Information credential multi-cloud side-end cooperative intelligent isolation domain scheduling method and device

Abstract

The invention relates to the technical field of credential and credential cloud platforms, in particular to a credential and credential multi-cloud edge-end cooperative intelligent isolation domain scheduling method and device, and the method comprises the steps: constructing a fault isolation domain through a standardized resource region for diversified science and technology product application classification; extracting effective data from resource application information submitted by a user based on a Kafka distributed event flow platform; and obtaining a deployed product resource layout sub-graph based on a Spark distributed memory calculation engine, combining the deployed product resource layout sub-graph with the effective data, performing intelligent scheduling on a to-be-deployed product based on a spectral clustering algorithm, and recommending an optimal fault isolation domain for multi-active deployment of the to-be-deployed product. Through combination of a clustering idea and edge calculation, real-time calculation and accurate recommendation are carried out at a user side, so that the scientific and technological products have a multi-activity operation framework better, the overall online period of the scientific and technological products is greatly shortened, and the intelligent and digital operation capability of a creative cloud platform is practically enhanced.

Description

Technical Field

[0001] The present application relates to the technical field of trusted cloud platform, and in particular to a trusted multi-cloud edge collaborative intelligent isolation domain scheduling method and device. Background Art

[0002] Current data center technology products face multi-cloud, multi-core, and multi-active deployments, and usually rely on technology product managers and resource planning administrators to manually determine the deployment locations of newly launched applications. This requires a series of evaluation and approval processes. At the same time, both product managers and resource planning administrators need to have an accurate understanding of the current deployment structure of the technology product.

[0003] Almost the entire chain of resource planning on traditional cloud platforms relies on human planning. Although relevant personnel have also conducted detailed and comprehensive evaluations based on the actual existing structure, facing the current reality of the booming development of new quality productivity and the flourishing of science and technology products, there are mainly two problems. First, the efficiency is low. Since a series of comprehensive evaluations are required by product managers, deployment personnel, resource planning administrators, etc. related to science and technology products, data centers usually connect the evaluation suggestions and conclusions of all relevant personnel through the work order process when new applications are deployed and launched. When a large number of new applications are launched, each person needs to handle and evaluate a large number of work orders, and sometimes the circulation even takes a month. Therefore, the efficiency of manual evaluation is relatively low and cannot meet the actual development needs; on the other hand, the personnel in the entire chain cannot guarantee that all the existing layouts of science and technology products are fully and accurately controlled in real time. For example, all new applications need to be launched through resource administrators, which brings a huge workload. At the same time, resource administrators need to accurately grasp the existing layouts of all products in the data center in real time. Therefore, in practice, the resource pool location of the final deployment will miss the multi-active deployment architecture. Summary of the invention

[0004] In order to overcome the shortcomings of the prior art, the present application provides a method and device for scheduling intelligent isolation domains in a coordinated manner among multiple clouds and edges. By combining clustering ideas and edge computing, it can perform real-time computing and accurate recommendations on the user side, enabling technological products to better have a multi-active operating architecture.

[0005] In a first aspect, the present application provides a method for scheduling intelligent isolated domains in a multi-cloud edge-end collaborative manner, the method comprising the following steps:

[0006] Use standardized resource areas classified for diverse technology product applications to build fault isolation domains; each computer room in the data center is divided into multiple standardized resource areas, each of which is composed of multiple rows of cabinets;

[0007] Extract valid data from the resource application information submitted by the user based on the Kafka distributed event stream platform; the valid data includes product information, submodule information, application information, application computing resource chip type and demand;

[0008] A resource layout subgraph of deployed products based on the Spark distributed memory computing engine is obtained, and the resource layout subgraph of deployed products is combined with the valid data, and intelligent scheduling is performed on the products to be deployed based on the spectral clustering algorithm, and an optimal fault isolation domain is recommended for its multi-active deployment.

[0009] In a possible implementation, a fault isolation domain is constructed by using standardized resource areas for application classification of diverse technology products, including the following steps:

[0010] Based on the layout comparison table, the constructed fault isolation domain is persistently written into the database for modeling, and all fault isolation domain layout diagrams are updated to the memory database; wherein, the HBase database is used, and the layout comparison table records the correspondence between each fault isolation domain and the computer room, data center, and region.

[0011] In a possible implementation, a point-of-study log is made in advance on the resource application side where the user submits resource application information to collect product information, submodule information, application information, and the type and demand of computing resource chips applied for.

[0012] In a possible implementation, a deployed product resource layout subgraph based on a Spark distributed memory computing engine includes the following steps:

[0013] By calling the API gateways of different cloud platforms, obtain the list of virtual machines of all deployed products on each fault isolation domain;

[0014] Read and clean the virtual machine Json data in the virtual machine list of all fault isolation domains, flatten the format to generate a virtual machine information configuration file, split the virtual machine information configuration file to generate a virtual machine fault isolation domain mapping file, and upload it to the distributed file system;

[0015] Processing the virtual machine fault isolation domain mapping file based on the Spark distributed memory computing engine to obtain a product fault isolation domain mapping file;

[0016] According to the product fault isolation domain mapping file, the constructed deployed product resource layout subgraph is stored in the HBase database; wherein, the deployed layout subgraph of each product in each fault isolation domain is recorded by setting a table.

[0017] In a possible implementation, intelligent scheduling is performed on the product to be deployed based on a spectral clustering algorithm to recommend an optimal fault isolation domain for its active-active deployment, including the following steps:

[0018] Each type of valid data collected is taken as a vertex, and its weight correlation is calculated with the deployed product resource layout subgraph;

[0019] According to the established fault isolation domain layout and the remaining resources in the current fault isolation domain, all modules are deployed in the same fault isolation domain of a product.

[0020] According to the priorities of the module application graph, the dual-center dual-active module application graph, the remote disaster recovery dual-active module application graph, and the resource layout sub-graph of deployed products, a weight correlation calculation is performed until the maximum weight is selected for each valid data, and the optimal fault isolation domain for the multi-active deployment of the deployed products is obtained.

[0021] In a possible implementation, the method further includes the following steps:

[0022] After deploying the optimal fault isolation domain for the product to be deployed, update the corresponding deployment information to the HBase database.

[0023] In a possible implementation, the method further includes the following steps:

[0024] The subgraph structure of the deployed products on each fault isolation domain is visualized by fault isolation domain.

[0025] In a second aspect, the present application provides a credible multi-cloud edge-end collaborative intelligent isolation domain scheduling device, the device comprising:

[0026] A construction module is used to construct a fault isolation domain using standardized resource areas classified for diversified technology product applications; wherein each computer room in the data center is divided into a plurality of standardized resource areas, and each standardized resource area is composed of a plurality of rows of cabinets;

[0027] An extraction module is used to extract valid data from the resource application information submitted by the user based on the Kafka distributed event stream platform; the valid data includes product information, submodule information, application information, application computing resource chip type and demand;

[0028] The scheduling module is used to obtain a deployed product resource layout subgraph based on the Spark distributed memory computing engine, and combine the deployed product resource layout subgraph with the valid data, perform intelligent scheduling on the deployed product based on the spectral clustering algorithm, and recommend the optimal fault isolation domain for its multi-active deployment.

[0029] In the third aspect, the present application provides an electronic device comprising: a processor, a memory and a bus, the memory storing machine-readable instructions executable by the processor, when the electronic device is running, the processor and the memory communicate through the bus, and when the machine-readable instructions are executed by the processor, the steps of the trusted innovation multi-cloud edge collaborative intelligent isolation domain scheduling method as described in any one of the first aspects are performed.

[0030] In a fourth aspect, the present application provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the method for coordinated intelligent isolation domain scheduling of multi-cloud and edge terminals as described in any one of the first aspects are executed.

[0031] The present embodiment provides a method and device for scheduling intelligent isolation domains for multi-cloud and edge-end collaboration, which utilizes standardized resource areas classified for diversified technology product applications to construct a fault isolation domain; wherein each computer room in the data center is divided into multiple standardized resource areas, and each standardized resource area is composed of multiple columns of cabinets; based on the Kafka distributed event stream platform, valid data is extracted from the resource application information submitted by the user; the valid data includes product information, sub-module information, application information, the type and demand of the computing resource chip applied for; a deployed product resource layout subgraph based on the Spark distributed memory computing engine is obtained, and the deployed product resource layout subgraph is combined with the valid data, and the deployed products are intelligently scheduled based on the spectral clustering algorithm to recommend the optimal fault isolation domain for its multi-active deployment. It uses the idea of ​​spectral clustering algorithm, combined with the specific data in the resource application information submitted by users or product managers, and the information subgraphs of different resource pools of existing technology products, to realize real-time edge computing when resources are applied, and accurately and quickly promote the appropriate resource pool for the launch of new applications. This greatly improves the accuracy of the multi-active deployment architecture of technology products, greatly shortens the overall launch cycle of technology products, and at the same time provides accurate data basis for personnel evaluation in related processes, effectively enhancing the intelligent digital operation capabilities of the Xinchuang Cloud Platform. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0033] Figure 1 A flow chart of the method for scheduling intelligent isolated domains in a multi-cloud edge-end collaboration described in an embodiment of the present application is shown;

[0034] Figure 2An architecture diagram of the trusted multi-cloud edge-end collaborative intelligent isolation domain scheduling system described in an embodiment of the present application is shown;

[0035] Figure 3 A flowchart showing an embodiment of the present application for intelligently scheduling products to be deployed based on a spectral clustering algorithm and recommending an optimal fault isolation domain for their multi-active deployment is shown;

[0036] Figure 4 A deployment flow chart of each module in an embodiment of the present application is shown;

[0037] Figure 5 A schematic diagram of the structure of the intelligent isolation domain scheduling device for multi-cloud edge collaboration described in an embodiment of the present application is shown;

[0038] Figure 6 A structural block diagram of an electronic device described in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0039] To make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. It should be understood that the drawings in the present application only serve the purpose of explanation and description and are not used to limit the scope of protection of the present application. In addition, it should be understood that the schematic drawings are not drawn in real proportion. The flowchart used in this application shows the operations implemented according to some embodiments of the present application. It should be understood that the operations of the flowchart can be implemented out of sequence, and the steps without logical context can be reversed in order or implemented simultaneously. In addition, those skilled in the art can add one or more other operations to the flowchart under the guidance of the content of the present application, or remove one or more operations from the flowchart.

[0040] In addition, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. The components of the embodiments of the present application described and shown in the drawings here can be arranged and designed in various configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application claimed for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present application.

[0041] It should be noted that the term "comprising" will be used in the embodiments of the present application to indicate the existence of the features declared thereafter, but does not exclude the addition of other features.

[0042] In view of the technical problems raised by the background technology, the present application provides a method and device for scheduling intelligent isolated domains in a coordinated manner among multiple clouds and edges. By combining clustering ideas and edge computing, when a user submits an application for deployment of a technological product, the application performs real-time calculation and accurate recommendation on the user side based on the distribution of existing multi-resource pools and details of new requirements, so that technological products can better have an operating architecture with dual active-active in the same center, dual active-active in the same city, and multiple active-active in different locations, thus achieving truly non-stop external services.

[0043] In one embodiment, see the attached specification Figure 1 The present application provides a method for scheduling intelligent isolated domains in a multi-cloud edge-end collaborative manner, the method comprising the following steps:

[0044] S1. Use standardized resource areas classified for diversified technology product applications to build fault isolation domains; wherein each computer room in the data center is divided into multiple standardized resource areas, each of which is composed of multiple rows of cabinets;

[0045] S2. Extract valid data from the resource application information submitted by the user based on the Kafka distributed event stream platform; the valid data includes product information, submodule information, application information, application computing resource chip type and demand;

[0046] S3. Obtain a deployed product resource layout subgraph based on the Spark distributed memory computing engine, combine the deployed product resource layout subgraph with the valid data, perform intelligent scheduling on the products to be deployed based on the spectral clustering algorithm, and recommend an optimal fault isolation domain for its multi-active deployment.

[0047] See the instruction manual Figure 2 This application provides a method for scheduling intelligent isolated domains in a multi-cloud edge-end collaboration, and deploys its architecture in the actual production environment of a large insurance company's data center. After the traditional cloud platform resource application is submitted, the method fully realizes intelligent optimal computing recommendation scheduling. After the user submits a resource application, the real-time edge computing module deployed on the front end of the cloud management first extracts valid data such as product information, sub-module information, application information, resource type (mainly the type of chip for the application computing resource), and resource quantity, and then combines the layout sub-graphs of all technological products in the data center in different resource pools constructed in the early stage through the spectral clustering algorithm. The spectral clustering algorithm is used again to classify and segment the sub-graphs of new resources, and provides accurate scheduling recommendations for the multi-active deployment of new resources in an intelligent and data-based manner.

[0048] Specifically, in step S1, it involves the layout of the persistence layer construction module of the intelligent isolation domain. This is because the current data center is facing the rapid development of information and innovation quality productivity. The traditional basic platform architecture can no longer meet the user's high continuity and high persistence service needs. Therefore, this application proposes a standardized resource area (referred to as MiniDC) for the classification of diversified technology product applications, which is constructed into an intelligent fault isolation domain, referred to herein as a "minimalist distributed architecture". This architecture divides each computer room in the data center into M standardized resource areas, each of which is composed of N rows of cabinets. Each area can be used as an independent sub-data center with complete network isolation and access to the global data center. In one embodiment, according to the actual project, four columns are used as a standardized resource area, and a computer room has a total of three standardized resource areas, which correspond to different application functions.

[0049] Based on the intelligent fault isolation domain formed by the minimalist distributed architecture, each technological product can deploy at least one independent service in different computer rooms in the same center, different computer rooms in different centers in the same city, and different computer rooms in different centers in different locations. A unified load balancing can be used externally to flexibly schedule according to the access distance, which can achieve the deployment architecture and service conditions of dual active in the same center, dual active in the same city, and multi-active in different locations without stopping service. Therefore, before the intelligent scheduling of the newly launched application is performed in this application, the intelligent fault isolation domain formed by the minimalist distributed architecture must be persistently written to the database for modeling. Considering the large amount of data and subsequent real-time calculations, the HBase database is used for storage in this application.

[0050] HBase is a typical non-relational database that can be built on HDFS (distributed file system). The data it stores is between mapping and relational data, which can be understood as a mapping relationship between key and value. The HBase table has several attributes: row key (RowKey), column family (ColumnFamily), column (Column) and timestamp (TimeStamp). In this application, the table MinidcGraphDictionary is used to record the correspondence between each fault isolation domain and the computer room, data center, and region. Its RowKey value is the Minidc name minidcName. The column family minidcInfo records the region (Location), data center (DcRegion), computer room (ModuleNum), the remaining allocatable CPU memory storage of the first type of computing resource pool (kunpeng_XCXTXT), and the remaining allocatable CPU memory storage of the second type of computing resource pool (hugon_XCXTXT). The first deployment reads the Minidc layout comparison table and writes it to the database. After deployment, the real-time update module refreshes the table data. For specific table structure, see

[0051] Table 1.

[0052]

[0053] (Table 1)

[0054] At the same time, in order to ensure the real-time performance of edge computing in this application, all MiniDC layout graphs need to be minidcX Update to the memory database. In this application, Redis is used as the cache database. The Key stores minidcName, i.e., the region information, and the Value stores the resource information in minidcInfo. The specific storage format is shown in Table 2.

[0055]

[0056] (Table 2)

[0057] In step S2, it involves the construction module of the resource layout subgraph of deployed technology products based on the Spark distributed memory computing engine. This is because usually before the deployment of the intelligent isolation domain scheduling application, after the fault isolation domain MiniDC is built, some technology products will be deployed in advance. Therefore, for the existing deployed products, it is necessary to first perform the basic construction of the resource layout subgraph. Usually in the enterprise data center, the host names of the virtual machines that have been deployed and put online need to be specified according to the naming constraints. The data center where this application is deployed agrees that the naming specification of the online virtual machine host name is: host name = product abbreviation + sub-application abbreviation + function abbreviation + number. Since this application applies to more than 100 fault isolation domain MiniDCs in two locations and three centers of the data center, and each MiniDC has deployed a large number of technology product application virtual machines, this module uses the Spark distributed memory computing engine in the calculation process. First, by calling the ApiGateway gateway of different cloud platforms, a list of all existing virtual machines on each MiniDC is obtained. The partial display format of the data obtained by one of the MiniDCs is as follows:

[0058]

[0059]

[0060] (Table 3)

[0061] Through an example virtual machine record in Table 3, where "region_name" is the MiniDC name, you can see that the required host name, fault isolation domain MiniDC name, and tenant resource set where the virtual machine is located can be captured in the list of all virtual machines in each MiniDC.

[0062] This module first reads and cleans all the virtual machine Json data obtained under Minidc, and discards the virtual machine information that lacks host name information and does not conform to the naming specification. This part is usually a test machine or an erroneous creation that needs to be rectified later. If it is included, it will form dirty data that interferes with the construction of the subgraph. All Json data is pre-processed using the Spark distributed memory computing engine. It is flattened according to the format of Table 4 to generate VmInfoProFile. After uploading it to HIDS (distributed file system), VmInfoProFile is used as input, and each row of hostname is sliced. According to Table 5, the product, module, application and other information of each virtual machine are split to generate VmMappingMinidcFile, and the results are uploaded to HIDS (distributed file system) again.

[0063] hostname|minidcName

[0064] (Table 4)

[0065] productName|moduleName|appName|minidcName

[0066] (Table 5)

[0067] Sample data generated according to Table 5:

[0068] CSRS|IAS|Nginx|Comprehensive Services

[0069] (Table 6)

[0070] Since this application faces more than a hundred data centers, the preprocessing module adopts the Spark distributed memory computing engine to exponentially improve data computing and preprocessing capabilities. Spark Core is based on the MapReduce framework in traditional Hadoop distributed computing and has been improved and optimized. Compared with traditional distributed computing, each operation stage requires reading files from HDFS. Spark Core stores metadata and intermediate data in memory first, unless the memory is insufficient, it will replace part of the data on the disk; in addition, Spark no longer divides each data parsing process into Map and Reduce stages, but extracts the program logic through the Driver module in Spark Core, and builds a directed acyclic graph (Build DAG) of program dependencies, and then calls the DAGScheduler module to cut the graph into different Stages (Split DAG to Stage) and distribute it to each node for distributed computing, thereby greatly improving the computing efficiency.

[0071] The input data of the preprocessing module in this application is the HDFS file VmMappingMinidcFile after the flattening of the collected virtual machine data, and its output is an HDFS file ProductMappingMinidcFile formed by the relationship diagram of technology products, submodules, application modules and fault isolation domain Minidc. The specific file format is shown in Table 7:

[0072] productName-moduleName-minidcName-appName countNum

[0073] (Table 7)

[0074] According to the data result ProductMappingMinidcFile, the constructed stock technology product resource layout subgraph needs to be stored in the HBase database. In this application, the table ProductMappingMinidcGraphy is used to record the deployed layout subgraph of each technology product in each Minidc. Its RowKey value is the product name and submodule name. The column family productInfo mainly records the Minidc where the submodule corresponding to the technology product has been deployed and the application name. The table structure is shown in Table 8:

[0075]

[0076] (Table 8)

[0077] In addition, step S2 also involves a demand data collection module based on the Kafka distributed event stream platform. Specifically, this module mainly makes point logs on the resource application side in the work order system. When a user enters the work order system to apply for cloud platform resources, its product information, submodule information, application information, application computing resource chip type, and demand are collected. See Table 9 for details:

[0078] (Table 9)

[0079] In this application, the tracking program will collect ProductName, ModuleName, AppName, ChipType, and ApplyValue in the technology product resource application work order submitted by the user. The above fields are all judged not to be empty in the front-end tracking. If they are empty, the application work order is illegal. Since this application needs to intelligently calculate the allocable scheduling isolation domain based on the input data on the user resource application side in real time, real-time intelligent calculation and processing are required for the collected tracking data. The Flink stream computing engine in big data real-time computing is used here. Flink is mainly used to process streaming data, that is, "data flow" (Data Flow). That is, the source data is not collected regularly, but arrives one by one and is processed one by one like a stream of water. In order to ensure the efficiency and real-time of the entire collection and calculation process, the data collected by the tracking point in this application no longer uses traditional logs or interface collection, but adopts data collection preprocessing based on the Kafka distributed event stream platform. Kafka is a distributed message queue (MessageQueue) based on the publish / subscribe mode, which is mainly used in the field of big data real-time processing. There are three main roles in the information collection and preprocessing process: Producer, Topic, and Consumer.

[0080] Specifically, this module plays the role of the producer. In this application, the producer is the application information for technology product resources that the user fills in in real time in the work order system. After the user completes the filling in the front end and clicks the next step to confirm that the information data is correct, the embedding program will immediately publish a message of the Map data structure to the Kafka cluster. Its Key is "productName_moduleName_appName_chipType" and its Value is "applyValue". If the user applies for multiple product resources, multiple sub-module resources, multiple application resources, or multiple chip types in the same batch, it will be split into multiple messages according to the application items, and each Key value combination will generate a message. A topic can be understood as a queue. The producer and consumer are facing the same Topic, and different consumers only subscribe to messages in the Topic they are interested in. This application only involves one information collection Topic. When the producer receives the information submitted by the user, it will publish the message to the topic queue. The Flink distributed computing cluster can subscribe to the application message from the topic queue for parallel real-time stream computing. The consumer in this application is the Flink real-time intelligent isolation domain dynamic scheduling edge computing module based on the spectral clustering algorithm.

[0081] In step S3, the Flink real-time intelligent isolation domain dynamic scheduling edge computing module based on the spectral clustering algorithm is involved. Specifically, the main idea of ​​the spectral clustering algorithm is to regard all data as points in space, and then define the edge weights between different points in combination with the actual weight calculation mode. If the weight is 0, it means that there is no correlation between the two points according to the current calculation dimension, and there is no edge connection between the two points; otherwise, the larger the weight, the stronger the correlation and the closer the distance, and the smaller the weight, the weaker the correlation and the farther the distance. Finally, the graph is cut. The principle of cutting the graph is that the weight in the subgraph is the largest and the weight between subgraphs is the smallest; that is, the sum of the weights in all subgraphs is the largest, and the sum of the weights between all subgraphs is the smallest. Common spectral clustering algorithms may usually use polynomial kernel functions, Gaussian kernel functions, and Sigmoid kernel functions for weight calculation. For the graph cutting scheme, the n eigenvalues ​​and n eigenvectors of the Laplacian matrix composed of the weight graph are usually calculated to finally cluster the point set.

[0082] This application combines the actual needs and takes the key of each message of the collected needs as a vertex. At the same time, it needs to calculate the weight correlation with the constructed current technology product resource layout subgraph. Combined with the constructed intelligent isolation domain layout, based on the fact that the current MiniDC remaining resources can meet the demand resources, all modules are deployed in the same MiniDC of a certain technology product in turn. According to the module application diagram, the dual-center dual-active module application diagram, the remote disaster recovery dual-active module application diagram, and the priority of the same-city dual-active module application diagram, the correlation weight calculation is performed with the existing MiniDC and the technology product subgraph, until the maximum weight is finally selected for each message, and the fault isolation domain MiniDC with the best multi-active deployment of technology products is selected.

[0083] In addition, the Flink real-time computing module used in this application mainly relies on its core stream processing interface set DataStreamAPI when processing the Topic unbounded data stream (Unbounded Data Stream) delivered in real time by the Kafka producer. The DataStream API is embedded with the processing function (Process Function); the underlying processing function (ProcessFunction) is integrated with the DataStream API, which can abstract the operation and allow users to use custom states to process events from one or more data streams, and the state has consistency and fault tolerance guarantees. Flink cluster processing jobs (Task) mainly rely on several key components: client (Client), job manager (JobManager) and task manager (TaskManager). JobManager performs central scheduling management of jobs, TaskManager performs real-time calculations on jobs, and Client is mainly used in session mode (Session Mode) deployment and single job mode (Per Job Mode) deployment. The code is submitted to JobManager through Client. The Flink cluster deployment mode used in this application is the Application Mode. In this mode, jobs are no longer submitted through the Client, which greatly reduces the resource consumption of the node where the client is located. Instead, a JobManager is established for each task, and the job is directly submitted to its corresponding JobManager, avoiding multiple jobs from occupying JobManager resources, which greatly improves the computing efficiency. For details, see the attached manual. Figure 3 The implementation of the Flink real-time intelligent isolation domain dynamic scheduling edge computing module based on the spectral clustering algorithm is divided into three parts:

[0084] (1) Graph Construction minidcX-ProducX Technology product resource layout sub-map:

[0085] This application combines the idea of ​​spectral clustering to make scheduling judgments on new applications in real time. Compared with the global clustering idea of ​​the traditional spectral clustering algorithm, when a user clicks on the resource application information confirmation in the work order system, when constructing the Graph minidcX-ProducXWhen the JobManger receives the information collection Topic message, it first goes to the memory database to pull the subgraph information of the required product. If it is available, it directly extracts the information. If it is not available, it needs to be initialized first to fuzzily query the data with the productName in the HBase database table ProductMappingMinidcGraphy RowKey, extract all the data associated with the product and Minidc to the HDFS file ProductMappingMinidcFileInput, and output each row of data in the ColumnFamily under RowKey one by one. The specific file format is shown in Table 10:

[0086] productName-moduleName-minidcName appName

[0087] (Table 10)

[0088] Then, the values ​​with the same key are merged through distributed computing, and different applications under the same product, the same submodule and the same MiniDC are merged into one piece of data to generate an intermediate state subgraph with submodule dimensions. The specific format of the intermediate state file ModuleMappingMinidcMiddleFile is shown in Table 11:

[0089] productName-moduleName-minidcName appName1+appName2+…+appNameN

[0090] (Table 11)

[0091] Based on the intermediate file ModuleMappingMinidcFileMiddleFile, the Key of each row is cut again to form basic data based on the product and fault isolation domain Minidc dimensions. The specific format of the intermediate file ProductMappingMinidcMiddleFile is shown in Table 12:

[0092] productName-minidcName moduleName:appName1+appName2+…+appNameN

[0093] (Table 12)

[0094] Graph minidcX-ProducX The JobManager is initially constructed with the relationship subgraph of technology products and fault isolation domains. Different modules under the same Minidc of the same product and the sub-applications under the modules are merged into a row of data. A row of data is a subgraph. The format of the final subgraph output file ProductMappingMinidcGraphFile is shown in Table 13:

[0095]

[0096] (Table 13)

[0097] (2) Update Graph minidcX-ProducX Subgraph to in-memory database:

[0098] Through the above steps, a subgraph between a technology product that is not in the memory database and a MiniDC has been built. For real-time calculation, it needs to be refreshed to the memory database. In this method, Redis is used as the cache database. Here, Graph minidcX-ProducX Although the idea of ​​spectral clustering is a segmented subgraph, it can be seen that this method is simplified into a tree with a three-layer structure. Of course, a tree is also a kind of graph. The first-layer parent node is the product and MiniDC, the second layer is the submodule, and the third layer is the application. The storage here is based on redis<key,Value> In the form of, since each child node has a unique parent node in the tree structure, and each parent node has multiple child nodes, each node Key stores its parent node and its grandparent node, and Value stores all the child nodes of the parent node. The storage format according to the hierarchical structure is shown in Table 14:

[0099]

[0100] (Table 14)

[0101] The above has completed the initialization of the layout subgraph of a technology product and a MiniDC into the memory. When the resource application flow in each work order system is completed and the application location is determined, the data update module will also add the newly added nodes or subgraphs to the memory to ensure that the memory database is the most real-time subgraph forest.

[0102] (3) Real-time intelligent isolation domain dynamic scheduling based on spectral clustering algorithm:

[0103] Previously, we implemented the construction and cache storage of technology products and MiniDC subgraphs. minidcX-ProducX When the JobManager receives a message from the information collection topic, it regards a message, that is, a resource type of an application of a module of a product, as a data point, and thus needs to calculate the edge weights of the data point to the product in different MiniDCs, and select the one with the largest weight, which is the optimal architecture deployment for the technology product. First, it is necessary to build the overall graph of all MiniDC layouts built in the intelligent isolation domain layout persistence layer construction module minidcXPull from the Redis library to MinidcGraphMap, then use the Kafka demand data collection module to collect productName, perform fuzzy query in the Redis database, and store all productName-moduleName data that matches the key fuzzy match in ProductMinidcGraphMap, that is, screen the Minidcs that have deployed the submodule of the product. Usually, the best domain is selected from the Minidc collection that has deployed the submodule of the product but not the application, so as to complete the complete subgraph of the product under a Minidc first. If no result is found, follow up with the screening in the remaining Minidc collections that have not deployed the technology product. The specific weight scoring is calculated according to the following announcement 1. The weights of the newly added data points and the existing subgraphs are calculated one by one:

[0104] Edge increment =applySore*(centerSore+offsiteSore+citySore)

[0105] (Formula 1)

[0106] Here, applySore means that MiniDC can apply for resource scores, and the score is 1 if satisfied and 0 if not satisfied; centerSore means the score of active-active in the same center, and the score is 0 if satisfied and 1 if not satisfied. At the same time, if centerSore scores 1, it means that the highest priority active-active in the same center is not satisfied, then the same-center MiniDC scores the highest, and offsiteSoreden and citySore scores are 0; if offsiteSore scores 1, it means that the higher priority active-active disaster recovery in different locations is not satisfied, then citySore three-center or multi-center multi-active scores are 0; if centerSore and offsiteSore both score 0, it means that active-active in the same center and active-active disaster recovery in different locations are satisfied, then citySore scores 1 if not satisfied and 0.1 if satisfied. Here, 0.1 is used to indicate that the multi-active architecture has been fully satisfied, and you can choose any MiniDC to deploy. Therefore, Edge increment The final result set values ​​are {0 (resource conditions are not met), 1 (some type of multi-active deployment), 0.1 (all meet the optional resource pool)}. The specific program implementation steps are as follows:

[0107] Traverse MinidcGraphMap, compare the collected applyValue with the remaining allocatable resources of each Minidc, and score the applySore of each Minidc;

[0108] Use the key value "productName-moduleName" composed of the newly added data point information to fuzzily match the deployed Minidc of the corresponding module of the product from all the key values ​​of ProductMinidcGraphMap, generate the Map subset ProductMinidcGraphSubMap, traverse all the values ​​of ProductMinidcGraphSubMap, determine whether each Minidc has deployed the collected appName to be applied, and extract the Minidc names of all deployed appNames into ProductMinidcGraphSubList;

[0109] Traverse ProductMinidcGraphSubList, extract the location of Minidc, data center park name (dcRegion), and module room number (moduleNum) from MinidcGraphMap according to minidcName, generate ProductMinidcLocationSubList, and the Value stored in List is "location_dcRegion_moduleNum";

[0110] Create a function that returns an Integer flag value, isActive, with a default value of -1. Traverse ProductMinidcLocationSubList to determine whether the same-center active-active is met. If not, return 000. If the same-center active-active and remote disaster recovery active-active are not met, the isActive value is 100. If both the same-center active-active and remote active-active are met, the isActive value is 110 if the same-city active-active is not met, and 111 is returned if all active-active are met.

[0111] Prioritize the selection of the optimal scheduling domain from the MiniDC where the technology product submodule has been deployed. Here, in order to prioritize the construction of the global subgraph of the corresponding technology product under a MiniDC, when selecting the optimal MiniDC, first extract the List collection of all the technology product submodules deployed from ProductMinidcGraphMap, and then make a difference with the ProductMinidcGraphSubList where the application has been deployed to generate ProductMinidcDiffSubList where the product submodule is deployed but the application is not deployed. Traverse the List, combine Public Announcement 1, the judgment value of isActive, and the deployment status of ProductMinidcGraphSubList to select the optimal MiniDC deployment resource pool according to the priority of dual active in the same center, dual active for disaster recovery in different locations, and dual active in the same city. If there are multiple results with the same priority, select the MiniDC with the largest remaining allocatable resources and feedback the MiniDC name;

[0112] If the result returned in the previous step is NULL, traverse the remaining Minidc information in MinidcGraphMap, and use Formula 1 and the optimal Minidc deployment resource pool with the same-center active-active, remote disaster recovery active-active, and same-city active-active priorities. If there are multiple results with the same priority, select the Minidc with the largest remaining allocatable resources and feedback the Minidc name.

[0113] Finally, the optimal MiniDC deployment result is fed back to the work order system in real time according to one or more pieces of information in the resource work order, and fault isolation domain information is recommended one by one, providing a basis for intelligent analysis results for the final technology product deployment.

[0114] Furthermore, in this application, a data update module is also involved. When the technology product resource application process is completed in the work order system and the deployment fault isolation domain is finally finalized, the module collects the final deployment information through the collection module, and combines with another Update in Flink to update the JobManger in real time, and updates the final deployment information to the HBase tables ProductMappingMinidcFile and MinidcGraphDictionary of Modules 1 and 2. The ProductMappingMinidcFile table supplements and updates the deployment information, and the MinidcGraphDictionary table updates the Minidc resource information. In addition, if a new Minidc is added, it is also added to the MinidcGraphDictionary, and finally the relevant information is refreshed to the Redis cache database.

[0115] Furthermore, the present application also involves a data visualization module, which is composed of a back-end microservice developed based on Springboot and a front-end chart display module based on ECharts. Through the MiniDC and technology product deployment sub-graph information stored in the memory database Redis, the sub-graph structure of the technology products deployed on each MiniDC is displayed in the MiniDC dimension, which is used to visualize the information in real time.

[0116] In addition, see the instructions for Figure 4 , this application deploys the modules involved in the following ways:

[0117] Step S01: Since this method uses the HDFS file system of the Hadoop platform, and Kafka deployment depends on Zookeeper, Flink cluster deployment requires the Yarn component based on Hadoop. Zookeeper is also an important component of Hadoop. Therefore, the Hadoop platform is deployed first. The installation and configuration steps in this method are summarized as follows:

[0118] Install JDK on all nodes and configure environment variables. This method uses JDK version "1.8.0_202"; configure password-free login between cluster nodes, turn off firewalls and SeLinux on cluster nodes, add all node host name information to the first master node / etc / hosts file, and distribute the file to each node via scp; decompress Hadoop installation files on all nodes. This method uses version number "hadoop-3.2.2", use command "tar-zxvf hadoop-3.2.2.tar.gz" to decompress, rename to hadoop, configure hadoop environment variables; modify Hadoop configuration files, specify service access address, HDFS file copy number and other configuration information. Four configuration files, hadoop-env.sh, core-site.xml, hdfs-site.xml, and mapred-site.xml, need to be updated and distributed to each node; execute command "hadoopnamenode-format" to format the cluster, and execute "start-all.sh" to start the cluster.

[0119] Step S02: In this method, HBase database is used for data storage, so HBase database needs to be installed. This method uses the "hbase-2.3.6" version. The main steps are summarized as follows: unzip and install "hbase-2.3.6-bin.tar.gz" on all nodes, rename it to hbase, and configure hbase environment variables; modify HBase configuration files such as hbase-env.sh, hbase-site.xml, regionservers, etc., update node information, access address, master node information, number of replicas, etc. and distribute them to each node; Source environment variables, switch to the "$HBASE_HOME / bin" directory, and use ". / start-hbase.sh" to start the HBase cluster.

[0120] Step S03: Create the MinidcGraphDictionary table and the ProductMappingMinidcGraphy table in HBase;

[0121] Step S04: Install Scala, the language that Spark development depends on. The Scala version used in this method is "scala-2.12.14". Similarly, all nodes unzip the installation package, configure environment variables, and use "scala-version" to verify that the installation is successful;

[0122] Step S05: Deploy and install the Spark distributed memory computing engine. This method uses the version "spark-3.1.2-bin-hadoop3.2". The main steps are summarized as follows: unzip the installation package on all nodes, rename it to spark, and configure the spark environment variables; modify the configuration file / software / spark / conf / workers to configure the computing nodes; modify the first master node configuration file spark-env.sh and distribute it to other nodes. The specific content of the configuration file is shown in Table 15:

[0123]

[0124] (Table 15)

[0125] Run the start-all.sh command to start all nodes in the cluster and use jps to check the process status to confirm that the cluster is normal.

[0126] Step S06: Deploy the distributed message queue Kafka cluster. The installation and configuration steps in this method are summarized as follows: Unzip the Kafka installation file on all nodes. This method uses the version number "kafka_2.12-3.0.0". Use the command "tar-zxvfkafka_2.12-3.0.0.tgz" to unzip it and rename it to kafka, and configure the kafka environment variables; modify the server.properties configuration file in the config directory. The specific configuration is shown in Table 16:

[0127]

[0128]

[0129] (Table 16)

[0130] Distribute the configuration file to other nodes and modify broker.id, starting from the first node id and increasing in sequence without duplication; start the cluster (start Zookeeper first). The cluster startup command here can be written into a script for batch start and stop: bin / kafka server start.sh-daemon config / server.properties.

[0131] Step S07: Deploy the distributed stream computing Flink cluster used in this method. This method uses the Application Mode mode for deployment. The installation and configuration steps are summarized as follows:

[0132] Unzip the Flink installation file on all nodes. This method uses the version number "flink-1.13.0-bin-scala_2.12". Use the command "tar-zxvf Unzip "flink-1.13.0-bin-scala_2.12.tgz" and rename it to flink, configure flink environment variables; go to the unzipped directory and modify the configuration file flink-conf.yaml under conf, modify jobmanager.rpc.address as follows: jobmanager.rpc.address:hadoop102; add the master node information to the masters file as follows: hadoop102:8081; add the computing node host name to the workers file as follows: hadoop103, hadoop104; optimize the configuration of the JobManager and TaskManager components in the cluster in the flink-conf.yaml file according to actual conditions. The main configuration items are as follows: jobmanager.memory.process.size and taskmanager.memory.process.size are respectively used to configure the total memory available to the JobManager and TaskManager processes, including JVM metaspace and other overhead. The default value is 1600M, which can be adjusted appropriately according to the cluster scale. taskmanager.numberOfTaskSlots: configures the number of slots that each TaskManager can allocate. The default value is 1, which can be determined based on the number of CPUs that the machine where the TaskManager is located can provide to Flink. Slot is the computing resource allocated to a specific task in the TaskManager.

[0133] Step S08: Build and deploy a one-master, two-slave and three-sentinel Redis database cluster to store cache data of each module for real-time computing;

[0134] Step S09: deploy and use "java-jar" to run the intelligent isolation domain layout persistence layer construction module, the deployed technology product resource layout subgraph construction module based on the Spark distributed memory computing engine, the demand data collection module based on the Kafka distributed event stream platform, and the data update module;

[0135] Step S10: Submit the Flink stream computing job. This method submits the "Construct Graph minidcX-ProducXThere are two JobManagers: "Update job" and "Update real-time update job". The method of submitting each job is as follows: put the corresponding job processing Jar package into the lib / directory, execute the command ". / bin / standalone job.sh start job entry class" to start JobManager; use the command ". / bin / taskmanager.sh start" to start TaskManager;

[0136] Step S11: Use nginx to deploy and publish the data visualization module front-end chart display module.

[0137] It can be seen that the method for intelligent isolation domain scheduling of multi-cloud edge-end collaboration provided by this application utilizes the idea of ​​spectral clustering algorithm, combined with the specific data in the resource application information submitted by users or product managers, and the information subgraphs of different resource pools of existing technology products, to realize real-time edge calculation when resources are applied, and accurately and quickly promote the appropriate resource pool for the launch of new applications, thereby greatly improving the accuracy of the multi-active deployment architecture of technology products, greatly shortening the overall launch cycle of technology products, and at the same time providing accurate data basis for personnel evaluation in related processes, effectively enhancing the intelligent digital operation capabilities of the trusted cloud platform.

[0138] Based on the same inventive concept, an embodiment of the present application also provides a trusted multi-cloud edge collaborative intelligent isolation domain scheduling device. Since the principle of solving the problem by the device in the embodiment of the present application is similar to the above-mentioned trusted multi-cloud edge collaborative intelligent isolation domain scheduling method in the embodiment of the present application, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be repeated.

[0139] As the instruction manual Figure 5 As shown, an embodiment of the present application provides a credible multi-cloud edge-end collaborative intelligent isolation domain scheduling device, the device comprising:

[0140] A construction module 501 is used to construct a fault isolation domain by using standardized resource areas classified for diversified technology product applications; wherein each computer room in the data center is divided into a plurality of standardized resource areas, and each standardized resource area is composed of a plurality of rows of cabinets;

[0141] Extraction module 502, used to extract valid data from resource application information submitted by users based on Kafka distributed event stream platform; the valid data includes product information, submodule information, application information, application computing resource chip type and demand;

[0142] The scheduling module 503 is used to obtain a deployed product resource layout subgraph based on the Spark distributed memory computing engine, and combine the deployed product resource layout subgraph with the valid data, perform intelligent scheduling on the deployed product based on the spectral clustering algorithm, and recommend the optimal fault isolation domain for its multi-active deployment.

[0143] Based on the same concept of the present invention, the specification is attached Figure 6 As shown, the structure of an electronic device 600 provided in an embodiment of the present application includes: at least one processor 601, at least one network interface 604 or other user interface 603, a memory 605, and at least one communication bus 602. The communication bus 602 is used to realize the connection and communication between these components. The electronic device 600 optionally includes a user interface 603, including a display (for example, a touch screen, LCD, CRT, holographic imaging (Holographic) or projection (Projector), etc.), a keyboard or a pointing device (for example, a mouse, a trackball (trackball), a touch pad or a touch screen, etc.).

[0144] The memory 605 may include a read-only memory and a random access memory, and provides instructions and data to the processor 601. A portion of the memory 605 may also include a non-volatile random access memory (NVRAM).

[0145] In some implementations, the memory 605 stores the following elements, executable modules or data structures, or a subset thereof, or an extended set thereof:

[0146] Operating system 6051, including various system programs for implementing various basic services and processing hardware-based tasks;

[0147] The application module 6052 includes various application programs, such as a launcher, a media player, a browser, etc., which are used to implement various application services.

[0148] In an embodiment of the present application, by calling the program or instructions stored in the memory 605, the processor 601 is used to execute steps in a method for scheduling intelligent isolation domains in a multi-cloud and edge-end collaborative manner. By combining clustering ideas and edge computing, it can perform real-time computing and accurate recommendations on the user side, so that technological products can better have a multi-active operating architecture.

[0149] The present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the steps in the method for intelligent isolation domain scheduling of multi-cloud edge-end collaboration are executed.

[0150] Specifically, the storage medium can be a general storage medium, such as a mobile disk, a hard disk, etc. When the computer program on the storage medium is run, it can execute the above-mentioned trusted multi-cloud edge collaborative intelligent isolation domain scheduling method.

[0151] In the embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interfaces, and the indirect coupling or communication connection of devices or units can be electrical, mechanical or other forms.

[0152] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0153] In addition, each functional unit in the embodiments provided in the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0154] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, and other media that can store program codes.

[0155] Finally, it should be noted that the above embodiments are only specific implementation methods of the present application, which are used to illustrate the technical solutions of the present application, rather than to limit them. The protection scope of the present application is not limited thereto. Although the present application is described in detail with reference to the aforementioned embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the aforementioned embodiments within the technical scope disclosed in the present application, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application. They should all be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. A method for scheduling intelligent isolated domains in a multi-cloud edge-end collaborative manner, characterized in that: The method comprises the following steps: Use standardized resource areas classified for diverse technology product applications to build fault isolation domains; each computer room in the data center is divided into multiple standardized resource areas, each of which is composed of multiple rows of cabinets; Extract valid data from the resource application information submitted by the user based on the Kafka distributed event stream platform; the valid data includes product information, submodule information, application information, application computing resource chip type and demand; A resource layout subgraph of deployed products based on the Spark distributed memory computing engine is obtained, and the resource layout subgraph of deployed products is combined with the valid data, and intelligent scheduling is performed on the products to be deployed based on the spectral clustering algorithm, and an optimal fault isolation domain is recommended for its multi-active deployment.

2. According to the method of claim 1, the method is characterized in that: Using standardized resource areas for diverse technology product application classification, constructing fault isolation domains includes the following steps: Based on the layout comparison table, the constructed fault isolation domain is persistently written into the database for modeling, and all fault isolation domain layout diagrams are updated to the memory database; wherein, the HBase database is used, and the layout comparison table records the correspondence between each fault isolation domain and the computer room, data center, and region.

3. According to claim 2, a method for scheduling intelligent isolated domains in a multi-cloud edge-end collaboration is characterized in that: in, Make a point log in advance on the resource application side where the user submits resource application information to collect product information, sub-module information, application information, application computing resource chip type and demand.

4. According to claim 3, a method for scheduling intelligent isolated domains in a multi-cloud edge-end collaboration is characterized in that: The deployed product resource layout subgraph based on the Spark distributed memory computing engine includes the following steps: By calling the API gateways of different cloud platforms, obtain the list of virtual machines of all deployed products on each fault isolation domain; Read and clean the virtual machine Json data in the virtual machine list of all fault isolation domains, flatten the format to generate a virtual machine information configuration file, split the virtual machine information configuration file to generate a virtual machine fault isolation domain mapping file, and upload it to the distributed file system; Processing the virtual machine fault isolation domain mapping file based on the Spark distributed memory computing engine to obtain a product fault isolation domain mapping file; According to the product fault isolation domain mapping file, the constructed deployed product resource layout subgraph is stored in the HBase database; wherein, the deployed layout subgraph of each product in each fault isolation domain is recorded by setting a table.

5. According to claim 4, a method for scheduling intelligent isolated domains in a multi-cloud edge-end collaboration is characterized in that: Based on the spectral clustering algorithm, intelligent scheduling is performed on the deployed products to recommend the optimal fault isolation domain for their active-active deployment, including the following steps: Each type of valid data collected is taken as a vertex, and its weight correlation is calculated with the deployed product resource layout subgraph; According to the established fault isolation domain layout and the remaining resources in the current fault isolation domain, all modules are deployed in the same fault isolation domain of a product. According to the priorities of the module application graph, the dual-center dual-active module application graph, the remote disaster recovery dual-active module application graph, and the resource layout sub-graph of deployed products, a weight correlation calculation is performed until the maximum weight is selected for each valid data, and the optimal fault isolation domain for the multi-active deployment of the deployed products is obtained.

6. According to claim 5, a method for scheduling intelligent isolated domains in a multi-cloud edge-end collaboration is characterized in that: The method further comprises the following steps: After deploying the optimal fault isolation domain for the product to be deployed, update the corresponding deployment information to the HBase database.

7. According to claim 6, a method for scheduling intelligent isolated domains in a multi-cloud edge-end collaboration, characterized in that: The method further comprises the following steps: The subgraph structure of the deployed products on each fault isolation domain is visualized by fault isolation domain.

8. A multi-cloud edge-end collaborative intelligent isolation domain scheduling device, characterized in that: The device comprises: A construction module is used to construct a fault isolation domain using standardized resource areas classified for diversified technology product applications; wherein each computer room in the data center is divided into a plurality of standardized resource areas, and each standardized resource area is composed of a plurality of rows of cabinets; An extraction module is used to extract valid data from the resource application information submitted by the user based on the Kafka distributed event stream platform; the valid data includes product information, submodule information, application information, application computing resource chip type and demand; The scheduling module is used to obtain a deployed product resource layout subgraph based on the Spark distributed memory computing engine, and combine the deployed product resource layout subgraph with the valid data, perform intelligent scheduling on the deployed product based on the spectral clustering algorithm, and recommend the optimal fault isolation domain for its multi-active deployment.

9. An electronic device, characterized in that: include: A processor, a memory and a bus, the memory storing machine-readable instructions executable by the processor. When the electronic device is running, the processor and the memory communicate through the bus. When the machine-readable instructions are executed by the processor, the steps of the method for coordinated intelligent isolation domain scheduling of multi-cloud edge devices as described in any one of claims 1 to 7 are performed.

10. A computer-readable storage medium, characterized in that: A computer program is stored on the computer-readable storage medium, and when the computer program is executed by the processor, the steps of the method for intelligent isolation domain scheduling of multi-cloud edge-end collaboration as described in any one of claims 1 to 7 are executed.

Images