A business data processing method, apparatus, computer device, and storage medium

Abstract

The present invention discloses a method, apparatus, computer device and storage medium for processing service data; embodiments of the present application can obtain a service data processing request under a target service and a first service identifier of the target service; based on the first service identifier, determine a target service model mapping relationship corresponding to the target service from pre-stored service model mapping relationships, wherein the establishment of the pre-stored service mapping relationships can be implemented based on machine learning techniques in the field of artificial intelligence; if there is a corresponding relationship between the target model identifier information and the target version information in the first corresponding relationship, determine the access address of the target version sub-model based on the target model identifier information; based on the target version information and the access address, call the target version sub-model to process the to-be-processed service data through the target version sub-model. This solution can improve the processing efficiency of service data.

Description

Technical Field

[0001] This application relates to the field of communication technologies, and in particular, to a method and apparatus for processing service data, a computer device, and a storage medium. Background Art

[0002] With the rapid development of information technology, models can be deployed, and then various service data can be processed by running these deployed models.

[0003] In the process of researching and practicing related technologies, the inventors of this application found that the current model deployment technology includes relatively low-level model deployment tools, but the relatively low-level model deployment tools cannot perform large-scale model management and deployment, and the efficiency of model processing service data is relatively low. Summary of the Invention

[0004] Embodiments of this application provide a method and apparatus for processing service data, a computer device, and a storage medium, which can improve the processing efficiency of service data.

[0005] Embodiments of this application provide a method for processing service data, including:

[0006] Obtain a service data processing request under a target service, and a first service identifier of the target service, where the service data processing request includes service data to be processed, target model identifier information of a target data processing model, and target version information of a target version sub-model for processing the service data to be processed under the target data processing model;

[0007] Based on the first service identifier, determine a target service model mapping relationship corresponding to the target service from pre-stored service model mapping relationships, where the target service model mapping relationship includes a first correspondence between model identifier information of data processing models deployed under the target service and version information of at least one version sub-model deployed under each data processing model;

[0008] If there is a correspondence between the target model identifier information and the target version information in the first correspondence, determine an access address of the target version sub-model based on the target model identifier information;

[0009] Based on the target version information and the access address, call the target version sub-model to process the service data to be processed through the target version sub-model.

[0010] Correspondingly, embodiments of this application provide a service data processing apparatus, including:

[0011] An acquisition unit, configured to acquire a service data processing request under a target service, and a first service identifier of the target service, where the service data processing request includes to-be-processed service data, target model identifier information of a target data processing model, and target version information of a target version sub-model for processing the to-be-processed service data under the target data processing model;

[0012] A relationship determination unit, configured to determine, based on the first service identifier, a target service model mapping relationship corresponding to the target service from pre-stored service model mapping relationships, where the target service model mapping relationship includes a first corresponding relationship between model identifier information of data processing models deployed under the target service and version information of at least one version sub-model deployed under each data processing model;

[0013] An address determination unit, configured to, if there is a corresponding relationship between the target model identifier information and the target version information in the first corresponding relationship, determine an access address of the target version sub-model based on the target model identifier information;

[0014] An invocation unit, configured to invoke the target version sub-model based on the target version information and the access address, and process the to-be-processed service data through the target version sub-model.

[0015] In one embodiment, the acquisition unit includes:

[0016] A reception sub-unit, configured to receive a service data processing request under a target service through a model access interface of a model access management module, where the model access interface is an interface shared by at least two deployed services, and the deployed services include the target service.

[0017] In one embodiment, the address determination unit includes:

[0018] An address determination sub-unit, configured to determine an access address of a target container required to run the target data processing model based on the target model identifier information and the second corresponding relationship, and use it as the access address of the target version sub-model.

[0019] In one embodiment, the invocation unit includes:

[0020] A first invocation sub-unit, configured to invoke the target version sub-model in the target container based on the target version information and the access address of the target container, and process the to-be-processed service data through the target version sub-model.

[0021] In one embodiment, the invocation unit includes:

[0022] The second call subunit is used to access the model management module through the model, call the target version sub-model based on the target version information and the access address, and process the to-be-processed service data through the target version sub-model.

[0023] In one embodiment, the address determination unit further includes:

[0024] A receiving subunit, configured to receive a model deployment instruction, where the model deployment instruction includes: a second service identifier of a first service to which the to-be-deployed data processing model belongs, model identifier information of the to-be-deployed data processing model, version information of a to-be-deployed version sub-model under the to-be-deployed processing model and model file acquisition information, and container indication information of a first container for running the to-be-deployed version sub-model;

[0025] A first acquisition subunit, configured to acquire an access address of the first container based on the container indication information;

[0026] A loading subunit, configured to load the to-be-deployed version sub-model in the first container based on the model file acquisition information and the access address, to obtain a deployed version sub-model;

[0027] A first update subunit, configured to update the pre-stored service model mapping relationship based on the second service identifier of the first service, the model identifier information of the to-be-deployed data processing model, and the version information of the deployed version sub-model;

[0028] A second update subunit, configured to update the second corresponding relationship based on the access address of the first container and the model identifier information of the to-be-deployed data processing model.

[0029] In one embodiment, the first acquisition subunit is further configured to, if there is a deployed model version under the to-be-deployed data processing model, acquire the access address of the first container from a corresponding relationship between preset container identifier information and the access address of the container based on the container identifier information of the first container; if there is no deployed model version under the to-be-deployed data processing model, acquire the image file of the first container based on the file identifier information; create the first container based on the image file, and acquire the access address of the created first container.

[0030] In one embodiment, the address determination unit further includes:

[0031] A second acquisition subunit, configured to acquire a model version offline instruction, where the model version offline instruction includes: first version information of a to-be-offline version sub-model, and first model identifier information of a first data processing model to which the to-be-offline version sub-model belongs, and a third service identifier of a service to which the first data processing model belongs;

[0032] A third acquisition subunit, configured to acquire the access address of the second container of the sub-model of the version to be taken offline.

[0033] A first uninstallation subunit, configured to uninstall the sub-model of the version to be taken offline from the second container based on the access address of the second container and the first version information of the sub-model of the version to be taken offline.

[0034] A third update subunit, configured to update the pre-stored business model mapping relationship and the second corresponding relationship based on the uninstalled sub-model of the version to be taken offline.

[0035] In an embodiment, the address determination unit further includes:

[0036] A second uninstallation subunit, configured to uninstall the second container if the sub-model of the version to be taken offline is the last version sub-model of the first data processing model in the second container, and set the resources of the uninstalled second container as available idle resources.

[0037] A deletion subunit, configured to delete the first corresponding relationship between the model identification information of the first data processing model and the version information of the sub-model of the version to be taken offline based on the uninstalled second container, and delete the second corresponding relationship between the first data processing model and the access address of the second container.

[0038] A fourth update subunit, configured to update the pre-stored business model mapping relationship based on the deleted first corresponding relationship and the deleted second corresponding relationship.

[0039] Correspondingly, an embodiment of the present application further provides a computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the steps in any of the business data processing methods provided by the embodiments of the present application.

[0040] Correspondingly, an embodiment of the present application further provides a storage medium, which stores multiple instructions, and the instructions are suitable for being loaded by a processor to execute the steps in any of the business data processing methods provided by the embodiments of the present application.

[0041] This solution can deploy the data processing models corresponding to each service in a unified manner, establish the mapping relationship of service models under each service, and pre-store them. When a service data processing request under the target service is received, the access address of the target version sub-model for processing the service data to be processed under the target service can be obtained based on the pre-stored service model mapping relationship, so as to call the target version sub-model to process the service data to be processed. This can solve the trouble of each service independently managing the data processing model, and also eliminate the redundant work of the service in managing the deployment resources, and can improve the processing efficiency of service data. Brief Description of the Drawings

[0042] To more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present application. For those skilled in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0043] Figure 1 is a schematic diagram of the scenario of the service data processing method provided by the embodiment of the present application;

[0044] Figure 2a is a flowchart of the service data processing method provided by the embodiment of the present application;

[0045] Figure 2b is a technical framework diagram of the service data processing method provided by the embodiment of the present application;

[0046] Figure 2c is a model deployment page diagram of the service data processing method provided by the embodiment of the present application;

[0047] Figure 2d is a model call flowchart of the service data processing method provided by the embodiment of the present application;

[0048] Figure 2e is a schematic diagram of the container page of the service data processing method provided by the embodiment of the present application;

[0049] Figure 2f is a model version release schematic diagram of the service data processing method provided by the embodiment of the present application;

[0050] Figure 2g is a model version offline schematic diagram of the service data processing method provided by the embodiment of the present application;

[0051] Figure 2h is a container resource management framework diagram of the service data processing method provided by the embodiment of the present application;

[0052] Figure 3It is another flowchart of the service data processing method provided by the embodiments of the present application;

[0053] Figure 4a It is a device diagram of the service data processing method provided by the embodiments of the present application;

[0054] Figure 4b It is another device diagram of the service data processing method provided by the embodiments of the present application;

[0055] Figure 4c It is another device diagram of the service data processing method provided by the embodiments of the present application;

[0056] Figure 4d It is another device diagram of the service data processing method provided by the embodiments of the present application;

[0057] Figure 5 It is a schematic structural diagram of a computer device provided by the embodiments of the present application. Detailed implementation manners

[0058] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative efforts belong to the scope of protection of the present application.

[0059] The embodiments of the present application provide a service data processing method, device, computer device, and storage medium. Specifically, the embodiments of the present application provide a service data processing device applicable to a computer device. Among them, the computer device can be a device such as a terminal or a server. The server can be an independent physical server, or a server cluster or a distributed system composed of multiple physical servers. The terminal can be a smart phone, a tablet computer, a notebook computer, a desktop computer, a smart speaker, a smart watch, etc., but is not limited thereto. The terminal and the server can be directly or indirectly connected through wired or wireless communication methods, and the present application does not limit this here.

[0060] Refer to Figure 1, taking a computer device as an example of a server, the server can obtain a business data processing request under a target business, as well as a first business identifier of the target business. The business data processing request includes business data to be processed, as well as target model identifier information of a target data processing model, and target version information of a target version sub-model for processing the business data to be processed under the target data processing model; based on the first business identifier, determine, from the pre-stored business model mapping relationships, the target business model mapping relationship corresponding to the target business. The target business model mapping relationship includes a first corresponding relationship between model identifier information of data processing models deployed under the target business and version information of at least one version sub-model deployed under each data processing model; if there is a corresponding relationship between the target model identifier information and the target version information in the first corresponding relationship, determine, based on the target model identifier information, the access address of the target version sub-model; based on the target version information and the access address, call the target version sub-model to process the business data to be processed through the target version sub-model.

[0061] Among them, the establishment of the pre-stored business mapping relationship can be implemented based on machine learning (ML) technology in the field of artificial intelligence. For example, the target label can be determined from the label set, and then based on natural language processing technology, the target label set having a word inclusion relationship with the target label can be determined from the label set.

[0062] Among them, artificial intelligence (AI) is to utilize a digital computer or a machine model controlled by a digital computer to extend and expand human intelligence, sense the environment, acquire knowledge and use knowledge to obtain the best effect of theory, method, technology and application system. Artificial intelligence technology is a comprehensive discipline, involving a wide range of fields, including both hardware-level technologies and software-level technologies. Artificial intelligence software technologies mainly include directions such as natural language processing, machine learning / deep learning, etc.

[0063] Machine learning is a multi-disciplinary cross-discipline, involving multiple disciplines such as probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory, etc. It specializes in studying how a computer simulates or realizes human learning behaviors to acquire new knowledge or skills and reorganize the existing knowledge structure to continuously improve its own performance. Machine learning is the core of artificial intelligence and the fundamental way to make a computer intelligent, and its applications cover all fields of artificial intelligence. Machine learning and deep learning usually include technologies such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and rote learning.

[0064] As can be seen from the above, the embodiments of the present application can deploy the data processing models corresponding to each service uniformly, establish the service model mapping relationships under each service, and pre-store them. When a service data processing request under a target service is received, the access address of the target version sub-model for processing the service data to be processed under the target service can be obtained based on the pre-stored service model mapping relationships, so as to call the target version sub-model to process the service data to be processed, which can solve the trouble of each service independently managing the data processing model, and also save the redundant work of the service in managing the deployment resources, and can improve the processing efficiency of service data.

[0065] The present embodiment will be described in detail separately below. It should be noted that the description order of the following embodiments does not limit the preferred order of the embodiments.

[0066] The embodiments of the present application provide a service data processing method. This method can be executed by a terminal or a server, or jointly executed by a terminal and a server. The embodiments of the present application will be described by taking the service data processing method executed by the server as an example. Specifically, it is executed by a service data processing device integrated in the server. As Figure 2a shown, the specific process of this service data processing method can be as follows:

[0067] 201. Obtain a service data processing request under a target service, and a first service identifier of the target service. The service data processing request includes service data to be processed, target model identifier information of a target data processing model, and target version information of a target version sub-model for processing the service data to be processed under the target data processing model.

[0068] Among them, the service data processing request can be a request sent by a device of the target service. This service data processing request can be used to request a version sub-model under the corresponding target model to process the service data to be processed. For example, based on the target model identifier information and target version information included in the obtained service data processing request, it can be determined whether the target version sub-model corresponding to the target version information is deployed. After determining the deployment, the access address of the target version sub-model can be obtained, and based on this access address, the target version sub-model can be run, and the service data to be processed can be processed by the running target version sub-model.

[0069] Among them, the business data to be processed is the data that needs to be processed under the target business. The business data to be processed can be processed by the version sub-model under the target data processing model. The model for processing the business data to be processed can be determined according to the target model identification information included in the business data processing request, that is, the target version sub-model. When deploying each data processing model, there can be multiple version sub-models of the target data processing model. Therefore, in the process of processing the business data to be processed by the target version sub-model under the target data processing model, the target version sub-model under the target data processing model for processing the business data to be processed can be determined according to the target version information included in the business data processing request.

[0070] In one embodiment, the business data processing request under the target business can also be received through the model access interface of the model access management module. The step of "obtaining the business data processing request under the target business" can include:

[0071] Receiving the business data processing request under the target business through the model access interface of the model access management module, where the model access interface is an interface shared by at least two deployed services, and the deployed services include the target business.

[0072] In one example, as Figure 2b shown, for each business, the model can generally be uploaded and deployed through api / web. For example, it can support cephFS (Ceph File System, a distributed file system) for data sharing or through network transmission, and other methods can also be used. The specific usage method can be determined by the user. When uploading the model, it is necessary to specify the project to which the model belongs and the version of the model. The same model can only belong to one project, but can include multiple versions. The information related to the model can be saved in the mysql database. For example, one business can be equal to one project, and one business can include multiple projects. For example, for the ball game prediction business, it can include predicting the ball game in Brazil (one project) and predicting the game in Argentina (another project).

[0073] In one example, when deploying each data processing model, the model can be deployed in units of model versions to implement a three-layer architecture model management service of project-model-model version. Among them, the results of project-model-model version are displayed on the web as Figure 2c shown. Under one project, there can be multiple data processing models, and each data processing model can include multiple model versions, that is, each data processing model can include multiple version sub-models of different versions. Information such as the model name, model identification, and online status (such as online or offline) of each data processing model can also be displayed on the web page.

[0074] In one example, asFigure 2d As shown, after each data processing model is deployed, it can provide services through the inference interface. To avoid each business from maintaining metadata such as the service address of model deployment, a unified inference module can be provided. The inference module uniformly manages the metadata of the service addresses of model deployment, and its process can include: The device of the target business can send a business data processing request to the inference module. After receiving the request, the inference module can determine whether the target model under the model version corresponding to the target version information is deployed based on the target model identification information and the target version information included in the request. After determining the deployment, it can obtain the access address of the target version sub-model of the target data processing model corresponding to the target version information from the database or cache. Finally, based on this access address, the target version sub-model is run, and the business data to be processed is processed through the running target version sub-model.

[0075] In an example, each business can access through a unified interface and provide model_id (model identification information) and model version (version information) parameters during access, that is, the business data processing request can include model_id and model version parameters, and can send this business data processing request to the inference module. After receiving the request, the inference module will determine whether the current model version is deployed. After determining the deployment, it will obtain the actual address of model deployment from the database or cache, then request the inference service of the model in the way within the cluster, and finally return the result to the caller.

[0076] 202. Based on the first business identifier, determine the target business model mapping relationship corresponding to the target business from the pre-stored business model mapping relationships. The target business model mapping relationship includes the first correspondence between the model identification information of the data processing models already deployed under the target business and the version information of at least one version sub-model already deployed under each data processing model.

[0077] Among them, the first business identifier can represent the target business. According to the first business identifier in the pre-stored business model mapping relationships, the business mapping relationship with the first business identifier can be queried, and this business mapping relationship with the first business identifier is the target business model mapping relationship corresponding to the target business.

[0078] Among them, the pre-stored business model mapping relationships include the correspondence between the model identification information of each data processing model already deployed under each business and the version information of at least one version sub-model of each data processing model. When the version information of a model version is known, based on this correspondence, the data processing model corresponding to this version information and the model version can be determined.

[0079] Among them, the target data processing model may include version sub-models of multiple versions, and the target data processing model is a general term for multiple version sub-models.

[0080] In one embodiment, in order to unify the management and deployment of data processing models for each business and enable the upper-layer business to focus more on the business, the business data processing method may further include:

[0081] Receiving a model deployment instruction, where the model deployment instruction includes: a second business identifier of a first business to which the data processing model to be deployed belongs, model identifier information of the data processing model to be deployed, version information of the version sub-model to be deployed under the data processing model to be deployed, model file acquisition information, and container indication information of a first container for running the version sub-model to be deployed;

[0082] Based on the container indication information, obtaining the access address of the first container;

[0083] Based on the model file acquisition information and the access address, loading the version sub-model to be deployed in the first container to obtain the deployed version sub-model;

[0084] Based on the second business identifier of the first business, the model identifier information of the data processing model to be deployed, and the version information of the deployed version sub-model, updating the pre-stored business model mapping relationship;

[0085] Based on the access address of the first container and the model identifier information of the data processing model to be deployed, updating the second corresponding relationship.

[0086] Among them, the first container may be a docker container. All version sub-models of a model can be in the same container, and the access address of the first container can be stored. The Docker container is an open-source application container engine that allows developers to package their applications and dependent packages in a unified manner into a portable container and then publish it to any server installed with the docker engine.

[0087] Among them, if the first container has been created, the access address of the first container may be obtained from the pre-set corresponding relationship between the container identifier information and the access address of the container based on the container identifier information of the first container. If the first container has not been created yet, the mirror file of the first container may be obtained according to the file identifier information of the mirror file of the first container, and then the first container may be created based on the mirror file, and the access address of the created first container may be obtained.

[0088] In one example, the data processing model m may include version sub-model 1 and version sub-model 2. When the model is deployed, it is deployed in units of versions. One data processing model corresponds to one Docker container, and one Docker container may include models of multiple model versions corresponding to the data processing model. For example, one Docker container may include version sub-model 1 and version sub-model 2 of the data processing model m, and the deployed model version is specified through environment variables.

[0089] Among them, the specific configured environment configuration may be as Figure 2e shown. The version number is included in the environment variable, indicating that it has been released. If it is not in the environment variable, it means that it has been taken offline or the version does not exist. The content in box C1 may represent the container name of the creation server, that is, the container name of the Docker container. One container corresponds to one image, and the image becomes a container when it runs. When it runs, it has a name, just like a server has an IP. The content in box C2 may represent the name of the image. The image includes the code files required for running. The image can be understood as the operating system. The image is static, and when it runs, it becomes a container. The m in box C3 represents the model name of the data processing model in this container. The content in box C4 may represent the model path of the data processing model in this container. The 1 and 2 in box C5 represent two model versions under the data processing model in this container, and so on.

[0090] In one embodiment, the container indication information includes the container identification information of the first container and the file identification information of the image file of the first container. The step of "obtaining the access address of the first container based on the container indication information" may include:

[0091] If there is a deployed model version under the data processing model to be deployed, based on the container identification information of the first container, obtain the access address of the first container from the corresponding relationship between the preset container identification information and the access address of the container;

[0092] If there is no deployed model version under the data processing model to be deployed, then based on the file identification information, obtain the image file of the first container;

[0093] Create the first container based on the image file, and obtain the access address of the created first container.

[0094] In one embodiment, it can support the business to upload different versions of the model through the API and the web, manage different versions of the model in units of models, and can also perform model release, scale-up / down, version update, and take offline in units of model versions.

[0095] In one example, the deployment and management of the model are generated in the context of a self - controllable computing resource pool. The management of the resource pool can be based on Kubernetes, and different versions of the model correspond to a docker container.

[0096] 203. If there is a correspondence between the target model identification information and the target version information in the first correspondence, based on the target model identification information, determine the access address of the target version sub - model.

[0097] Among them, the access address of the target version sub - model can be used to access the target version sub - model, and based on this access address, the target version sub - model can be run. For example, the access address can be Figure 2b the address where the target version sub - model is stored in the tfscluster. After knowing the access address of the target version sub - model stored in the tfs cluster, the target version sub - model can be run to process the to - be - processed service data of the target business.

[0098] In one embodiment, the version sub - models under the same data - processing model share a container for running the model. The target business model mapping relationship further includes: a second correspondence between the model identification information of the data - processing model and the access address of the container for running the data - processing model. The step of "based on the target model identification information, determine the access address of the target version sub - model" may include:

[0099] Based on the target model identification information and the second correspondence, determine the access address of the target container required to run the target data - processing model, and use it as the access address of the target version sub - model.

[0100] 204. Based on the target version information and the access address, call the target version sub - model to process the to - be - processed service data through the target version sub - model.

[0101] Among them, the to - be - processed service data can be sent to the target version sub - model in the target container based on the access address, that is, based on the access address, the to - be - processed service data and the target sample information can be sent to the target container, triggering the target container to transfer the to - be - processed service data to the target version sub - model based on the target version information, so that the target version sub - model processes the to - be - processed service data.

[0102] In one embodiment, in order to process the to - be - processed service data, the corresponding target version sub - model can be run to process the to - be - processed service data. The specific step of "based on the target version information and the access address, call the target version sub - model to process the to - be - processed service data" may include:

[0103] Based on the target version information and the access address of the target container, call the target version sub-model in the target container, and process the business data to be processed through the target version sub-model.

[0104] In one embodiment, the step of "based on the target version information and the access address, call the target version sub-model, and process the business data to be processed through the target version sub-model" may include:

[0105] Through the model access management module, based on the target version information and the access address, call the target version sub-model, and process the business data to be processed through the target version sub-model.

[0106] In one embodiment, whenever a version sub-model under a data processing model is taken offline, the pre-stored business model mapping relationship and the second corresponding relationship can be updated. Specifically, the business data processing method may further include:

[0107] Obtain a model version offline instruction, where the model version offline instruction includes: the first version information of the version sub-model to be taken offline, the first model identification information of the first data processing model to which the version sub-model to be taken offline belongs, and the third business identification of the business to which the first data processing model belongs;

[0108] Obtain the access address of the second container of the version sub-model to be taken offline;

[0109] Based on the access address of the second container and the first version information of the version sub-model to be taken offline, uninstall the version sub-model to be taken offline from the second container;

[0110] Based on the uninstalled version sub-model to be taken offline, update the pre-stored business model mapping relationship and update the second corresponding relationship.

[0111] In one example, as Figure 2f shown, since the deployment of the data processing model is released in units of model versions, it is necessary to distinguish whether there are other versions released during the release. If there are other versions released, only the environment variables need to be updated, and then the container is restarted; if there are no other versions released, a container needs to be created instead of simply updating the environment variables.

[0112] Similarly, the model offline is also in units of versions, and the process during processing is a reverse process of the model version release. As Figure 2g shown, when the data processing model is taken offline, if it is the last version, the container can be recycled to update the status of the data processing model; if there are other versions online, the environment variables can be updated, and the version status of the data processing model can be updated.

[0113] In one embodiment, to ensure that resources can be fully utilized, resources can be recycled. After the step of "unloading the sub-model of the version to be taken offline from the second container based on the access address of the second container and the first version information of the sub-model of the version to be taken offline", the following steps may further be included:

[0114] If the sub-model of the version to be taken offline is the last version sub-model of the first data processing model in the second container, unload the second container, and set the resources of the unloaded second container as available idle resources;

[0115] Based on the unloaded second container, delete the first corresponding relationship between the model identification information of the first data processing model and the version information of the sub-model of the version to be taken offline, and delete the second corresponding relationship between the access address of the first data processing model and the second container;

[0116] Based on the deleted first corresponding relationship and the deleted second corresponding relationship, update the pre-stored business model mapping relationship.

[0117] In one example, as Figure 2h shown, for the computing resources required for model deployment, in order to better manage the entire life cycle of the computing resources, they can be abstracted here as a separate module. The main functions include:

[0118] (1) Parse out the latest image that meets the conditions according to the image parameters provided in the model deployment (models generated by tensorflow1.x and tensorflow2.x correspond to different images).

[0119] (2) Create resources, update containers, query status, and recycle.

[0120] Here, to facilitate the management of model deployment resources, when creating a service, the task_uuid of the model deployment will be stored as a foreign key in the database for creating resources, which is convenient for subsequent updates, queries, and recycling. All are associated through task_uuid to achieve low coupling in the interaction between modules.

[0121] The embodiments of the present application introduce a solution for unified management of trained data processing models, including management of data processing models and model versions, as well as the entire life cycle management of deployment, including functions such as allocation, update, scaling, and other services of resources required for deployment. It solves the trouble of each business independently managing data processing models and eliminates the redundant work of business in managing deployment resources. At the same time, it can also provide a unified external inference interface, eliminating the trouble for upper-layer services to manage inference addresses additionally, and helping upper-layer services focus more on business. After using the solution of the embodiments of the present application, the model management and deployment work of each business can be unified, enabling upper-layer services to focus more on business. The business side does not need to manage each version of the model and only needs to request the model inference service through a general interface during the inference stage. In addition, when deploying the model, the resource pool for model deployment is uniformly managed and allocated to ensure the full utilization of resources and save costs for the enterprise.

[0122] As can be seen from the above, the embodiments of the present application can deploy the data processing models corresponding to each business in a unified manner, establish the mapping relationship of business models under each business, and pre-store them. When receiving a business data processing request under the target business, based on the pre-stored business model mapping relationship, the access address of the target version sub-model for processing the business data to be processed under the target business can be obtained, so as to call the target version sub-model to process the business data to be processed, which can solve the trouble of each business independently managing data processing models and eliminate the redundant work of business in managing deployment resources, and can improve the processing efficiency of business data.

[0123] According to the above-introduced content, the following will give examples to further illustrate the business data processing method of the present application. Refer to Figure 3 , a business data processing method, and the specific process can be as follows:

[0124] 301. The server receives a business data processing request under the target business. The business data processing request includes the business data to be processed of the target business, the target model identification information of the target data processing model, and the target version information of the target version sub-model for processing the business data to be processed under the target data processing model.

[0125] In one example, taking the model access management module as the inference module as an example for illustration, as Figure 2dAs shown, each service can access through the model access interface of the model access management module, that is, the access interface of the inference module. The server can receive a service data processing request under the target service through this access interface of the inference module. The service data processing request may include model identification information and target version information of the target data processing model for processing the service data to be processed.

[0126] 302. The server determines the access address of the target version sub-model for processing the service data to be processed based on the corresponding relationship between the created services, the data processing models deployed under each created service, and the access addresses of at least one model version under each data processing model, as well as the target model identification and target model version information.

[0127] In one example, the corresponding relationship between the created services, the data processing models deployed under each created service, and the access addresses of at least one model version under each data processing model can be created and then pre-stored to obtain a pre-stored service model mapping relationship. When receiving a service data processing request under the target service, the access address of the target version sub-model for processing the service data to be processed can be determined based on the target model identification and target model version information included in the service data processing request and the pre-stored service model mapping relationship.

[0128] 303. The server calls the target version sub-model based on the target version information and the access address of the target version sub-model, and processes the service data to be processed through the target version sub-model.

[0129] In one example, the server can determine the access address of the target container required to run the target data processing model based on the target model identification information, as well as the second corresponding relationship between the model identification information of the data processing model and the access address of the container running the data processing model, and use it as the access address of the target version sub-model. Then, based on the target version information and the access address of the target container, the target version sub-model in the target container can be called to process the service data to be processed.

[0130] In one example, the server can also receive a service data processing request under the target service through the model access interface of the model access management module, and then, through the model access management module, call the target version sub-model based on the target version information and the access address.

[0131] Through the embodiments of the present application, both the management of data processing models and model versions can be completed, and the management of the entire life cycle of the deployment of data processing models can be supported. The specific problems to be solved may include:

[0132] (1) By analyzing and abstracting the model deployment requirements of each business, a set of general model / model version management systems can be formed, which well solves the problem of each business independently managing models / model versions.

[0133] (2) When managing models, a three-level architecture system of project -> model -> model version is formed, enabling the business to view and manage its own model / model version services by project.

[0134] (3) Model management provides two forms: web and api. Both can create projects, upload models / model versions, and provide functions such as model release, scale-up / scale-down, and offline. It supports large-scale model management and deployment.

[0135] (4) In cooperation with functions such as model release, scale-up / scale-down, and offline, a set of management services for the entire life cycle of computing resources is provided.

[0136] (5) In the inference stage, the service addresses of model deployments are uniformly managed, and a unified interface is provided externally, saving the redundant work of the upper-layer business side in managing deployed models.

[0137] As can be seen from the above, the embodiments of the present application can uniformly deploy the data processing models corresponding to each business, establish the business model mapping relationships under each business, and pre-store them. When a business data processing request under the target business is received, based on the pre-stored business model mapping relationships, the access address of the target version sub-model for processing the business data to be processed under the target business can be obtained, so as to call the target version sub-model to process the business data to be processed, which can solve the trouble of each business independently managing the data processing models and save the redundant work of the business in managing deployment resources, and can improve the processing efficiency of business data.

[0138] To better implement the above method, correspondingly, the embodiments of the present application further provide a business data processing device. Specifically, the business data processing device can be integrated in a server. Refer to Figure 4a , the business data processing device may include an acquisition unit 401, a relationship determination unit 402, an address determination unit 403, and a call unit 404, as follows:

[0139] (1) Acquisition unit 401;

[0140] The acquisition unit 401 is used to acquire a business data processing request under the target business, and a first business identifier of the target business. The business data processing request includes the business data to be processed, as well as the target model identifier information of the target data processing model, and the target version information of the target version sub-model for processing the business data to be processed under the target data processing model.

[0141] In one embodiment, as Figure 4b shown, the obtaining unit 401 includes:

[0142] A receiving subunit 4011, configured to receive a service data processing request under a target service through a model access interface of a model access management module, where the model access interface is an interface shared by at least two deployed services, and the deployed services include the target service.

[0143] (2) A relationship determination unit 402;

[0144] The relationship determination unit 402 is configured to determine a target service model mapping relationship corresponding to the target service from a pre-stored service model mapping relationship based on a first service identifier, where the target service model mapping relationship includes a first correspondence between model identifier information of a data processing model deployed under the target service and version information of at least one version sub-model deployed under each data processing model.

[0145] (3) An address determination unit 403;

[0146] The address determination unit 403 is configured to, if there is a correspondence between target model identifier information and target version information in the first correspondence, determine an access address of the target version sub-model based on the target model identifier information.

[0147] In one embodiment, as Figure 4c shown, the address determination unit 403 includes:

[0148] An address determination subunit 4031, configured to determine an access address of a target container required to run the target data processing model based on the target model identifier information and a second correspondence, and use it as the access address of the target version sub-model.

[0149] In one embodiment, as Figure 4c shown, the address determination unit 403 further includes:

[0150] A receiving subunit 4032, configured to receive a model deployment instruction, where the model deployment instruction includes: a second service identifier of a first service to which the data processing model to be deployed belongs, model identifier information of the data processing model to be deployed, version information of a version sub-model to be deployed under the data processing model to be deployed, model file acquisition information, and container indication information of a first container for running the version sub-model to be deployed;

[0151] A first acquisition subunit 4033, configured to acquire an access address of the first container based on the container indication information;

[0152] A loading subunit 4034, configured to obtain information and access addresses based on a model file, and load a sub-model of a version to be deployed in a first container to obtain a deployed sub-model of the version;

[0153] A first update subunit 4035, configured to update a pre-stored business model mapping relationship based on a second business identifier of a first business, model identifier information of a data processing model to be deployed, and version information of the deployed sub-model of the version;

[0154] A second update subunit 4036, configured to update a second corresponding relationship based on the access address of the first container and the model identifier information of the data processing model to be deployed.

[0155] In an embodiment, the first acquisition subunit 4033 is further configured to, if there is a deployed model version under the data processing model to be deployed, obtain the access address of the first container from the corresponding relationship between the preset container identifier information and the access address of the container based on the container identifier information of the first container; if there is no deployed model version under the data processing model to be deployed, obtain the image file of the first container based on the file identifier information; create the first container based on the image file, and obtain the access address of the created first container.

[0156] In an embodiment, as Figure 4c shown, the address determination unit 403 further includes:

[0157] A second acquisition subunit 4037, configured to obtain a model version offline instruction, where the model version offline instruction includes: first version information of a sub-model of a version to be taken offline, first model identifier information of a first data processing model to which the sub-model of the version to be taken offline belongs, and a third business identifier of a business to which the first data processing model belongs;

[0158] A third acquisition subunit 4038, configured to obtain the access address of a second container of the sub-model of the version to be taken offline;

[0159] A first uninstall subunit 4039, configured to uninstall the sub-model of the version to be taken offline from the second container based on the access address of the second container and the first version information of the sub-model of the version to be taken offline;

[0160] A third update subunit 40310, configured to update the pre-stored business model mapping relationship and update the second corresponding relationship based on the uninstalled sub-model of the version to be taken offline.

[0161] In an embodiment, the address determination unit 403 further includes:

[0162] A second uninstall subunit 40311, configured to, if the sub-model of the version to be taken offline is the last version sub-model of the first data processing model in the second container, uninstall the second container and set the resources of the uninstalled second container as available idle resources;

[0163] Delete the sub-unit 40312, which is used for the second container based on offloading, to delete the first corresponding relationship between the model identification information of the first data processing model and the version information of the sub-model to be taken offline, and delete the second corresponding relationship between the access addresses of the first data processing model and the second container;

[0164] The fourth update sub-unit 40313 is used to update the pre-stored business model mapping relationship based on the deleted first corresponding relationship and the deleted second corresponding relationship.

[0165] (4) Invocation unit 404;

[0166] The invocation unit 404 is used to invoke the target version sub-model based on the target version information and the access address, and process the to-be-processed service data through the target version sub-model.

[0167] In one embodiment, as Figure 4d shown, the invocation unit 404 includes:

[0168] The first invocation sub-unit 4041 is used to invoke the target version sub-model in the target container based on the target version information and the access address of the target container, and process the to-be-processed service data through the target version sub-model.

[0169] In one embodiment, as Figure 4d shown, the invocation unit 404 includes:

[0170] The second invocation sub-unit 4042 is used to invoke the target version sub-model through the model access management module based on the target version information and the access address, and process the to-be-processed service data through the target version sub-model.

[0171] As can be seen from the above, the acquisition unit 401 of the service data processing device according to the embodiment of the present application acquires a service data processing request under a target service, as well as a first service identifier of the target service. The service data processing request includes service data to be processed, as well as target model identifier information of a target data processing model and target version information of a target version sub-model for processing the service data to be processed under the target data processing model. Then, the relationship determination unit 402 determines, based on the first service identifier, a target service model mapping relationship corresponding to the target service from the pre-stored service model mapping relationships. The target service model mapping relationship includes a first correspondence relationship between model identifier information of data processing models deployed under the target service and version information of at least one version sub-model deployed under each data processing model. If there is a correspondence relationship between the target model identifier information and the target version information in the first correspondence relationship, the address determination unit 403 determines the access address of the target version sub-model based on the target model identifier information. The running unit 404 calls the target version sub-model based on the target version information and the access address, and processes the service data to be processed through the target version sub-model.

[0172] This solution can uniformly deploy the data processing models corresponding to each service, establish service model mapping relationships under each service, and pre-store them. When receiving a service data processing request under a target service, it is possible to obtain the access address of the target version sub-model for processing the service data to be processed under the target service based on the pre-stored service model mapping relationship, so as to call the target version sub-model to process the service data to be processed, which can solve the trouble of each service independently managing the data processing model and eliminate the redundant work of the service in managing deployment resources, and can improve the processing efficiency of service data.

[0173] In addition, the embodiment of the present application further provides a computer device, which can be a device such as a terminal or a server, as Figure 5 shown, which shows a schematic structural diagram of the computer device involved in the embodiment of the present application. Specifically:

[0174] The computer device may include a processor 501 with one or more processing cores, a memory 502 with one or more storage media, a power supply 503, an input unit 504, and other components. Those skilled in the art can understand that Figure 5 the structural diagram of the computer device shown in does not constitute a limitation on the computer device, and may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements. Among them:

[0175] The processor 501 is the control center of the computer device, connecting various parts of the entire computer device through various interfaces and circuits. By running or executing software programs and / or modules stored in the memory 502, and by invoking the data stored in the memory 502, it executes various functions of the computer device and processes data, thereby conducting an overall inspection of the computer device. Optionally, the processor 501 may include one or more processing cores; preferably, the processor 501 may integrate an application processor and a modem processor. Among them, the application processor mainly processes the operating system, user interface, application programs, etc., and the modem processor mainly processes wireless communications. It can be understood that the above-mentioned modem processor may not be integrated into the processor 501 either.

[0176] The memory 502 can be used to store software programs and modules. The processor 501 executes various functional applications and data processing by running the software programs and modules stored in the memory 502. The memory 502 may mainly include a program storage area and a data storage area. Among them, the program storage area can store the operating system, application programs required for at least one function (such as the sound playback function, image playback function, etc.), etc.; the data storage area can store the data created according to the use of the computer device. In addition, the memory 502 may include high-speed random access memory, and may also include non-volatile memory, such as at least one magnetic disk storage device, flash memory device, or other volatile solid-state storage devices. Correspondingly, the memory 502 may also include a memory controller to provide the processor 501 with access to the memory 502.

[0177] The computer device further includes a power supply 503 for supplying power to each component. Preferably, the power supply 503 can be logically connected to the processor 501 through a power management system, so as to realize functions such as management of charging, discharging, and power consumption management through the power management system. The power supply 503 may also include any components such as one or more DC or AC power supplies, a recharge system, a power failure detection circuit, a power converter or inverter, and a power status indicator.

[0178] The computer device may further include an input unit 504, which can be used to receive input digital or character information, and generate keyboard, mouse, joystick, optical or trackball signal inputs related to user settings and function controls.

[0179] Although not shown, the computer device may also include a display unit, etc., which will not be elaborated here. Specifically, in this embodiment, the processor 501 in the computer device will load the executable files corresponding to the processes of one or more application programs into the memory 502 according to the following instructions, and the processor 501 will run the application programs stored in the memory 502 to achieve various functions as follows:

[0180] Obtain a business data processing request under a target business, as well as a first business identifier of the target business. The business data processing request includes to-be-processed business data, target model identifier information of a target data processing model, and target version information of a target version sub-model for processing the to-be-processed business data under the target data processing model; based on the first business identifier, determine, from pre-stored business model mapping relationships, a target business model mapping relationship corresponding to the target business. The target business model mapping relationship includes a first corresponding relationship between model identifier information of data processing models deployed under the target business and version information of at least one version sub-model deployed under each data processing model; if there is a corresponding relationship between the target model identifier information and the target version information in the first corresponding relationship, determine an access address of the target version sub-model based on the target model identifier information; based on the target version information and the access address, call the target version sub-model to process the to-be-processed business data through the target version sub-model.

[0181] As can be seen from the above, the embodiments of the present application can deploy data processing models corresponding to each business uniformly, establish business model mapping relationships under each business, and pre-store them. When receiving a business data processing request under a target business, an access address of a target version sub-model for processing the to-be-processed business data under the target business can be obtained based on the pre-stored business model mapping relationship, so as to call the target version sub-model to process the to-be-processed business data. This can solve the trouble of each business independently managing data processing models and eliminate the redundant work of business deployment resource management, and can improve the processing efficiency of business data.

[0182] Those of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructions or by controlling related hardware through instructions. The instructions can be stored in a storage medium and loaded and executed by a processor.

[0183] For this reason, the embodiments of the present application provide a storage medium in which multiple instructions are stored. The instructions can be loaded by a processor to execute the steps in any of the business data processing methods provided by the embodiments of the present application. For example, the instructions can execute the following steps:

[0184] Obtain a business data processing request under a target business, as well as a first business identifier of the target business. The business data processing request includes to-be-processed business data, as well as a target model identifier information of a target data processing model and a target version information of a target version sub-model for processing the to-be-processed business data under the target data processing model; Based on the first business identifier, determine, from the pre-stored business model mapping relationships, a target business model mapping relationship corresponding to the target business. The target business model mapping relationship includes a first corresponding relationship between the model identifier information of the data processing models deployed under the target business and the version information of at least one version sub-model deployed under each data processing model; If there is a corresponding relationship between the target model identifier information and the target version information in the first corresponding relationship, determine the access address of the target version sub-model based on the target model identifier information; Based on the target version information and the access address, call the target version sub-model to process the to-be-processed business data through the target version sub-model.

[0185] Among them, the storage medium may include: read-only memory (ROM, Read Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or optical disc, etc.

[0186] Since the instructions stored in the storage medium can execute the steps in any of the business data processing methods provided in the embodiments of the present application, the beneficial effects that can be achieved by any of the business data processing methods provided in the embodiments of the present application can be realized. For details, see the previous embodiments and will not be elaborated here.

[0187] Among them, according to one aspect of the present application, a computer program product or a computer program is provided. The computer program product or the computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the business data processing method provided in the above-mentioned invention content and embodiments.

[0188] The above has introduced in detail a business data processing method, device, computer device and storage medium provided by the embodiments of the present application. Specific examples are used in this article to elaborate on the principle and implementation manner of the present application. The description of the above embodiments is only used to help understand the method and its core idea of the present application; at the same time, for those skilled in the art, according to the idea of the present application, there will be changes in the specific implementation manner and application scope. In summary, the content of this specification should not be construed as a limitation to the present application.

Claims

1. A method for processing service data, characterized in that, it includes: obtaining a service data processing request under a target service, and a first service identifier of the target service, where the service data processing request includes service data to be processed, target model identifier information of a target data processing model, and target version information of a target version sub-model for processing the service data to be processed under the target data processing model; based on the first service identifier, determining, from a pre-stored service model mapping relationship, a target service model mapping relationship corresponding to the target service, where the target service model mapping relationship includes a first correspondence relationship between model identifier information of data processing models deployed under the target service and version information of at least one version sub-model deployed under each data processing model; wherein, version sub-models under the same data processing model share a container of an operating model, and the target service model mapping relationship further includes: a second correspondence relationship between model identifier information of a data processing model and an access address of the container running the data processing model; if there is a correspondence relationship between the target model identifier information and the target version information in the first correspondence relationship, determining an access address of the target version sub-model based on the target model identifier information, including: determining an access address of a target container required to run the target data processing model based on the target model identifier information and the second correspondence relationship, and using it as the access address of the target version sub-model; invoking the target version sub-model based on the target version information and the access address, and processing the service data to be processed through the target version sub-model.

2. The method according to claim 1, characterized in that, the invoking the target version sub-model based on the target version information and the access address, and processing the service data to be processed through the target version sub-model includes: invoking the target version sub-model in the target container based on the target version information and the access address of the target container, and processing the service data to be processed through the target version sub-model.

3. The method according to claim 2, characterized in that, the method further includes: receiving a model deployment instruction, where the model deployment instruction includes: a second service identifier of a first service to which a data processing model to be deployed belongs, model identifier information of the data processing model to be deployed, version information of a version sub-model to be deployed under the data processing model to be deployed, model file acquisition information, and container indication information of a first container for running the version sub-model to be deployed; obtaining an access address of the first container based on the container indication information; loading the version sub-model to be deployed in the first container based on the model file acquisition information and the access address to obtain a deployed version sub-model; updating the pre-stored service model mapping relationship based on the second service identifier of the first service, the model identifier information of the data processing model to be deployed, and the version information of the deployed version sub-model. Update the second corresponding relationship based on the access address of the first container and the model identification information of the data processing model to be deployed.

4. The method according to claim 3, wherein, the container indication information includes the container identification information of the first container and the file identification information of the image file of the first container; the obtaining the access address of the first container based on the container indication information includes: if there is a deployed model version under the data processing model to be deployed, obtaining the access address of the first container from the preset corresponding relationship between the container identification information and the access address of the container based on the container identification information of the first container; if there is no deployed model version under the data processing model to be deployed, obtaining the image file of the first container based on the file identification information; creating the first container based on the image file and obtaining the access address of the created first container.

5. The method according to claim 4, wherein, the method further includes: obtaining a model version offline instruction, where the model version offline instruction includes: the first version information of the sub-model of the version to be taken offline, the first model identification information of the first data processing model to which the sub-model of the version to be taken offline belongs, and the third business identification of the business to which the first data processing model belongs; obtaining the access address of the second container of the sub-model of the version to be taken offline; unloading the sub-model of the version to be taken offline from the second container based on the access address of the second container and the first version information of the sub-model of the version to be taken offline; updating the pre-stored business model mapping relationship and the second corresponding relationship based on the unloaded sub-model of the version to be taken offline.

6. The method according to claim 5, wherein, after unloading the sub-model of the version to be taken offline from the second container based on the access address of the second container and the first version information of the sub-model of the version to be taken offline, the method further includes: if the sub-model of the version to be taken offline is the last version sub-model of the first data processing model in the second container, unloading the second container and setting the resources of the unloaded second container as available idle resources; deleting the first corresponding relationship between the model identification information of the first data processing model and the version information of the sub-model of the version to be taken offline, and deleting the second corresponding relationship between the first data processing model and the access address of the second container based on the unloaded second container; updating the pre-stored business model mapping relationship based on the deleted first corresponding relationship and the deleted second corresponding relationship.

7. The method according to claim 1, wherein, the obtaining the business data processing request under the target business includes: receiving the business data processing request under the target business through the model access interface of the model access management module, where the model access interface is an interface shared by at least two deployed services, and the deployed services include the target business; Based on the target version information and the access address, the target version sub-model is called, and the to-be-processed service data is processed by the target version sub-model, including: Through the model access management module, based on the target version information and the access address, the target version sub-model is called, and the to-be-processed service data is processed by the target version sub-model.

8. A service data processing device Characterized in that It includes: An acquisition unit, configured to acquire a service data processing request under a target service, and a first service identifier of the target service, where the service data processing request includes to-be-processed service data, a target model identifier information of a target data processing model, and target version information of a target version sub-model for processing the to-be-processed service data under the target data processing model; A relationship determination unit, configured to determine, based on the first service identifier, a target service model mapping relationship corresponding to the target service from pre-stored service model mapping relationships, where the target service model mapping relationship includes a first correspondence relationship between model identifier information of a data processing model deployed under the target service and version information of at least one version sub-model deployed under each data processing model; wherein, version sub-models under the same data processing model share a container of a running model, and the target service model mapping relationship further includes: a second correspondence relationship between model identifier information of a data processing model and an access address of a container running the data processing model; An address determination unit, configured to, if there is a correspondence relationship between the target model identifier information and the target version information in the first correspondence relationship, determine an access address of the target version sub-model based on the target model identifier information, including: determining an access address of a target container required to run the target data processing model based on the target model identifier information and the second correspondence relationship, as the access address of the target version sub-model; A call unit, configured to call the target version sub-model based on the target version information and the access address, and process the to-be-processed service data through the target version sub-model.

9. A computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, Wherein, When the processor executes the program, it implements the service data processing method according to any one of claims 1 to 7.

10. A storage medium Characterized in that The storage medium stores multiple instructions, and the instructions are suitable for being loaded by a processor to execute the service data processing method according to any one of claims 1 to 7.

11. A computer program product, including computer instructions Characterized in that When the computer instructions are executed by a processor, they implement the service data processing method according to any one of claims 1 to 7.

Images