities.
[0103] In the embodiments of this application, the resource nodes can be devices from different manufacturers, of different types, and using different protocols. These resource nodes can collectively provide computing resources for the execution of services or computing tasks included in those services. Therefore, the resource nodes in the embodiments of this application can be referred to as heterogeneous nodes. Correspondingly, the computing resources provided by these resource nodes can be referred to as heterogeneous computing resources.
[0104] II. Node Marking Status
[0105] Node marking status refers to the current node status of a resource node. In this embodiment, the node marking status may include at least a first node status and a second node status. The first node status may indicate that the resource node is in a sub-healthy state; specifically, the first node status may indicate that the computing power resources within the resource node are in a sub-healthy state. The second node status may indicate that the resource node is in a healthy state; specifically, the second node status may indicate that the computing power resources within the resource node are in a healthy state.
[0106] The difference between a resource node in the first node state and a resource node in the second node state lies in the fact that both can provide computing power services for business (such as model optimization business), but their computing power service capabilities differ. In one implementation, the computing power service capability of a resource node in the first node state is lower than that of a resource node in the second node state; that is, the computing power service capability of a resource node in a sub-healthy state is lower than that of a resource node in a healthy state.
[0107] The computing power service capabilities include, but are not limited to, service scale and computational load. In this embodiment, the computing power resources of a node in the first node state can support scenarios with a small service scale or low computational load, while the computing power resources of a node in the second node state can support scenarios with a large service scale or high computational load. Alternatively, the difference between a resource node in the first node state and a resource node in the second node state is that the computing power resources of the resource node in the first node state are different from those of the resource node in the second node state.
[0108] With the increasing development of large-scale AI (Artificial Intelligence) models, the scale of heterogeneous computing resources is also constantly increasing. Ensuring the stable operation of computing cards is a fundamental requirement. While most anomalies in the resource nodes where computing cards reside can be detected by setting resource detection indicators, a small number of anomalies cannot be detected by building resource detection indicators at the resource layer and only become apparent when anomalies occur during certain business computation processes. Therefore, this application provides a business processing solution to address the above-mentioned problems. This solution can construct a state detection model, which can perform state detection of resource nodes at both the resource and business levels, bridging the gap between resource detection and business computation anomaly detection, improving the accuracy of resource node state detection, and thus improving the efficiency of business execution.
[0109] Specifically, such as Figure 1The diagram illustrates a state detection model provided in this application embodiment. This model defines two types of detection rules: resource-level detection rules (i.e., basic rules) and business-level detection rules (i.e., scenario-based rules). These rules primarily utilize big data, aggregating stress test data from stress tests on abnormal resource nodes and business execution data generated by each resource node's business operations. Specifically, resource detection rules are used to detect multiple resource nodes in the business processing system at the resource layer, while business detection rules are used to detect multiple resource nodes at the business layer. These two detection rules ensure accurate detection of resource nodes in the business processing system. When a resource node in the business processing system matches one or more of the above two detection rules during operation, its node status is determined to be a first node status (e.g., sub-healthy state). When a resource node in the business processing system does not match either of the above two detection rules during operation, its node status is determined to be a second node status (e.g., healthy state). The computing resources of resource nodes in the first node status and resource nodes in the second node status are different.
[0110] It should be noted that when the node state of a resource node includes node states other than the first node state and the second node state, the state detection model in this application embodiment can be a state detection model for the first node state, that is, as shown in the example. Figure 1 In this process, as long as a resource node matches the detection rules defined in the state detection model, its node label state can be determined as the first node state. Optionally, such as... Figure 1 In the process, a marker indicating the first node state is added to the resource node. As for the detection rules in the resource node mis-hit state detection model, if the node marker state of the resource node is determined to be not the first node state, further detection rules besides detecting the first node state are needed.
[0111] Furthermore, upon receiving a model optimization service request, suitable resource nodes can be selected from resource nodes with different node marking states to provide computing power services for the model optimization service. In one implementation, the service parameters of the model optimization service (such as priority, number of computing cards required, etc.) are obtained; based on the service parameters, the node requirement state of the resource nodes supporting the model optimization service is determined, such as a first node state or a second node state; based on the node requirement state and the node marking state of each resource node, multiple target resource nodes for executing the model optimization service are determined from the multiple resource nodes; for example, if the node requirement state is a first node state, then resource nodes in the first node state can be selected from the multiple resource nodes, and these selected resource nodes in the first node state are determined as the target resource nodes for executing the model optimization service. Then, the model optimization service is distributed to the service processing containers corresponding to the multiple target resource nodes, and the service processing containers are used to execute the model optimization service using the computing power resources of the corresponding target resource nodes.
[0112] In some alternative implementations, a resource node in the first node state refers to a resource node that can provide some computing power (e.g., the graphics card has a problem, but the CPU is available). This means that the resource node in the first node state is in a sub-healthy state, so it can be like this: Figure 1 As shown, the resource node in the first node state is repaired, and after the repair is completed, the first node state mark of the resource node is removed.
[0113] As can be seen, the solution provided in this application can achieve the following beneficial effects: 1) This solution innovatively uses big data from the operation of the business and the resource operation process of resource nodes as the source data for building the state detection model, and innovatively proposes the concept of resource service with node marking status (such as different computing power resource services for resource nodes with different node marking statuses). In this way, the problem of abnormal business operation but undetectable resource indicators can be judged by building a state detection model, thus making up for the deviation between resource detection and business computing anomalies and improving the accuracy of resource node state detection. 2) In terms of business experience, this solution supports scenarios where business is abnormal but resource indicators cannot be captured by building a state detection model. This effectively avoids allocating resource nodes with insufficient computing power resources to the business, which is conducive to improving the R&D efficiency of the business and enabling the business training program to run and iterate quickly. 3) In terms of computing power service operation, this solution optimizes the shortcomings of the state detection scheme from a technical perspective, avoiding the operational work from getting bogged down in the long-term joint positioning of anomalies by business and computing power platform personnel, which is conducive to improving the quality of computing power service operation. 4) In terms of cost, this solution not only alleviates operational and maintenance pressure and reduces operational and maintenance manpower costs, but also avoids idle computing power during the process of locating problems after business computing anomalies, thereby improving the utilization of computing power resources in each resource node and reducing the cost of computing power usage for the business. 5) By constructing a status detection model, it is beneficial to monitor computing power availability from the perspectives of resource detection and big data, which is conducive to the large-scale deployment of computing power and supports the service capabilities of massive computing power. 6) Through business parameters, the node demand status of resource nodes supporting model optimization services can be determined. Then, based on the node demand status and the node marking status of each resource node, multiple target resource nodes for executing model optimization services can be determined from multiple resource nodes. This allows for the rational use of computing power resources in each resource node, achieving flexible allocation of computing power resources for model optimization services.
[0114] The business processing system provided in the embodiments of this application will be described next. Please refer to... Figure 2 This is an architecture diagram of the business processing system provided in an embodiment of this application. Figure 2 In this context, the business processing system comprises a business layer, a task execution layer, a resource layer, and a storage layer. Specifically:
[0115] The business layer, also known as the product layer, defines the supported business processes. These processes can include model optimization, which refers to the training of models, such as... Figure 2In this context, the models used in model optimization services can include hybrid large-scale models, game models, medical models, NPL (Need-Problem-Solution Thinking Model) models, and so on. The model optimization service can include at least one computational task, each requiring different, but also potentially the same, computational resources.
[0116] The task execution layer is used to execute model optimization tasks; more specifically, it executes the computational tasks within these tasks. This layer includes multiple business processing containers (also known as heterogeneous containers, which are containers that share different types of computing resources, such as sharing CPUs and GPUs) for task execution. Model optimization tasks can be executed within these business processing containers, or in other words, the computational tasks included in the model optimization tasks can be executed within these containers.
[0117] The resource layer includes multiple resource nodes, such as Figure 2 In this model, the resource layer includes node1, node2, node3, and node4. These resource nodes may have different or identical node information, which can include: the source of the resource node (e.g., node1 from manufacturer A, node2 from manufacturer B), model, node type, computing power resources, etc. Furthermore, each resource node can have a corresponding node marker state, which can be determined based on a state detection model. Specifically, the state detection model can be used to detect resource nodes. When a resource node matches a detection rule in the state detection model, its node marker state is determined to be the first node state. For example... Figure 2 In the process, after nodes 1, 2, 3, and 4 are detected according to the detection rules in the state detection model, nodes 1, 2, and 3 are marked as second node states, while node 4 is marked as first node state. Furthermore, when the resource layer receives container production requests for model optimization, resource nodes with the appropriate marked states can be selected from multiple resource nodes. Pods can then be built based on these selected resource nodes; that is, pods can run on the selected resource nodes. Figure 2 In this system, four pods (pod1-pod4) can run on four nodes, with one pod running on each node. When business logic is executed at the task execution layer, the pods built at the resource layer become the business processing containers within the task execution layer.
[0118] The storage layer is used to store the detection rules defined by the state detection model, such as Figure 2As shown, the storage layer includes a storage database (DB). The detection rules defined by the state detection model are stored in the storage database. The detection rules defined in this state detection model include resource detection rules at the resource level and business detection rules at the business level, such as... Figure 3 The diagram shown illustrates a detection rule provided in an embodiment of this application. Resource detection rules may include, but are not limited to: the cumulative number of errors in resource nodes, the number of abnormal node failures, the cumulative number of node resource scheduling failures, the cumulative number of failed business computations, etc. Business detection rules may include, but are not limited to: the failure frequency of a specific business, the total number of failures of a specific business, the performance degradation of a specific business, etc.
[0119] In this embodiment, the resource node can be a server or a terminal device. The terminal device can be a smartphone, tablet, laptop, desktop computer, etc. The server can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms, etc.
[0120] In some embodiments, this application also provides an architecture diagram of another business processing system. Please refer to [link / reference]. Figure 4 This business processing system includes a business layer, a resource layer, filters, a storage database (DB), a rule logic processor, a rule generator, and so on. Among them:
[0121] The business layer can include various model optimization services, such as speech training, vision training, and medical training. The resource layer includes multiple resource nodes, and can schedule appropriate resource nodes for container generation to allocate computing resources for model optimization services.
[0122] The filter is used to select the node demand status of resource nodes that support model optimization services based on filtering rules. These filtering rules may include, but are not limited to: business-defined rules, interface parameter rules, business priority rules, and business computing card demand rules. Business-defined rules define the identifiers of each node's marked status. For example, if a node's marked status includes a first node status and a second node status, then the business-defined rules define the identifiers for the first and second node statuses. Interface parameter rules define interface parameters, which can be used to configure the node demand status of resource nodes that support the service. Business priority rules define the node demand status of resource nodes that can be used by services with lower priority. Business computing card demand rules define the node demand status of resource nodes that can be used by the service when the service's computing card demand is lower than a preset number. Furthermore, after selecting the node demand status of resource nodes that support model optimization services, the filter can distribute the model optimization service to the resource layer for resource node scheduling and production.
[0123] The rule generator primarily draws from load testing data after resource nodes experience anomalies and from business execution data generated when resource nodes perform business operations. The rule generator can include a load testing generator and a business generation generator. The load testing generator is used for stress test evaluation; that is, it performs stress tests on resource nodes with concentrated errors and determines detection rules from the stress data obtained. The business generation generator obtains business execution data from the resource nodes and determines detection rules from this data.
[0124] The rule logic processor separates the rules generated by the rule generator into resource detection rules and business detection rules, and stores these rules in the storage database (DB). The rule logic processor includes at least one processing process that separates the rules generated by the rule generator. The reason for separating the detection rules here is that under resource detection rules, the computing power of resource nodes is the same when serving different businesses, while under business detection rules, the computing power of resource nodes varies when serving different businesses; for example, the computing power of a resource node serving a specific business under business detection rules will be lower than its computing power in other business scenarios.
[0125] The resource layer can provide computing power services by scheduling suitable resource nodes for model optimization services based on the node demand status of the resource nodes that support model optimization services.
[0126] As can be seen, the embodiments of this application can improve the accuracy of resource node status detection through the collaborative work of the business layer, resource layer, filter, storage DB, rule logic processor and rule generator, and can flexibly allocate resource nodes to model optimization tasks according to the node requirement status of resource nodes supporting model optimization business, thereby improving business execution efficiency.
[0127] In one embodiment, combined with such Figure 2 and Figure 4 The system architecture shown is a schematic diagram illustrating the business processing flow provided in the embodiments of this application. Please refer to [link / reference]. Figure 5 The business process may include the following parts:
[0128] (1) Summarize candidate detection rules at the resource level.
[0129] In specific implementations, such as Figure 4 As shown, when resource nodes experience a concentrated number of errors, stress testing can be performed on the resource nodes to obtain stress test data, and candidate detection rules can be determined from this data. Specifically, stress testing can be performed on resource nodes using a stress test generator. For example, if the stress test data includes 5 memory error occurrences, then the number of memory error occurrences can be used as a candidate detection rule. Among other things, such as... Figure 4 In this context, stress testing can include computing power stress testing and communication performance testing. Computing power testing refers to the performance testing of computing power cards in resource nodes, while communication performance testing refers to the communication testing between resource nodes and other resource nodes, such as bandwidth testing, latency testing, and so on.
[0130] (2) Summarize the candidate detection rules at the business level.
[0131] In specific implementations, such as Figure 4 As shown, it is possible to obtain business execution data of resource nodes when executing business operations. This involves running different business operations on resource nodes to obtain business execution data, and then extracting candidate detection rules for the business layer from this data. For example, if a resource node experiences performance degradation when executing business operation A, but performs business operation B normally, then the performance degradation of business operation A can be used as a candidate detection rule for the business layer. Alternatively, if a resource node experiences performance degradation when executing business operation A, then the performance degradation of business operation A can be used as a candidate detection rule for the business layer.
[0132] (3) Perform rule logic filtering to obtain at least one detection rule.
[0133] In this embodiment, it is necessary to perform rule filtering on candidate detection rules at the resource level and candidate detection rules at the business level to obtain at least one detection rule. Specifically, a first condition and a second condition are defined. If there is a candidate detection rule at the resource level that satisfies the first condition, then the candidate detection rule that satisfies the first condition is determined as a detection rule. If there is a candidate detection rule at the business level that satisfies the second condition, then the candidate detection rule that satisfies the second condition is determined as a detection rule.
[0134] As can be seen, the embodiments of this application can utilize various tests on resource nodes to achieve the computing power service capabilities of resource nodes through (1)-(3). Furthermore, the embodiments of this application collect a large amount of stress test data and business execution data of resource nodes, which can successfully rely on big data to aggregate the different performances of resource nodes in different businesses (or computing tasks). The resulting detection rules are more representative and can better detect resource nodes in the future.
[0135] (4) Resource detection rules and business detection rules are separated from at least one detection rule.
[0136] Specifically, resource detection rules and business detection rules can be separated from at least one detection rule using the processing steps in the rule logic processor. For example, if at least one detection rule includes detection rules obtained from stress test data, then that detection rule can be identified as a resource detection rule. Similarly, if the number of abnormal business operations occurring on resource nodes under a certain detection rule in at least one detection rule is less than a preset number, then that rule is identified as a business detection rule. Separating resource detection rules and business detection rules facilitates the subsequent determination of the node marking status of each resource node under different business types.
[0137] Optionally, resource detection rules and business detection rules can be stored in a storage database.
[0138] (5) Detect resource nodes based on resource detection rules and business detection rules to obtain the node marking status of resource nodes.
[0139] In one implementation, resource detection rules and business detection rules are used to detect resource nodes in the first node state. If a resource node matches one or more of the resource detection rules and business detection rules during operation, its node marking state can be determined as the first node state. If a resource node does not match either the resource detection rules or the business detection rules during operation, its node marking state can be determined as the second node state.
[0140] Furthermore, a status detection model is constructed based on resource detection rules and business detection rules. This model can perform status detection on resource nodes and, after determining the node marker status of a resource node, associate that marker status with the resource node, such as by marking the resource node with a node marker status. Additionally, when the first node status indicates that a resource node is in a sub-healthy state, maintenance is performed on that resource node. After maintenance, the status detection model can be used to remove the first node marker status from the resource node.
[0141] It should be understood that resource nodes are periodically checked to ensure that the node marking status of resource nodes is the most recent node marking status.
[0142] (6) Initiate the execution of the model optimization task. When a model optimization task exists, the task's parameters can be obtained.
[0143] (7) Determine the node requirement status of resource nodes that support model optimization services.
[0144] In one implementation, the node requirement status of resource nodes supporting model optimization services can be determined based on business parameters. Specifically, when there are multiple model optimization services, a filter can be invoked to determine the node requirement status of resource nodes supporting each model optimization service based on its business parameters. The filter defines filtering rules, which determine the node requirement status of resource nodes supporting model optimization services based on both business parameters and the filtering rules. For example, if business parameters include business priority, and the filtering rules define the node requirement status of resource nodes available to low-priority services as the first node status, then when the priority of a model optimization service is low, the node requirement status of the resource nodes supporting that service is determined to be the first node status.
[0145] In one implementation, if the model optimization task includes at least one computation task, the node requirement status of the resource nodes supporting each computation task can be determined based on the business parameters of each computation task.
[0146] (8) Deploy container production for model optimization business.
[0147] After executing (7), the filter can issue a container production request for the model optimization service to the resource layer. The container production request includes the node requirement status of the resource nodes that support the model optimization service. In one implementation, a container production request for the computing task in the model optimization service can be issued to the resource layer. The container production request for the computing task includes the node requirement status of the corresponding resource node that supports the computing task.
[0148] (9) Select the resource node production business processing container that matches the node's required state.
[0149] As mentioned above, each resource node has a node tag status. The resource node that matches the node requirement status can be selected from multiple resource nodes, and a pod (i.e., a business processing container) can be created on the selected resource node.
[0150] (10) Determine whether container production is complete. If yes, execute (12); otherwise, execute (11).
[0151] Specifically, it determines whether container production has been completed according to business requirements. For example, if the business requirement is to produce 4 containers, then container production is completed if 4 containers have been produced, otherwise it is not.
[0152] (11) Select resource nodes with other node marking status according to business needs to produce business processing containers.
[0153] It should be understood that the computing power of the selected resource nodes in other node-marked states must be stronger than that of the resource nodes in the node-marked states that match the node's demand state; or, in other words, the computing power of the selected resource nodes in other node-marked states must be stronger than that of the resource nodes in the node-marked states that match the node's demand state. In one implementation, if the number of production business processing containers for the selected resource nodes in the node-marked states that match the node's demand state is insufficient to meet business requirements, then production business processing containers for resource nodes in other node-marked states can be selected. For example, if the node's demand state is the first node state, and the selected resource nodes in the first node state (e.g., sub-healthy state) do not have enough production business processing containers, then resource nodes in the second node state (e.g., healthy state) can be selected.
[0154] (12) Execute model optimization tasks in the business processing container using the computing resources of the corresponding resource nodes. More specifically, execute the computing tasks included in the model optimization business in the business processing container using the computing resources of the corresponding resource nodes.
[0155] As can be seen, the embodiments of this application can utilize various tests on resource nodes to cover the computing power service capabilities of resource nodes, and by collecting a large amount of stress test data and business execution data generated during the testing process, it can successfully rely on big data to aggregate the different performances of resource nodes in different businesses, obtain more comprehensive resource detection rules and business detection rules, improve the accuracy of resource node status detection, and then flexibly allocate resource nodes to the model optimization business according to the node marking status of each resource node, providing computing power resources that meet the execution of the model optimization task.
[0156] The business processing methods provided in the embodiments of this application will be described in detail below.
[0157] Please see Figure 6 This is a flowchart illustrating a business processing method provided in an embodiment of this application. The business processing method can be executed by a computer device and can be applied to a business processing system. The business processing system may include multiple resource nodes. In some implementations, the business processing system can be a resource heterogeneous system. A resource heterogeneous system refers to a system where some or all of the resource nodes have different node information. This node information may include, but is not limited to, the source, model, node type, computing power resources, etc., of the resource nodes. The business processing method may include the following steps S601-S604:
[0158] S601. Upon receiving a model optimization service request, obtain the service parameters of the model optimization service request.
[0159] The model optimization service refers to the training of models, which can include, but are not limited to, speech models, visual models, and text models. Service parameters can include at least one of the following: node status identifiers, the priority of the model optimization service, the required computing power cards for the model optimization service, node status indication information, and the computational load of the service.
[0160] In one implementation, when submitting a model optimization request, service parameters can be set to determine the node tagging status of resource nodes supporting the request. For example, if the service parameters include node status identifiers, then when submitting the request, these identifiers can be configured to uniquely identify a node's status. Correspondingly, when the computer receives the request, it can obtain the configured service parameters. In another implementation, upon receiving the request, the service parameters can be determined directly based on the request itself. For example, if the service parameter is a priority, the priority of the optimization request can be determined based on its service type. Alternatively, the priority can be determined based on the importance of the optimization request; more important requests have higher priorities.
[0161] S602. Based on the business parameters, determine the node requirement status of the resource nodes that support the model optimization business.
[0162] The node demand status can include either a first node status or a second node status. The first node status indicates that the resource node is in a sub-healthy state; the second node status indicates that the resource node is in a healthy state. The following section details how to determine the node demand status of resource nodes supporting model optimization based on business parameters:
[0163] (1) Business parameters include node status identifiers.
[0164] In this embodiment, the node requirement status of usable resource nodes can be customized according to business needs. In this case, different node statuses correspond to different identifiers. When receiving a model optimization task, the node status identifier can be carried. The specific implementation of step S602 includes: if the node status identifier is the identifier of the first node status, then the node requirement status of the resource node supporting the model optimization business is determined to be the first node status; if the node status identifier is the identifier of the second node status, then the node requirement status of the resource node supporting the model optimization business is determined to be the second node status.
[0165] (2) Business parameters include the priority of model optimization business.
[0166] In this embodiment, services with low priority can use resource nodes in a first node state, and / or services with high priority can use resource nodes in a second node state. Low priority can mean a priority less than a preset priority, and high priority can mean a priority greater than or equal to a preset priority. The preset priority can be set according to requirements, and this embodiment does not impose any limitations on it.
[0167] In this implementation, step S602 may be implemented as follows: if the priority of the model optimization service is lower than the preset priority, the node demand state of the resource node supporting the model optimization service is determined to be the first node state; if the priority of the model optimization service is higher than the preset priority, the node demand state of the resource node supporting the model optimization service is determined to be the second node state.
[0168] (3) Business parameters include the amount of computing power cards required for model optimization business.
[0169] The number of computing cards included in different resource nodes can vary, or they can be the same. In this embodiment, it can be defined that if the computing card demand required by the service is less than a preset number, a resource node in the first node state is used, and / or if the computing card demand required by the service is higher than or equal to the preset number, a resource node in the second node state is used. The specific implementation of step S602 can include: if the computing card demand required by the model optimization service is less than the preset number, then the node demand state of the resource node supporting the model optimization service is determined to be the first node state; if the computing card demand required by the model optimization service is higher than or equal to the preset number, then the node demand state of the resource node supporting the model optimization service is determined to be the second node state; for example, if the computing card demand used by the model optimization service is at the hundred-card level and less than the preset number, then the node demand state of the resource node supporting the model optimization service can be determined to be the first node state.
[0170] (4) Business parameters include node status indication information.
[0171] In this embodiment, the service processing system provides interface parameters for configuring the node requirement status of the resource nodes that can be used. This status indication information is obtained by configuring the interface parameters provided by the service processing system. For example, when a user wants to use a resource node in a first node state to execute an optimization model service, the status indication information for using the resource node in the first node state can be configured through the interface parameters. A specific implementation of step S602 may include: if the status indication information indicates the use of a resource node in the first node state, then the node requirement status of the resource node supporting the model optimization service is determined to be the first node state; if the status indication information indicates the use of a resource node in a second node state, then the node requirement status of the resource node supporting the model optimization service is determined to be the second node state.
[0172] It should be understood that, for model optimization services, these services may include computational tasks. Since the service parameters include the service parameters of the computational tasks, the node requirement status of the resource nodes determined in step S602 can refer to the node requirement status of the resource nodes supporting the computational tasks in the model optimization service. The number of computational tasks can be one or more, and the computing resources required by each computational task may be different or the same. Therefore, the node requirement status of the resource nodes supporting each computational task can be determined based on the service parameters of each computational task.
[0173] S603. Based on the node demand status and the node marking status of each resource node, determine multiple target resource nodes from multiple resource nodes to perform model optimization business; the node marking status includes at least the first node status and the second node status; the computing power resources of resource nodes in the first node status and resource nodes in the second node status are different.
[0174] The first node status indicates that the resource node is in a sub-healthy state, and the second node status indicates that the resource node is in a healthy state. The difference in computing power between the resource node in the first node status and the resource node in the second node status can be: the computing power of the resource node in the first node status is lower than that of the resource node in the second node status; or, the resource service capacity provided by the computing power of the resource node in the first node status is lower than that provided by the computing power of the resource node in the second node status.
[0175] In one implementation, step S603 may include: determining multiple candidate resource nodes for performing model optimization services from multiple resource nodes based on the node demand status and the node tag status of each resource node; and determining multiple target resource nodes from the multiple candidate resource nodes for performing model optimization services based on the number of service demands in the service parameters. Specifically, candidate resource nodes may be randomly selected as target resource nodes from multiple candidate resource nodes based on the number of service demands in the service parameters. In this case, the multiple target resource nodes include multiple resource nodes in the node demand status, that is, the number of target resource nodes can be N, where N is an integer greater than 1, and the number of resource nodes in the node demand status (such as the first node status) is also N.
[0176] In another implementation, this application embodiment supports resource nodes with different node marking states jointly executing model optimization services. In this case, the multiple target resource nodes include at least one resource node in a first node state and at least one resource node in a second node state. As one implementation, step S603 may include: selecting at least one resource node from the multiple resource nodes whose node marking state matches the node demand state; for example, selecting all resource nodes whose node marking state matches the node demand state, or randomly selecting one or two resource nodes from all resource nodes whose node marking state matches the node demand state, etc. This application embodiment does not impose any limitations on this; if the number of selected resource nodes is less than the number of service requirements, then based on the difference between the number of selected resource nodes and the number of service requirements, determining the remaining resource nodes from the unselected resource nodes in other node marking states, and determining the remaining resource nodes and the selected resource nodes whose node marking state matches the node demand state as target resource nodes. For example, taking the node demand state as the first node state as an example, the multiple resource nodes include 3 resource nodes in the first node state and 3 resource nodes in the second node state; the number of business demands is 4. From the multiple resource nodes, 3 resource nodes in the first node state are selected; then, one resource node is determined from the resource nodes that have not been selected and are in other node marking states (i.e., the second node state); then the 4 target resource nodes include 3 resource nodes in the first node state and 1 resource node in the second node state.
[0177] As another implementation, the number of business requirements includes a first quantity and a second quantity. The first quantity indicates the number of resource nodes required to optimize the business using the execution model, and the second quantity indicates the number of resource nodes that need to be selected in the node requirement state. Step S603 may include: selecting resource nodes corresponding to the node marking state that match the node requirement state from multiple resource nodes according to the first quantity, and determining the remaining resource nodes from resource nodes in other node marking states according to the second quantity. For example, taking the node requirement state as the first node state as an example, such as... Figure 4 As shown, the multiple resource nodes include 3 resource nodes in the first node state (node4, node7, and node8) and 5 resource nodes in the second node state; the first quantity is 2 and the second quantity is 2; based on the first quantity, 2 resource nodes in the first node state can be selected from the multiple resource nodes; then 2 resource nodes (i.e., the remaining resource nodes mentioned above) can be determined from the 5 resource nodes in the second node state.
[0178] Among them, the computing power of the remaining resource nodes must be stronger than that of the resource nodes in the first node state. In other words, taking the node demand state as the first node state as an example, the first node state indicates that the resource node is in a sub-healthy state, and the second node state indicates that the resource node is in a healthy state. After selecting the resource nodes in the sub-healthy state, the remaining resource nodes must be determined from the resource nodes in the healthy state.
[0179] In one implementation, for a given resource node, its node marking status may differ under different business types. For example, the resource node might be marked as a first node state under business type A and a second node state under business type B. The first node state under business type A indicates that the resource node is in a sub-healthy state under business type A; the second node state under business type B indicates that the resource node is in a healthy state under business type B. Of course, the node marking status of the resource node is the same across all business types. For example, the resource node might be marked as a first node state across all types. In this implementation, the node status of each resource node includes its node marking status under different business types. The specific implementation of step S603 may include: determining the node marking status of each resource node under the business type to which the model optimization business belongs, based on the node marking status of each resource node under different business types; and determining multiple target resource nodes from multiple resource nodes for executing the model optimization business, based on the node demand status and the node marking status of each resource node under the business type to which the model optimization business belongs.
[0180] S604. Distribute the model optimization service to the service processing containers corresponding to multiple target resource nodes. The service processing containers are used to execute the model optimization service by utilizing the computing resources of the corresponding target resource nodes.
[0181] Specifically, business processing containers can be deployed on each target resource node, and the computational tasks in the model optimization business can be distributed to each business processing container. Each business processing container can utilize the computing resources of the corresponding target resource node to execute the computational task, thereby completing the execution of the model optimization business.
[0182] In this embodiment, upon receiving a model optimization service, the service parameters of the model optimization service are obtained; based on the service parameters, the node demand status of resource nodes supporting the model optimization service is determined; based on the node demand status and the node marking status of each resource node, multiple target resource nodes for executing the model optimization service are determined from multiple resource nodes; the node marking status includes at least a first node status and a second node status; resource nodes in the first node status and resource nodes in the second node status have different computing resources; the model optimization service is distributed to the service processing containers corresponding to the multiple target resource nodes, and the service processing containers are used to execute the model optimization service using the computing resources of the corresponding target resource nodes. It can be seen that this embodiment utilizes service parameters and the node marking status of each resource node to achieve flexible selection of resource nodes required for the model optimization service, and the computing resources of resource devices in different node states are different. This allows for the reasonable utilization of the computing resources of resource nodes in different node marking states to execute the model optimization service, achieving flexible allocation of computing resources for the model optimization service.
[0183] Please see Figure 7 This is a flowchart illustrating a business processing method provided in an embodiment of this application. The business processing method can be executed by a computer device, which can be a terminal device or a server; this embodiment of the application does not impose any limitations on this. The business processing method can be applied to a business processing system, which includes multiple resource nodes. The business processing method may include the following steps S701-S707:
[0184] S701. Obtain resource detection rules and business detection rules. Resource detection rules are used to detect resources in resource nodes from the resource layer, while business detection rules are used to detect resource nodes from the business layer based on business execution status.
[0185] Resource detection rules and business detection rules can be stored in a storage database (DB), and can be retrieved from the DB. These rules can be configured as needed. In another implementation, this embodiment can aggregate big data rules generated by each resource node at the resource and business layers. Big data rules refer to the large amount of abnormal data generated in each resource node, which can be used as rules for subsequent detection of the resource nodes.
[0186] Step 1: Obtain the load test data of the resource nodes and the business execution data of the resource nodes when they are executing business functions.
[0187] The load test data is obtained by performing stress tests on resource nodes when a concentrated number of errors occur. Specifically, if a resource node experiences more than a preset number of errors within a target time period (e.g., 1 hour, 10 minutes, 1 week, etc.), the load test data can be obtained. This load test data can include abnormal data from resource nodes, such as the graphics card experiencing 3 errors, business calculation failures 6 times, and data scheduling failures from memory 9 times within a certain time period.
[0188] Stress testing can include at least one or more of the following: performance stress testing of computing cards in resource nodes, bandwidth stress testing of communication between resource nodes, and scenario-based testing of open-source model optimization services running on resource nodes. In one implementation, these three types of stress tests can be performed on resource nodes in a hybrid manner. This hybrid approach covers the computing power service capabilities of the resource node, allowing for more comprehensive acquisition of stress test data and the selection of more suitable testing rules. Figure 8 The diagram shown is a flowchart illustrating a resource node stress test according to an embodiment of this application. Stress testing a resource node may include the following steps: a concentrated number of errors occur within the resource node (the number of errors reported by the resource node within a target time period exceeds a preset number); the configuration information of the resource node and the environmental information of the environment in which the resource node resides are scanned; a stress testing program is configured based on the configuration and environmental information; the stress testing program is run to perform stress testing on the resource node; a communication performance stress test is performed on the resource node; and stress testing data generated during the stress testing of the resource node is obtained. In one implementation, a stress testing generator can be called to perform stress testing on the resource node.
[0189] Furthermore, embodiments of this application may rely on big data information of resource nodes when performing business operations, or in other words, on big data information of nodes when performing computational tasks in business operations, to aggregate the different performance of certain resource nodes in different businesses or different computational tasks. Please see [link to relevant documentation]. Figure 9This diagram illustrates the execution status of a resource node under different services, as provided in an embodiment of this application. The process for obtaining service execution data includes the following steps: An anomaly occurs when a resource node executes service A (i.e., an anomaly occurs when service A hits the resource node). This anomaly can include: decreased computing performance of the resource node and / or a computing hang (a computing hang means that service A or the computing task within service A suddenly stops responding or cannot continue during execution). The resource node is then reassigned to another service for computation, and the resource node executes the other service normally. At this point, service execution data for both service A and service B can be obtained. In one implementation, a service generator can be called to obtain service execution data for different services. The service execution data includes both abnormal and normal data for each service execution. For example, the service execution data may include: the resource device's performance degrades when executing service A, but its performance is normal when executing service B. Alternatively, the service execution data may include the total number of failures for a certain service over a period of time.
[0190] Step 2: Based on the load test data and business execution data, determine at least one detection rule for detecting the node status of resource nodes.
[0191] In this application embodiment, the conditions for confirming a detection rule are defined. In one implementation, if the load test data contains data that meets a first condition, then the data meeting the first condition can be determined as a detection rule. The first condition includes: the load test data is obtained by performing a load test when the number of errors reported by resource nodes within a target time period exceeds a preset number, and the data is intermittent abnormal data during the load test. Intermittent abnormal data means that the number of times the abnormal data occurs within a preset time period is less than a preset number during the entire load test process. For example, the load test data is obtained by performing a load test when the number of errors reported by resource nodes within a target time period exceeds a preset number, and the load test data includes graphics card errors occurring on resource nodes, and these graphics card errors occur 1-2 times per week. In this case, the occurrence of a graphics card error can be used as one detection rule, or the occurrence of two graphics card errors can be used as one detection rule. In another implementation, if the business execution data contains data that meets a second condition, then the data meeting the second condition can be determined as a detection rule. For example, in business execution data: if a resource node experiences performance degradation when executing business A, but performs normally when executing business B, then the performance degradation of business A can be used as a detection rule.
[0192] Step 3: Separate at least one detection rule to obtain resource detection rules and business detection rules.
[0193] In one implementation, a rule logic processor can separate at least one detection rule to obtain resource detection rules and business detection rules. It should be understood that the reason for separating at least one detection rule is that the node marker status of resource nodes that hit a resource detection rule is the same, meaning that the computing power service capability or computing resources provided by the resource node are the same when serving different businesses. However, the node marker status of resource nodes that hit a business detection rule is different under different businesses, meaning that the computing power service capability or computing resources of the resource node are different when serving different businesses. For example, the computing power service capability of a resource node that hits a business detection rule will be lower under one business than under other businesses. For instance, if the node marker status of a resource node that hits a resource detection rule is the first node status, indicating that the resource node is in a sub-healthy state, then the computing power service capability or computing resources of this resource node are the same when serving different businesses—all are in a sub-healthy state. The node marker status of a resource node that hits the business detection rule is the first node status under business A. The first node status indicates that the resource node is in a sub-healthy state. Therefore, the computing power service capability or computing resources of this resource node when serving business A are in a sub-healthy state. However, the computing power service capability or computing resources of this resource node when serving other businesses (excluding business A) are in a healthy state.
[0194] In this application, the embodiments define the separation criteria for resource detection rules and service detection rules, as shown in Table 1:
[0195] Table 1
[0196]
[0197] Separating at least one detection rule to obtain resource detection rules and business detection rules may specifically include: identifying the detection rule determined from the stress test data as the resource detection rule; determining the number of abnormal services that occur when the resource node executes different services under each detection rule; identifying the detection rule corresponding to the number of abnormal services being greater than or equal to a preset number as the resource detection rule, and identifying the detection rule corresponding to the number of abnormal services being less than the preset number as the business detection rule. The preset number N can be set according to requirements, such as N being 2, 3, etc. For example, setting N to 2 can avoid situations where the resource node corresponds to multiple services. If the resource node experiences abnormalities in services A, B, and C, and all of these involve 18-20 GPU failures; under the rule regarding GPU failure counts, the number of abnormal services that occur when the resource node executes different services is 3, which is greater than the preset number, then the rule for 18 GPU failures is identified as the resource detection rule.
[0198] Optionally, resource detection rules and business detection rules can be stored in a storage database. Additionally, such as... Figure 3 As shown, resource detection rules and business detection rules support dynamic updates, such as updating them within a preset period. These updates include storing newly added resource detection rules and business detection rules in the storage database, and deleting outdated rules to prevent rule accumulation and operational anomalies.
[0199] S702. Perform status detection on each resource node in the business processing system according to the resource detection rules and business detection rules to obtain the node marking status of each resource node.
[0200] In one implementation, the node marking state includes a first node state and a second node state. If a first resource node hits a resource detection rule during business execution, then the node marking state of the first resource node is determined to be the first node state under different business types. If a second resource node does not hit a resource detection rule but hits a business detection rule during business execution, then the node marking state of the second resource node is determined to be the first node state under the business type to which the business belongs, but the node marking state of the second resource node is the second node state under other business types. If a third resource node does not hit a resource detection rule or a business detection rule during operation, then the node marking state of the third resource node is determined to be the second node state under different business types. Here, the first resource node, the second resource node, and the third resource node are resource nodes in the business processing system.
[0201] S703, Store the node tag status of each resource node.
[0202] Specifically, the node marking status of each resource node is used for subsequent scheduling of resource nodes. In one implementation, each resource node can be marked with a node marking status.
[0203] Optionally, the status of each resource node can be checked within a preset period to determine the latest node tag status of each resource node.
[0204] S704. Upon receiving a model optimization service, obtain the service parameters of the model optimization service;
[0205] S705. Based on business parameters, determine the node requirement status of resource nodes that support model optimization business.
[0206] S706. Based on the node demand status and the node marking status of each resource node, determine multiple target resource nodes from multiple resource nodes to perform model optimization business; the node marking status includes at least a first node status and a second node status; the computing power resources of resource nodes in the first node status and resource nodes in the second node status are different.
[0207] S707. Distribute the model optimization service to the service processing containers corresponding to multiple target resource nodes. The service processing containers are used to execute the model optimization service by utilizing the computing resources of the corresponding target resource nodes.
[0208] The specific implementation methods of steps S704-S707 can be found above. Figure 6 Steps S601-S604 in the above steps will not be described again here.
[0209] In this embodiment, abnormal data (i.e., business execution data and stress test data) during the operation of resource nodes can be collected, and resource detection rules and business detection rules can be identified and separated from the abnormal data. Then, the status of each resource node in the business processing system can be detected according to the resource detection rules and business detection rules to obtain the node marking status of each resource node. By using the two types of detection rules, resource detection rules and business detection rules, the deviation between resource detection and business anomalies can be filled, and the node marking status of each resource node can be accurately detected. This makes it easier to determine multiple target resource nodes for executing model optimization business from multiple resource nodes based on the node demand status and the node marking status of each resource node. This effectively avoids resource nodes under a certain node marking status (such as the first node status) from serving various businesses indiscriminately, provides a guarantee of computing power stability for model optimization business, improves business experience, and enhances the R&D efficiency of AI (i.e., AI model).
[0210] The following section will describe the business processing apparatus provided in the embodiments of this application.
[0211] Please see Figure 10 , Figure 10 This is a schematic diagram of a business processing apparatus provided in an embodiment of this application. The business processing apparatus can be a computer program (including program code) in a computer device; for example, the business processing apparatus can be application software in a computer device. The business processing apparatus can be used to execute... Figure 6 or Figure 7 Some or all of the steps in the method embodiments shown. Please refer to [link / reference]. Figure 10 The business processing device includes the following units:
[0212] The acquisition unit 1001 is used to acquire the service parameters of the model optimization service when the model optimization service is received;
[0213] Processing unit 1002 is used to determine the node demand status of resource nodes that support model optimization services based on business parameters.
[0214] The processing unit 1002 is also used to determine multiple target resource nodes for performing model optimization business from multiple resource nodes based on the node demand status and the node marking status of each resource node; the node marking status includes at least a first node status and a second node status; the computing power resources of resource nodes in the first node status and resource nodes in the second node status are different.
[0215] The processing unit 1002 is also used to distribute the model optimization business to the business processing containers corresponding to multiple target resource nodes. The business processing containers are used to execute the model optimization business by utilizing the computing power resources of the corresponding target resource nodes.
[0216] The node status of each resource node includes the node marking status of each resource node under different business types; when the processing unit 1002 determines multiple target resource nodes for performing model optimization business from multiple resource nodes based on the node demand status and the node marking status of each resource node, it can be specifically used for:
[0217] Based on the node marking status of each resource node under different business types, determine the node marking status of each resource node under the business type to which the model optimization business belongs;
[0218] Based on the node demand status and the node marking status of each resource node under the business type to which the model optimization business belongs, multiple target resource nodes are determined from multiple resource nodes to perform the model optimization business.
[0219] The business parameters include node status identifiers; when determining the node requirement status of resource nodes supporting model optimization services based on the business parameters, the processing unit 1002 can specifically be used for:
[0220] If the node status identifier is the identifier of the first node status, then the node requirement status of the resource node supporting the model optimization business is determined to be the first node status.
[0221] If the node status identifier is the identifier of the second node status, then the node requirement status of the resource node supporting the model optimization business is determined to be the second node status.
[0222] The business parameters include the priority of the model optimization business. When the processing unit 1002 determines the node requirement status of the resource nodes supporting the model optimization business based on the business parameters, it can be specifically used for:
[0223] If the priority of the model optimization service is lower than the preset priority, then the node requirement status of the resource node supporting the model optimization service is determined to be the first node status.
[0224] If the priority of the model optimization service is higher than the preset priority, then the node requirement status of the resource node supporting the model optimization service is determined to be the second node status.
[0225] Among them, the business parameters include the number of computing cards required for model optimization business, and the number of computing cards included in different resource nodes is different;
[0226] When processing unit 1002 determines the node requirement status of resource nodes supporting model optimization services based on business parameters, it can specifically be used for:
[0227] If the demand for computing power cards required for model optimization services is lower than the preset number, then the node demand status of the resource nodes supporting model optimization services is determined to be the first node status.
[0228] If the demand for computing power cards required for model optimization services is higher than or equal to the preset number, then the node demand status of the resource nodes supporting model optimization services is determined to be the second node status.
[0229] Among them, the business parameters include node status indication information, which is configured according to the interface parameters provided by the business processing system.
[0230] When processing unit 1002 determines the node requirement status of resource nodes supporting model optimization services based on business parameters, it can specifically be used for:
[0231] If the status indication information indicates that a resource node in the first node state is used, then the node requirement status of the resource node supporting the model optimization business is the first node state.
[0232] If the status indication information indicates that a resource node in the second node state is used, then the node requirement status of the resource node supporting the model optimization business is the second node state.
[0233] The acquisition unit 1001 is also used to: acquire resource detection rules and business detection rules; the resource detection rules are used to detect resources in resource nodes from the resource layer, and the business detection rules are used to detect resource nodes from the business layer using business execution status;
[0234] The processing unit 1002 is also used to perform status detection on each resource node in the business processing system according to the resource detection rules and the business detection rules, and obtain the node marking status of each resource node.
[0235] The processing unit 1002 is also used to store the node tag status of each resource node.
[0236] Specifically, when processing unit 1002 performs status detection on each resource node in the business processing system according to resource detection rules and business detection rules to obtain the node tag status of each resource node, it can be used for:
[0237] If a first resource node hits the resource detection rule during the execution of its business, then the node marking status of the first resource node under different business types is determined to be the first node status.
[0238] If a second resource node fails to meet the resource detection rule but meets the business detection rule during the execution of a business, then the second resource node is identified. The node marking status under the business type to which the business belongs is the first node status, but the node marking status under other business types is the second node status.
[0239] If a third resource node fails to meet either the resource detection rule or the business detection rule during business execution, then the node marking status of the third resource node under different business types is determined to be the second node status.
[0240] Among them, the first resource node, the second resource node, and the third resource node are resource nodes in the business processing system.
[0241] Specifically, when acquiring resource detection rules and business detection rules, the acquisition unit 1001 can be used for:
[0242] Obtain load test data and business execution data of resource nodes when they are performing business operations; load test data refers to the data obtained when a resource node experiences more errors than a preset number within a target time period.
[0243] Based on load test data and business execution data, determine at least one detection rule for detecting the node status of resource nodes;
[0244] At least one detection rule is separated to obtain resource detection rules and business detection rules.
[0245] Specifically, when the acquisition unit 1001 separates at least one detection rule to obtain resource detection rules and business detection rules, it can be used for:
[0246] Among at least one detection rule, the detection rule determined from the stress test data is identified as the resource detection rule;
[0247] Determine the number of services that cause anomalies when resource nodes execute different services under each detection rule;
[0248] The detection rules corresponding to the number of abnormal services being greater than or equal to a preset number are determined as resource detection rules, and the detection rules corresponding to the number of abnormal services being less than a preset number are determined as service detection rules.
[0249] In this embodiment, upon receiving a model optimization service, the service parameters of the model optimization service are obtained. Based on the service parameters, the node demand status of the resource nodes supporting the model optimization service is determined. Thus, based on the node demand status and the node marking status of each resource node, multiple target resource nodes for executing the model optimization service are determined from multiple resource nodes. The node marking status includes at least a first node status and a second node status. Resource nodes in the first node status and resource nodes in the second node status have different computing resources. The model optimization service is distributed to the service processing containers corresponding to the multiple target resource nodes. The service processing containers are used to execute the model optimization service using the computing resources of the corresponding target resource nodes. Therefore, this embodiment utilizes service parameters to achieve flexible selection of resource nodes required for the model optimization service. Furthermore, resource devices in different node states have different computing resources, which allows for the reasonable utilization of the computing resources of resource nodes in different node marking states to execute the model optimization service, achieving flexible allocation of computing resources for the model optimization service.
[0250] The computer device provided in the embodiments of this application will now be described.
[0251] Furthermore, this application also provides a schematic diagram of the structure of a computer device, which can be found in [reference needed]. Figure 11 The computer device may include a processor 1101, an input device 1102, an output device 1103, and a memory 1104. The processor 1101, input device 1102, output device 1103, and memory 1104 are connected via a bus. The memory 1104 stores computer programs, which include program instructions. The processor 1101 executes the program instructions stored in the memory 1104.
[0252] In one embodiment, the processor 1101 performs the following operations by running program instructions stored in memory 1104:
[0253] Upon receiving a model optimization request, obtain the service parameters of the model optimization request;
[0254] Based on business parameters, determine the node requirement status of resource nodes that support model optimization business;
[0255] Based on the node demand status and the node marking status of each resource node, multiple target resource nodes are determined from multiple resource nodes to perform model optimization business; the node marking status includes at least a first node status and a second node status; the computing power resources of resource nodes in the first node status and resource nodes in the second node status are different;
[0256] The model optimization service is distributed to the service processing containers corresponding to multiple target resource nodes. The service processing containers are used to execute the model optimization service by utilizing the computing resources of the corresponding target resource nodes.
[0257] The node status of each resource node includes the node marking status of each resource node under different service types; when the processor 1101 determines multiple target resource nodes for performing model optimization services from multiple resource nodes based on the node demand status and the node marking status of each resource node, it can specifically perform the following steps:
[0258] Based on the node marking status of each resource node under different business types, determine the node marking status of each resource node under the business type to which the model optimization business belongs;
[0259] Based on the node demand status and the node marking status of each resource node under the business type to which the model optimization business belongs, multiple target resource nodes are determined from multiple resource nodes to perform the model optimization business.
[0260] The business parameters include node status identifiers; when determining the node requirement status of resource nodes supporting model optimization services based on the business parameters, the processor 1101 can specifically perform the following operations:
[0261] If the node status identifier is the identifier of the first node status, then the node requirement status of the resource node supporting the model optimization business is determined to be the first node status.
[0262] If the node status identifier is the identifier of the second node status, then the node requirement status of the resource node supporting the model optimization business is determined to be the second node status.
[0263] Among them, the business parameters include the priority of the model optimization business. When the processor 1101 determines the node requirement status of the resource nodes that support the model optimization business based on the business parameters, it can specifically perform the following operations:
[0264] If the priority of the model optimization service is lower than the preset priority, then the node requirement status of the resource node supporting the model optimization service is determined to be the first node status.
[0265] If the priority of the model optimization service is higher than the preset priority, then the node requirement status of the resource node supporting the model optimization service is determined to be the second node status.
[0266] Among them, the business parameters include the number of computing cards required for model optimization business, and the number of computing cards included in different resource nodes is different;
[0267] When determining the node requirement status of resource nodes that support model optimization services based on business parameters, processor 1101 can specifically perform the following operations:
[0268] If the demand for computing power cards required for model optimization services is lower than the preset number, then the node demand status of the resource nodes supporting model optimization services is determined to be the first node status.
[0269] If the demand for computing power cards required for model optimization services is higher than or equal to the preset number, then the node demand status of the resource nodes supporting model optimization services is determined to be the second node status.
[0270] Among them, the business parameters include node status indication information, which is configured according to the interface parameters provided by the business processing system.
[0271] When determining the node requirement status of resource nodes that support model optimization services based on business parameters, processor 1101 can specifically perform the following operations:
[0272] If the status indication information indicates that a resource node in the first node state is used, then the node requirement status of the resource node supporting the model optimization business is the first node state.
[0273] If the status indication information indicates that a resource node in the second node state is used, then the node requirement status of the resource node supporting the model optimization business is the second node state.
[0274] The processor 1101 can also perform the following operations:
[0275] Obtain resource detection rules and business detection rules; resource detection rules are used to detect resources in resource nodes from the resource layer, and business detection rules are used to detect resource nodes from the business layer based on business execution.
[0276] Based on resource detection rules and business detection rules, the status of each resource node in the business processing system is detected to obtain the node marking status of each resource node.
[0277] Store the node tag status of each resource node.
[0278] Specifically, when the processor 1101 performs status detection on each resource node in the service processing system according to the resource detection rules and service detection rules, and obtains the node tag status of each resource node, it can perform the following operations:
[0279] If a first resource node hits the resource detection rule during the execution of its business, then the node marking status of the first resource node under different business types is determined to be the first node status.
[0280] If a second resource node fails to meet the resource detection rule but meets the business detection rule during the execution of a business, then the second resource node is identified. The node marking status under the business type to which the business belongs is the first node status, but the node marking status under other business types is the second node status.
[0281] If a third resource node fails to meet either the resource detection rule or the business detection rule during business execution, then the node marking status of the third resource node under different business types is determined to be the second node status.
[0282] Among them, the first resource node, the second resource node, and the third resource node are resource nodes in the business processing system.
[0283] Specifically, when acquiring resource detection rules and service detection rules, processor 1101 can perform the following operations:
[0284] Obtain load test data and business execution data of resource nodes when they are performing business operations; load test data refers to the data obtained when a resource node experiences more errors than a preset number within a target time period.
[0285] Based on load test data and business execution data, determine at least one detection rule for detecting the node status of resource nodes;
[0286] At least one detection rule is separated to obtain resource detection rules and business detection rules.
[0287] When the processor 1101 separates at least one detection rule to obtain resource detection rules and service detection rules, it may specifically perform the following operations:
[0288] Among at least one detection rule, the detection rule determined from the stress test data is identified as the resource detection rule;
[0289] Determine the number of services that cause anomalies when resource nodes execute different services under each detection rule;
[0290] The detection rules corresponding to the number of abnormal services being greater than or equal to a preset number are determined as resource detection rules, and the detection rules corresponding to the number of abnormal services being less than a preset number are determined as service detection rules.
[0291] It should be understood that, in the embodiments of this application, the processor 1101 may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.
[0292] The memory 1104 may include read-only memory and random access memory, and provides instructions and data to the processor 1101. A portion of the memory 1104 may also include non-volatile random access memory.
[0293] The input device 1102 may include a microphone, etc., and inputs the service parameters of the model optimization service into the processor 1101; the output device 1103 may include a display, etc.
[0294] In specific implementations, the processor 1101, input device 1102, output device 1103, and memory 1104 described in the embodiments of this application can execute the implementation methods described in all the above embodiments, or they can execute the implementation methods described in the above devices, which will not be repeated here.
[0295] In this embodiment, upon receiving a model optimization service, the service parameters of the model optimization service are obtained. Based on the service parameters, the node demand status of the resource nodes supporting the model optimization service is determined. Thus, based on the node demand status and the node marking status of each resource node, multiple target resource nodes for executing the model optimization service are determined from multiple resource nodes. The node marking status includes at least a first node status and a second node status. Resource nodes in the first node status and resource nodes in the second node status have different computing resources. The model optimization service is distributed to the service processing containers corresponding to the multiple target resource nodes. The service processing containers are used to execute the model optimization service using the computing resources of the corresponding target resource nodes. Therefore, this embodiment utilizes service parameters to achieve flexible selection of resource nodes required for the model optimization service. Furthermore, resource devices in different node states have different computing resources, which allows for the reasonable utilization of the computing resources of resource nodes in different node marking states to execute the model optimization service, achieving flexible allocation of computing resources for the model optimization service.
[0296] In this application, the term "unit" refers to a computer program or part of a computer program that has a predetermined function and works with other related parts to achieve a predetermined goal, and can be implemented wholly or partially using software, hardware (such as processing circuitry or memory), or a combination thereof. Similarly, a processor (or multiple processors or memory) can be used to implement one or more units. Furthermore, each unit can be part of an overall unit that includes the functionality of that unit.
[0297] Furthermore, it should be noted that this application also provides a computer-readable storage medium storing a computer program, which includes program instructions. When a processor executes these program instructions, it can execute the aforementioned... Figure 6 or Figure 7 The methods described in the corresponding embodiments are therefore not repeated here. For technical details not disclosed in the computer-readable storage medium embodiments related to this application, please refer to the description of the method embodiments of this application. As an example, program instructions may be deployed on a computer device, executed on multiple computer devices located in one location, or executed on multiple computer devices distributed in multiple locations and interconnected through a communication network.
[0298] According to one aspect of this application, a computer program product is provided, comprising a computer program stored in a computer-readable storage medium. A processor of a computer device reads the computer program from the computer-readable storage medium, and the processor executes the computer program, enabling the computer device to perform the aforementioned... Figure 6 or Figure 7 The methods described in the corresponding embodiments are therefore not repeated here.
[0299] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.
[0300] The above-disclosed embodiments are merely preferred embodiments of this application and should not be construed as limiting the scope of this application. Therefore, any equivalent variations made in accordance with the claims of this application shall still fall within the scope of this application.
Claims
1. A business processing method, characterized in that, Applied to a business processing system, the business processing system including multiple resource nodes, the method includes: Upon receiving a model optimization request, obtain the service parameters of the model optimization request; Based on the business parameters, determine the node demand status of the resource nodes that support the model optimization business; Based on the node demand status and the node marking status of each resource node, multiple target resource nodes for executing the model optimization business are determined from the multiple resource nodes; the node marking status includes at least a first node status and a second node status; the computing power resources of resource nodes in the first node status and resource nodes in the second node status are different; The model optimization service is distributed to the service processing containers corresponding to the multiple target resource nodes. The service processing containers are used to execute the model optimization service by utilizing the computing resources of the corresponding target resource nodes.
2. The method as described in claim 1, characterized in that, The node status of each resource node includes the node marking status of each resource node under different business types; the step of determining multiple target resource nodes for executing the model optimization business from the multiple resource nodes based on the node demand status and the node marking status of each resource node includes: Based on the node marking status of each resource node under different business types, determine the node marking status of each resource node under the business type to which the model optimization business belongs; Based on the node demand status and the node marking status of each resource node under the business type to which the model optimization business belongs, multiple target resource nodes for executing the model optimization business are determined from the multiple resource nodes.
3. The method as described in claim 1, characterized in that, The business parameters include node status identifiers; determining the node demand status of resource nodes supporting the model optimization business based on the business parameters includes: If the node status identifier is the identifier of the first node status, then the node requirement status of the resource node supporting the model optimization service is determined to be the first node status. If the node status identifier is the identifier of the second node status, then the node requirement status of the resource node supporting the model optimization service is determined to be the second node status.
4. The method as described in claim 1, characterized in that, The business parameters include the priority of the model optimization business. Determining the node demand status of resource nodes supporting the model optimization business based on the business parameters includes: If the priority of the model optimization service is lower than the preset priority, then the node demand status of the resource node supporting the model optimization service is determined to be the first node status. If the priority of the model optimization service is higher than the preset priority, then the node requirement status of the resource node supporting the model optimization service is determined to be the second node status.
5. The method as described in claim 1, characterized in that, The business parameters include the number of computing cards required for the model optimization business, and the number of computing cards included in different resource nodes is different; The step of determining the node demand status of resource nodes supporting the model optimization service based on the business parameters includes: If the number of computing cards required for the model optimization service is lower than the preset number, then the node demand status of the resource node supporting the model optimization service is determined to be the first node status. If the demand for computing power cards required for the model optimization service is higher than or equal to the preset number, then the node demand status of the resource node supporting the model optimization service is determined to be the second node status.
6. The method as described in claim 1, characterized in that, The business parameters include node status indication information, which is configured according to the interface parameters provided by the business processing system. The step of determining the node demand status of resource nodes supporting the model optimization service based on the business parameters includes: If the status indication information indicates that a resource node in the first node state is used, then the node requirement status of the resource node supporting the model optimization service is the first node state. If the status indication information indicates that a resource node in the second node state is used, then the node requirement status of the resource node supporting the model optimization service is the second node state.
7. The method according to any one of claims 1-6, characterized in that, The method further includes: Obtain resource detection rules and business detection rules; the resource detection rules are used to detect resources in resource nodes from the resource layer, and the business detection rules are used to detect resource nodes from the business layer based on business execution status; The status of each resource node in the business processing system is detected according to the resource detection rules and the business detection rules to obtain the node tag status of each resource node; Store the node tag status of each resource node.
8. The method as described in claim 7, characterized in that, The step of performing state detection on each resource node in the business processing system according to the resource detection rules and the business detection rules to obtain the node marker state of each resource node includes: If a first resource node hits the resource detection rule during the execution of a business process, then the node marking status of the first resource node under different business types is determined to be the first node status. If a second resource node fails to match the resource detection rule but matches the business detection rule during the execution of a business, then the node is identified as having the first node status under the business type to which the business belongs, but the node is identified as having the second node status under other business types. If a third resource node fails to match the resource detection rule and the business detection rule during business execution, then the node marking status of the third resource node under different business types is determined to be the second node status. The first resource node, the second resource node, and the third resource node are resource nodes in the business processing system.
9. The method as described in claim 7, characterized in that, The resource acquisition detection rules and business detection rules include: Obtain load test data and business execution data of resource nodes when they are performing business operations; the load test data refers to the data obtained by performing a load test on the resource node when the number of errors occurring on the resource node within a target time period exceeds a preset number. Based on the stress test data and the service execution data, determine at least one detection rule for detecting the node status of resource nodes; The at least one detection rule is separated to obtain resource detection rules and business detection rules.
10. The method as described in claim 9, characterized in that, The separation of the at least one detection rule to obtain resource detection rules and service detection rules includes: The detection rule determined from the stress test data is selected from the at least one detection rule and designated as the resource detection rule. Determine the number of services that cause anomalies when resource nodes execute different services under each detection rule; The detection rules corresponding to the number of abnormal services being greater than or equal to a preset number are determined as resource detection rules, and the detection rules corresponding to the number of abnormal services being less than a preset number are determined as service detection rules.
11. A business processing apparatus, characterized in that, The device is applied to a business processing system, which includes multiple resource nodes, and includes: The acquisition unit is used to acquire the service parameters of the model optimization service when the model optimization service is received; The processing unit is used to determine the node demand status of resource nodes that support the model optimization service based on the business parameters. The processing unit is further configured to determine, from the plurality of resource nodes, a plurality of target resource nodes for performing the model optimization service based on the node demand status and the node marking status of each resource node; the node marking status includes at least a first node status and a second node status; the resource nodes in the first node status and the resource nodes in the second node status have different computing resources. The processing unit is further configured to distribute the model optimization service to the service processing containers corresponding to the plurality of target resource nodes, wherein the service processing containers are configured to utilize the computing power resources of the corresponding target resource nodes to execute the model optimization service.
12. A computer device, characterized in that, include: A processor is used to execute computer programs; A computer-readable storage medium storing a computer program, which, when executed by the processor, performs the business processing method according to any one of claims 1-10.
13. A computer-readable storage medium, characterized in that, The computer storage medium stores a computer program, which, when executed by a processor, performs the business processing method according to any one of claims 1-10.
14. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the business processing method according to any one of claims 1-10.
