Intelligent gray release method and system and computer equipment

Abstract

The embodiment of the invention provides an intelligent gray release method and system and computer equipment, an AI Agent device is arranged in the method, the AI Agent device comprises five perceptual bodies, namely, a knowledge perception agent, a multi-path strategy agent, a task planning agent, a tool use agent and a data decision agent, knowledge services provided in a knowledge base are utilized, and the knowledge services provided in the knowledge base are utilized to release the gray. And the specific steps and process of gray release implementation are determined without depending on artificial experience. The method comprises the following steps: respectively controlling a gray version and a formal version of a service system to enter a running state, and starting a flow mirroring function to generate first running data corresponding to the gray version and second running data corresponding to the formal version; the first real-time operation data information and the second real-time operation data information are obtained and sent to the AI Agents device, so that the AI Agents device determines whether to execute large-scale popularization of the switched gray version or not by using the knowledge base.

Description

Technical Field

[0001] This application relates to the field of grayscale deployment, and in particular to an intelligent grayscale deployment method, system, and computer device. Background Technology

[0002] With the rapid development of digital finance, the physical and time constraints of traditional financial services have been broken down. The online, real-time, mobile, and intelligent nature of financial services has greatly improved their convenience and accessibility. Financial technologies such as big data, cloud computing, blockchain, and artificial intelligence provide the underlying driving force for the development of digital finance. Especially in the research and development of financial products, services, and applications, the technical characteristics of containerization, microservices, and multi-terminal operation, along with the requirements for agile, seamless, and real-time services, have driven the continuous development of the technical architecture in the financial sector. This has resulted in an increasing complexity in system management and updates.

[0003] In the financial sector, the requirements for security, compliance, stability, and reliability place higher demands on the stable operation of business systems and the efficient handling of operational issues. In practical technology development within the financial sector, to effectively control review risks, ensure the stability of financial services, and minimize the impact of version defects, the canary release model has been incorporated into the product development process as a core capability that needs continuous development. Canary release is a gradual release process that expands the scope of pilot users in stages to achieve comprehensive verification of the quality of the new version. Simultaneously, it keeps issues discovered during verification within a small area and responds quickly, promptly fixing and rolling back the version. The capabilities of the canary release model are well-suited to proactively address the technical characteristics and service requirements of the financial sector.

[0004] Depending on the application deployment model and the degree of control over the scope of version releases, the industry offers several canary deployment models: canary deployment, rolling deployment, and blue-green deployment. In the process of building canary deployment capabilities during the development of financial business products, existing solutions suffer from a heavy reliance on manual experience for canary deployment actions. Summary of the Invention

[0005] In view of this, embodiments of this application provide an intelligent grayscale release method, system, and computer device to solve the problem that grayscale release operations heavily rely on human experience.

[0006] According to one aspect of this application, a smart canary release method is provided, comprising:

[0007] The grayscale version and the official version of the business system are controlled to enter the running state respectively, and the streaming mirror function is started to generate the first running data corresponding to the grayscale version and the second running data corresponding to the official version.

[0008] The first and second running data are acquired using a real-time data acquisition device and then pushed to the knowledge base module.

[0009] The knowledge base module processes the first running data and the second running data respectively, thereby obtaining first real-time running data information corresponding to the first running data and second real-time running data information corresponding to the second running data. The first real-time running data information and the second real-time running data information are sent to the AIAgents device, so that the AIAgents device uses the knowledge base to determine the switched grayscale version and whether to perform large-scale expansion and promotion of the switched grayscale version. The AI ​​Agents device includes a knowledge-aware intelligent agent, a multi-path strategy intelligent agent, a task planning intelligent agent, a tool-using intelligent agent, and a data-decision-making intelligent agent.

[0010] In some embodiments, the AIAgents device uses a knowledge base to determine the switched grayscale version and determines whether to perform a large-scale rollout of the switched grayscale version, including the following steps:

[0011] The multi-path strategy agent uses the first real-time running data information, the second real-time running data information, and the historical running data stored in the knowledge base to determine the covered and uncovered ranges corresponding to the historical running data;

[0012] The first and second real-time operating data information within the uncovered area are stored in the business product construction and operation knowledge base;

[0013] The first real-time operation data information and the second real-time operation data information within the covered area are compared with the historical operation data stored in the knowledge base to determine the pass rate;

[0014] If the pass rate is 100%, the knowledge-sensing intelligence agent will automatically form a gray-scale release strategy and a verification strategy, and call the task planning intelligence agent to form a complete release task and process arrangement result based on the gray-scale release strategy and the verification strategy.

[0015] The task planning agent is used to call the tool using agent, so that the tool using agent, based on the release task and process orchestration results planned by the task planning agent, schedules the version release tool, version update tool, application restart tool, health check tool, transaction verification tool and component verification tool to complete the switching and push of the gray version, and obtain the switched gray version;

[0016] The tool is used to invoke a data decision-making agent, which in turn obtains real-time operational data of the business system after the switch to the gray-scale version. Combined with the operational status after the gray-scale release strategy takes effect, the system operation stability index data of the business system is calculated. Based on the system operation stability index data, it is determined whether to expand the scale of the switch to the gray-scale version.

[0017] In some embodiments, before the steps of separately controlling the grayscale version and the official version of the business system to enter the running state, and starting the streaming mirror function to generate first running data corresponding to the grayscale version and second running data corresponding to the official version, the method further includes:

[0018] A knowledge base construction module is used to build a knowledge base. The knowledge base construction includes building a business product construction and operation knowledge base based on the construction data and operation history data of the business system, and building a gray release knowledge base based on the enterprise's historical gray release data. The knowledge base includes a business product construction and operation knowledge base and a gray release knowledge base.

[0019] Based on the business product construction and operation knowledge base and the gray release knowledge base, as well as the prompt word project, a fusion and collaboration device for multiple knowledge bases is constructed.

[0020] Construct real-time data acquisition devices;

[0021] Standard operating procedure (SOP) documents are prepared based on pre-existing experience.

[0022] Construct an AI Agents device for group decision-making, the AI ​​Agents device including a knowledge-aware agent, a multi-path policy agent, a task planning agent, a tool-using agent, and a data-decision agent.

[0023] In some embodiments, the step of building a business product construction and operation knowledge base based on the construction data and operation history data of the business system includes:

[0024] The data extraction module is used to receive construction data and historical operation data from the business system.

[0025] SID annotation is performed based on the construction data and operation history data of the aforementioned business system to obtain the original data;

[0026] Convert the raw data into a corresponding file stream;

[0027] The data processing and transformation unit uses the file stream to form a graph, which includes graph data nodes, and the graph data nodes include feature data, abstract data, fragmented data and raw data;

[0028] An indexing mechanism is established based on the graph using the index building module;

[0029] Based on the indexing mechanism, the RAG engine module is used to build a query engine and a dialogue engine.

[0030] In some embodiments, the step of transforming the file stream using data processing to form a graph includes:

[0031] The file stream is fragmented to obtain fragmented data, and a CID identifier is set on the fragmented data to obtain identifiable fragmented data;

[0032] The identified fragmented data is abstracted to obtain abstract data, and an AID identifier is set on the abstract data to obtain identified abstract data. The abstract data includes data blocks, data modalities, and data features. Currently, the fields of the data features are empty.

[0033] Extract feature data from the abstract data after identification, and fill the feature data into the data features to form new abstract data after identification;

[0034] The newly identified abstract data is used to construct a graph based on the subject information. The graph includes graph data nodes, which include feature data, abstract data, fragmented data, and original data.

[0035] In some embodiments, the step of converting the raw data into a corresponding file stream includes:

[0036] When the original data includes text data, the text data is converted into a first file stream;

[0037] When the original data includes audio data, the audio data is converted into text format data via ASR (Speech-to-Text Service), and the text format data is converted into a second file stream;

[0038] When the original data includes image data, the image data is converted, processed by a convolutional neural network, and then encoded using Base64 to become a third file stream.

[0039] When the original data includes video data, the audio portion of the video data is converted into a fourth file stream, and each image frame in the video data is processed by a convolutional neural network and then Base64 encoded to be converted into a fifth file stream, wherein the fourth file stream and the fifth file stream are associated by a time series identifier.

[0040] When the original data includes web page data, the text portion of the web page data is converted into a sixth file stream, and the image portion of the web page data is processed by a convolutional neural network and then Base64 encoded into a seventh file stream.

[0041] In some embodiments, the step of establishing an indexing mechanism based on the map includes:

[0042] Based on the graph, a knowledge graph index is constructed and stored in a graph database;

[0043] Using a large model, keywords are intelligently extracted from the feature data in the graph, and a keyword index is constructed and stored in Elasticsearch based on the keywords;

[0044] Based on the abstract data, fragmented data, and original data in the graph, a vectorized index is constructed using an embedding model and stored in a vectorized database;

[0045] A tree structure index is formed, wherein the tree structure index is rooted at the original data and sequentially concatenates the fragmented data, abstract data, and feature data.

[0046] Weights are assigned to the knowledge graph index, keyword index, vectorized index, and tree structure index respectively to construct an indexing mechanism, wherein the weights of the knowledge graph index, keyword index, vectorized index, and tree structure index decrease sequentially.

[0047] In some embodiments, the step of constructing the AIAgents device for group decision-making includes:

[0048] Supported by the AgentRunner base class, AgentWorker is expanded to form knowledge-aware intelligent agents, multi-path strategy intelligent agents, task planning intelligent agents, tool-using intelligent agents, and data-driven decision-making intelligent agents;

[0049] Based on a plug-in construction mechanism, SOP plug-ins and memory abstraction plug-ins are constructed. The SOP plug-ins are used to pre-configure the standard operation steps SOP files for the knowledge-aware agent, multi-path policy agent, task planning agent, tool-using agent, and data-decision agent, respectively. The memory abstraction plug-in abstracts four atomic operations—memory write, memory read, memory update, and memory delete—by shielding the usage differences of scene modes, and extracts the scene differences into parameters, which are then passed to the memory service.

[0050] According to another aspect of this application, an intelligent canary release system is provided for performing:

[0051] The grayscale version and the official version of the business system are controlled to enter the running state respectively, and the streaming mirror function is started to generate the first running data corresponding to the grayscale version and the second running data corresponding to the official version.

[0052] The first and second running data are acquired using a real-time data acquisition device and then pushed to the knowledge base module.

[0053] The knowledge base module processes the first running data and the second running data respectively, thereby obtaining first real-time running data information corresponding to the first running data and second real-time running data information corresponding to the second running data. The first real-time running data information and the second real-time running data information are sent to the AIAgents device, so that the AIAgents device uses the knowledge base to determine the switched grayscale version and whether to perform large-scale expansion and promotion of the switched grayscale version. The AI ​​Agents device includes a knowledge-aware intelligent agent, a multi-path strategy intelligent agent, a task planning intelligent agent, a tool-using intelligent agent, and a data-decision-making intelligent agent.

[0054] According to another aspect of this application, a computer device is provided, including a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, wherein the processor executes the program to implement the above-described intelligent grayscale release method.

[0055] According to another aspect of this application, a storage medium is provided that stores a computer program thereon, which, when executed by a processor, implements the above-described intelligent grayscale release method.

[0056] Using the above technical solutions, embodiments of this application provide an intelligent canary release method, system, and electronic device. The method includes an AIAgents device, which comprises five sensing agents: a knowledge-aware agent, a multi-path strategy agent, a task planning agent, a tool-using agent, and a data-decision agent. By utilizing the knowledge services provided in the knowledge base, the specific steps and processes for canary release implementation are determined without relying on human experience.

[0057] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0058] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0059] Figure 1 An exemplary flowchart of a smart canary release method provided according to some embodiments is shown;

[0060] Figure 2 An exemplary flowchart of yet another intelligent canary release method provided according to some embodiments is shown;

[0061] Figure 3 An exemplary flowchart of yet another intelligent canary release method provided according to some embodiments is shown;

[0062] Figure 4 An exemplary flowchart of yet another intelligent canary release method provided according to some embodiments is shown;

[0063] Figure 5 A schematic diagram of a spectrum provided according to some embodiments is shown for illustrative purposes;

[0064] Figure 6 An exemplary flowchart of yet another intelligent canary release method provided according to some embodiments is shown.

[0065] Figure 7 An example is shown in the overall architecture diagram of an intelligent canary release system. Detailed Implementation

[0066] The present application will be described in detail below with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in the embodiments of the present application can be combined with each other.

[0067] With the rapid development of digital finance, the physical and time constraints of traditional financial services have been broken down. The online, real-time, mobile, and intelligent nature of financial services has greatly improved their convenience and accessibility. Financial technologies such as big data, cloud computing, blockchain, and artificial intelligence provide the underlying driving force for the development of digital finance. Especially in the research and development of financial products, services, and applications, the technical characteristics of containerization, microservices, and multi-terminal operation, along with the requirements for agile, seamless, and real-time services, have driven the continuous development of the technical architecture in the financial sector. This has resulted in an increasing complexity in system management and updates.

[0068] In the financial sector, the requirements for security, compliance, stability, and reliability place higher demands on the stable operation of business systems and the efficient handling of operational issues. In practical technology development within the financial sector, to effectively control review risks, ensure the stability of financial services, and minimize the impact of version defects, the canary release model has been incorporated into the product development process as a core capability that needs continuous development. Canary release is a gradual release process that expands the scope of pilot users in stages to achieve comprehensive verification of the quality of the new version. Simultaneously, it keeps issues discovered during verification within a small area and responds quickly, promptly fixing and rolling back the version. The capabilities of the canary release model are well-suited to proactively address the technical characteristics and service requirements of the financial sector.

[0069] Depending on the application deployment model and the degree of control over the scope of version releases, the industry offers several canary deployment models: canary deployment, rolling deployment, and blue-green deployment. In the process of building canary deployment capabilities during the development of financial business products, existing solutions suffer from a heavy reliance on manual experience for canary deployment actions.

[0070] The heavy reliance on human experience can be understood as follows: In the financial field, after the design of a canary release is completed, the strategy needs to be configured in the relevant support system for canary release. Given the complexity of the relationships, all configurations require specialized human experience. Furthermore, even with the support of a canary release support system, the specific characteristics of the field necessitate verification of the accuracy of the expected results at each step, which also relies on human experience. Especially when problems arise during deployment and release, troubleshooting becomes heavily reliant on manual handling. Moreover, with the development of technical architectures, microservices, distributed systems, and containerization are widely used in financial business system construction. The complex system architecture and numerous service applications require significant time to locate the root cause of problems, followed by policy updates and gradual re-deployment.

[0071] Existing solutions, besides relying heavily on human experience for canary deployments, also suffer from limitations in fully leveraging the knowledge capabilities provided by historical data and insufficient utilization of online operational data from business systems. Timeliness and flexibility are significantly lacking in existing technical solutions. This issue is also related to the insufficient knowledge conversion capabilities of the aforementioned two types of data, resulting in the overall system still being unable to meet the requirements of efficient system construction and stable operation in the digital age.

[0072] The inability to fully leverage the knowledge provided by historical data can be understood as follows: With the advancement of information system digitization, the massive amounts of system data accumulated during the construction, operation, and continuous updating of information systems can provide crucial support for subsequent product development and iteration. In the field of canary releases, the various technical documents from the development and design process of existing business systems, the various audio-visual data, image data, structured business data, log document data accumulated during system operation, and the web page data used for displaying pages during business operations are all important data for system design and operation. If this data can be transformed into knowledge supporting system design and operation, the service capabilities of intelligent canary release systems can be further improved, eliminating potential problems and defects as much as possible before pushing canary services to users. However, current canary release devices do not have this capability.

[0073] The insufficient ability to utilize online operational data from business systems can be understood as follows: After a business system undergoes a gray-scale release or gray-scale version launch, users begin operating the system in a real production environment. For business data generated during the gray-scale version's operation, existing gray-scale models collect system operational data in real time; however, the data usage pattern is relatively simplistic, focusing on a single perspective such as runtime errors, performance deficiencies, or erroneous results to identify system problems, failing to develop knowledge of the system's operational domain. Integrating the collected incremental data with historical data to build knowledge services for the resulting system operation would further enhance the effectiveness of gray-scale releases. Real-time verification and improved efficiency in identifying system operational anomalies during user gray-scale service usage can ensure the stability and accuracy of financial services.

[0074] In summary, the design of gray-scale release solutions in the financial sector needs to address key challenges and difficulties such as the lack of knowledge derived from historical data, insufficient online data utilization capabilities, overly complex operations, and heavy reliance on human experience. This will enable continuous improvement in intelligence and precision, meeting the diverse technical characteristics and service needs of the financial industry.

[0075] To address the aforementioned technical issues, this application provides an intelligent canary release method. This method includes an AIAgents device, which comprises five types of sensing agents: a knowledge-aware agent, a multi-path strategy agent, a task planning agent, a tool-using agent, and a data-decision agent. By utilizing knowledge services provided in a knowledge base, the specific steps and processes for canary release implementation are determined without relying on human experience.

[0076] Figure 1 A flowchart of a smart grayscale publishing method provided according to some embodiments is illustrated. The method includes steps S100-S300.

[0077] S100: Control the grayscale version and the official version of the business system to enter the running state respectively, and start the streaming mirror function to generate the first running data corresponding to the grayscale version and the second running data corresponding to the official version.

[0078] In this embodiment, the business system controls the grayscale and official versions to enter the running state. For grayscale customers, the service carrier is the grayscale version. For official customers, the service carrier is the official version. Simultaneously, a streaming mirroring function is activated. That is, when grayscale customers use the grayscale version, their traffic is mirrored to their own official version, and business operations are automatically executed to obtain the corresponding execution status. Similarly, when official customers use the official version, their traffic is mirrored to their own grayscale version, and business operations are automatically executed to obtain the corresponding execution status.

[0079] In this embodiment, both a gray-scale version and a formal version are chosen because the formal version serves the core user group and must ensure stability and business continuity. The gray-scale version is for a small group of users (the gray-scale group), allowing for controllable risks (such as defects in new features). Bidirectional mirrored traffic ensures that even if the gray-scale version fails, the formal version can still operate independently, avoiding a global business interruption.

[0080] In this embodiment, both the grayscale version and the official version generate corresponding runtime data when entering the running state, namely, first runtime data corresponding to the grayscale version and second runtime data corresponding to the official version. In this embodiment, the first and second runtime data can be multimodal data, including text data, image data, audio data, video data, web page data, and / or structured data. Specifically, a business system may have function A. When function A is used, it generates corresponding background log records. These background log records may include data such as usage duration, operation path, system running status, interface response time, interface input and output, and interface display status. Collecting the actual performance of different versions of the system during operation allows for real-time verification of system version quality.

[0081] In some embodiments, Figure 2 A flowchart of yet another intelligent canary deployment method provided according to some embodiments is illustrated. The method further includes steps S400-S800 before performing step S100.

[0082] S400. A knowledge base construction module is used to construct a knowledge base. The construction of the knowledge base includes constructing a business product construction and operation knowledge base based on the construction data and operation history data of the business system, and constructing a gray release knowledge base based on the enterprise's historical gray release data. The knowledge base includes a business product construction and operation knowledge base and a gray release knowledge base.

[0083] In this embodiment, there are two knowledge bases: a business product construction and operation knowledge base and a canary release knowledge base. Both knowledge bases include three modules: data extraction, index construction, and a RAG engine. The difference is that the business product construction and operation knowledge base is based on multimodal data from the business system's construction data and operational history data. The canary release knowledge base is based on multimodal data from the enterprise's historical canary release data.

[0084] The following describes the construction process in detail, using the construction and operation knowledge base of a business product as an example. In some embodiments, Figure 3 A flowchart of yet another intelligent canary release method provided according to some embodiments is illustrated. The step of building a business product construction and operation knowledge base based on the construction data and operation history data of the business system includes S401-S406.

[0085] S401. Use the data extraction module to obtain the construction data and operation history data of the business system.

[0086] The construction data and operation history data of the business system are multimodal data, which include text data, image data, audio data, video data, web page data and / or structured data.

[0087] S402. Based on the construction data and operation history data of the business system, SID (Original Data Number) is generated to obtain the original data.

[0088] In this embodiment, edge computing units are used to label the construction data and operational history data of business systems with SIDs. Different business systems have different SIDs for their construction data and operational history data.

[0089] S403. Convert the raw data into a corresponding file stream.

[0090] In this embodiment, considering that file creation takes time and large file network transmission puts significant pressure on network bandwidth, these factors can lead to high latency in data acquisition and transmission. Therefore, by converting the raw data into a streaming transmission mode, transmission begins as soon as the file content changes, improving the timeliness and quality of data upload in real time and in small batches.

[0091] In some embodiments, the step of converting the raw data into a corresponding file stream includes:

[0092] When the original data includes text data, the text data is converted into a first file stream.

[0093] When the original data includes audio data, the audio data is converted into text format data via ASR (Speech-to-Text Service), and the text format data is converted into a second file stream.

[0094] When the original data includes image data, the image data is converted, processed by a convolutional neural network, and then encoded using Base64 to become a third file stream.

[0095] When the original data includes video data, the audio portion of the video data is converted into a fourth file stream. Each image frame in the video data is processed by a convolutional neural network and then Base64 encoded into a fifth file stream. The fourth and fifth file streams are associated using time-series identifiers. In this embodiment, both image frames and audio portions have corresponding time-series identifiers, and image frames and audio portions with the same time-series identifier are associated.

[0096] When the original data includes web page data, the text portion of the web page data is converted into a sixth file stream, and the image portion of the web page data is processed by a convolutional neural network and then Base64 encoded into a seventh file stream.

[0097] In this embodiment, the original data is transformed according to different data modalities (including text data, audio data, image data, video data, and web page data) to support subsequent streaming computing and processing.

[0098] S404. The data processing and conversion unit forms a graph based on the file stream. The graph includes graph data nodes, which include feature data, abstract data, fragmented data, and raw data.

[0099] In this embodiment, the data extraction module is used to solve the problem of multimodal data access and fusion of construction data and historical operation data of the business system, forming a real-time streaming data processing capability for multimodal data. After being processed by the data processing and conversion unit, the data is organized in a four-layer linked list pattern of raw data -> fragmented data -> abstract data -> feature data. At the most detailed feature data dimension, a graph is constructed according to the subject and relationship to form a complete data graph linked list data structure, providing structural support for subsequent index construction and retrieval.

[0100] In some embodiments, Figure 4A flowchart of yet another intelligent grayscale publishing method provided according to some embodiments is illustrated. The step of forming a map based on the file stream using a data processing and conversion unit includes S4041-S4044.

[0101] S4041. The file stream is split into data segments to obtain segmented data, and the segmented data is assigned a CID identifier (segmented data number) to obtain the identified segmented data.

[0102] In this embodiment, the data is split into fragments by using the dataSplit method in the data splitter, according to the size threshold, end character, and delimiter, and the fragmented data is assigned a CID identifier.

[0103] S4042. Data abstraction is performed on the identified fragmented data to obtain abstract data, and an AID identifier (abstract data number) is set on the abstract data to obtain the identified abstract data. The abstract data includes data blocks, data modalities, and data features. The current field of the data feature is empty.

[0104] In this embodiment, the identified fragmented data enters the data abstraction converter, and the dataAbstract method is used to abstract the fragmented data to obtain abstract data, and an AID identifier is set for the abstract data.

[0105] S4043. Extract feature data from the abstract data after identification, and fill the feature data into the data features to form new abstract data after identification.

[0106] In this embodiment, a feature extraction converter is used to extract feature data from the identified abstract data. This feature data may include subject, keywords, summary, domain, and vector. The extracted feature data is then filled into the fields of the feature data to form new identified abstract data.

[0107] S4044. Construct a graph from the newly identified abstract data according to the subject information. The graph includes graph data nodes, which in turn include feature data, abstract data, fragmented data, and original data. In this embodiment, the subject information includes information such as the business system name, functional operation information, and grayscale strategy.

[0108] In this embodiment, a data processing and transformation function unit is used to construct a data transformation channel, and with the technical support of large models, embedded models, and knowledge graphs, data transformation is completed. In this embodiment, a graph data converter is used to construct a graph from the newly identified abstract data according to the subject information, forming a graph as shown below. Figure 5 The graph shown is a graph linked list structure, and the graph includes multiple graph data nodes.

[0109] S405. Using the index building module, establish an indexing mechanism based on the map.

[0110] In this embodiment, a multi-level adaptive weighted indexing mechanism is formed through the index building module to improve data recall. An indexing mechanism is established for the graph generated after the data extraction module has run.

[0111] In some embodiments, the step of establishing an indexing mechanism based on the map includes:

[0112] Based on the graph, a knowledge graph index (GraphIndex) is constructed and stored in the graph database.

[0113] Using a large model, keywords are intelligently extracted from the feature data in the graph, and a keyword index is constructed based on the keywords and stored in Elasticsearch (search server).

[0114] Based on the abstract data, fragmented data, and original data in the graph, a vectorized index (VectorIndex) is constructed using an embedding model and stored in a vectorized database.

[0115] A tree structure index is formed, wherein the tree structure index is rooted at the original data and sequentially concatenates the fragmented data, abstract data, and feature data;

[0116] Weights are assigned to the knowledge graph index, keyword index, vectorized index, and tree structure index to construct an indexing mechanism, wherein the weights of the knowledge graph index, keyword index, vectorized index, and tree structure index decrease sequentially. This indexing mechanism improves efficiency and increases information hit rate.

[0117] S406. Using the RAG engine module based on the indexing mechanism, construct a query engine and a dialogue engine.

[0118] In this embodiment, the RAG engine module is used to form a service capability for mixed-pattern intent understanding, thereby constructing a high-quality knowledge base service. The RAG engine includes a query engine and a dialogue engine. The query engine refers to the one-time response content obtained through natural language queries and questions; the dialogue engine is a stateful query engine that provides multiple consecutive queries with context.

[0119] Specifically, the steps for building a query engine and a dialogue engine using the RAG engine module based on the aforementioned indexing mechanism include: inheriting from CustomQueryEngine and implementing the custom_query method, sequentially defining the knowledge graph index, keyword index, vectorized index, and tree structure index to build the query engine; the dialogue engine, by adding a context memory device, uses the best dialogue mode to build an Agent dialogue service mode based on the ReActAgent engine. The RAG engine module first performs the query engine and then calls the dialogue engine; this hybrid mode enhances intent understanding capabilities.

[0120] In this embodiment, with the support of the data extraction module, the index building module, and the RAG engine module, a business product construction and operation database is built based on the construction data and operation history data of the business system.

[0121] In this embodiment, both the business product construction and operation knowledge base and the canary release knowledge base are constructed using three modules: data extraction, index building, and a RAG engine. The difference lies in that the business product construction and operation knowledge base is based on multimodal data from the business system's construction and operational history. The canary release knowledge base is based on multimodal data from the enterprise's historical canary release data, but the process is the same and will not be elaborated further.

[0122] This embodiment innovatively proposes building a business product construction and operation knowledge base based on the construction data and operational history data of the business system, and a gray-scale release knowledge base based on the enterprise's historical gray-scale release data. Three technical devices—data extraction, index building, and RAG engine—are constructed sequentially to complete the construction of both types of knowledge bases based on data technology support. The data extraction device solves the problem of multimodal data access and fusion, forming a multimodal real-time streaming data processing capability; the index building module forms a multi-level adaptive weighted index mechanism to improve data recall rate; and the RAG engine module forms a service capability for understanding mixed-modal intents, constructing a high-quality knowledge base service, effectively solving the constraint problem that existing devices cannot fully utilize the knowledge capabilities provided by historical data. The high-quality knowledge base ensures the intelligence and accuracy of the gray-scale release system and services.

[0123] In this embodiment, two types of knowledge base services are built based on historical data for this work domain: a business product construction and operation knowledge base and a gray-scale release knowledge base. These two knowledge base services operate collaboratively. In building the business product construction and operation knowledge base, design document data from the system construction process, various structured business data, log document data, audio / video data, and image data generated during system operation, web page data responsible for page interaction services, and financial domain business attribute information of the business product are effectively integrated to build the domain knowledge base for this business product. This achieves the cleaning, alignment, parsing, and integration of various modal data to form knowledge-based services, such as the functional paths of historical user operations, key functional service links, overall historical system service performance, and usage analysis of new functions. This provides accurate domain knowledge services for product gray-scale release and verification, guiding future product release services through historical knowledge. Simultaneously, the company's accumulated experience in the gray-scale release domain is used in the same technical model to build a gray-scale release knowledge base, providing domain support for the generation and formulation of gray-scale strategies. The separate construction and integrated use of these two knowledge bases, while ensuring the purity of domain knowledge, provides highly efficient knowledge services. To effectively address the problem that insufficient historical data knowledge can hinder the effectiveness of the canary release system.

[0124] S500. Based on the business product construction and operation knowledge base and the gray release knowledge base, as well as the prompt word project, a fusion and collaboration device for multiple knowledge bases is constructed.

[0125] In this embodiment, the two types of knowledge bases are constructed separately for different domains, ensuring the professional depth of each domain's knowledge base. Then, an efficient integration and collaborative service mechanism is implemented to enable the combined use of knowledge from different domains, continuously improving the end-to-end service of gray-scale strategy customization, gray-scale version verification, gray-scale operation monitoring, and gray-scale version switching, thus promoting version updates while ensuring the stable operation of the business system.

[0126] In this embodiment, knowledge base fusion and collaborative services are achieved through prompt word engineering and a two-database configuration model. For different tasks undertaken by the canary release system, such as strategy planning, the following ProptTemplate is generated: "You are a planning robot. The system you are planning possesses the construction patterns and operational characteristics in knowledge base {A}. For existing canary release strategy knowledge, please use knowledge base {B}. Please form the corresponding version verification strategy from the above knowledge according to the positive group pattern."

[0127] S600, Construct a real-time data acquisition device.

[0128] In this embodiment, the real-time data acquisition device can be used to acquire first running data and second running data. The first running data corresponds to the grayscale version, and the second running data corresponds to the official version. In this embodiment, the online running data (first running data and second running data) of the official version and the grayscale version are collected and stored. In this embodiment, the underlying knowledge of the real-time running data of the two versions can be deeply mined to promote the real-time evolution of the knowledge base, so as to ensure the effectiveness of knowledge support for subsequent grayscale releases. This ensures the knowledge service's up-to-date nature.

[0129] In the specific implementation, the real-time data acquisition device mainly consists of two parts: one is a data acquisition device built based on the JavaAgent technology principle; the other is a memory storage and computing device built with the support of the memory computing component. All data in this device is stored in memory, and the data cleaning and transformation are completed through the memory computing device. The second part, the memory device, also provides memory-level data access services for the entire device.

[0130] In this embodiment, an incremental business data knowledge transformation service capability is built for the online operation environment of canary release services. This continuously improves the timeliness and accuracy of the knowledge base for business product construction and operation and canary release, ensuring the efficient service capability of domain knowledge for subsequent canary release systems. Through a real-time data acquisition device, real-time business data is transformed into knowledge domain services that support the operation of the entire canary release system. This device effectively ensures the in-depth application of online operational data, while the continuous enrichment and accumulation of real-time knowledge in the business product construction and operation knowledge base solves the timeliness problem of the entire system.

[0131] S700: Compile standard operating procedure (SOP) documents based on pre-set experience.

[0132] In this embodiment, to address issues arising during the operation of AIAgents devices (including knowledge-aware agents, multi-path policy agents, task planning agents, tool-using agents, and data-decision agents) due to task complexity, abnormal tool scheduling, and API interface scheduling failures, pre-configured Standard Operating Procedures (SOPs) based on expert experience are provided for the entire operational process of each agent, ensuring the high efficiency of the entire canary release system. All agents integrate the fallback standard operating procedures for this service via SOP plugins.

[0133] Standard Operating Procedures (SOPs) are developed based on existing expert experience to address issues such as task complexity, tool scheduling anomalies, and API interface scheduling failures during the operation of intelligent agents. For each intelligent agent's entire operational process, SOPs based on expert experience are pre-defined to ensure the high efficiency of the entire canary release system.

[0134] S800, Constructing a group decision-making AI Agents device, wherein the AI ​​Agents device includes a knowledge-aware intelligent agent, a multi-path policy intelligent agent, a task planning intelligent agent, a tool-using intelligent agent, and a data-decision intelligent agent.

[0135] In the construction of specific AI Agents devices, based on the ReActAgent class definition, various types of Agents (intelligent agents) are constructed. Five types of agents are defined sequentially: from_knowledge (knowledge-aware agent), from_strategy (multi-path strategy agent), from_tools (tool-using agent), from_data (data-driven decision-making agent), and from_step (construction task planning agent). The first is the Knowledge-aware Agent (from_knowledge), a basic agent used to retrieve domain knowledge in real time from two types of knowledge bases in a multi-knowledge-base fusion and collaboration device, and to perceive knowledge changes.

[0136] A multi-path strategy agent (from_strategy) is constructed. Under the service of the basic knowledge-aware agent, a multi-path gray-scale strategy is formed based on the gray-scale function definition, including two types: release strategy and verification strategy. The verification strategy is further divided into four types of operation strategies: group mode forward verification strategy, group mode reverse verification strategy, individual mode forward verification strategy, and individual mode reverse verification strategy. Among them, group mode and individual mode refer to the group behavior and individual behavior used by the business in the knowledge base, respectively, to support the global verification of comprehensive gray-scale services.

[0137] The construction task planning agent (from_step) automatically plans the release tasks and process orchestration based on the release and verification strategies generated by the multi-path strategy agent;

[0138] The tool uses intelligent agents (from_tools), which are mainly used to support the task planning intelligent agents in planning the task and process orchestration results, and to schedule existing tools for version release, version update, application restart, health check, transaction verification, and component verification to complete the switching and push of gray versions.

[0139] Finally, the data decision-making agent (from_data) acquires real-time operational data of the business system after the switchover and, combined with the operational status after the gray-scale release strategy takes effect, calculates the system's operational stability index data. Based on the system operational stability index data, it determines whether to scale up the rollout of the gray-scale version after the switchover. Alternatively, it can acquire real-time operational data of the business system after the switchover and, combined with the operational status after the gray-scale release strategy takes effect, determine whether to scale up the rollout of the gray-scale version and ultimately switch to the full version based on the results of different verification strategies. This is a device based on multi-agent collaboration to form a group verification and group decision-making process.

[0140] In this embodiment, AgentWorker is expanded to form multiple types of intelligent agents, supported by the AgentRunner base class. The system comprises: a knowledge-aware intelligent agent, a basic intelligent agent used to retrieve domain knowledge in real time from two types of knowledge bases and perceive knowledge changes; a multi-path strategy intelligent agent, which, under the service of the basic knowledge-aware intelligent agent, forms multi-path gray-scale strategies based on gray-scale function definitions, including release strategies and verification strategies. Verification strategies are further divided into four categories: group mode forward verification strategy, group mode reverse verification strategy, individual mode forward verification strategy, and individual mode reverse verification strategy; a task planning intelligent agent, which automatically plans release tasks and orchestrates processes based on the release and verification strategies generated by the multi-path strategy intelligent agent; a tool-using intelligent agent, mainly used to support the task and process orchestration results planned by the task planning intelligent agent, and to schedule existing tools for version release, version update, application restart, health check, transaction verification, and component verification to complete the gray-scale version switching and push; and finally, a data decision-making intelligent agent, which obtains real-time operating data of the business system after the gray-scale version is executed, and, combined with the operating status after the gray-scale release strategy takes effect, confirms whether to execute the large-scale expansion and promotion of the gray-scale version and the final full version switch. This is a device based on multi-agent collaboration to form group verification and group decision-making.

[0141] In some embodiments, the steps of constructing a group decision-making AI Agents device include: expanding AgentWorker to form knowledge-aware intelligent agents, multi-path policy intelligent agents, task planning intelligent agents, tool-using intelligent agents, and data-decision intelligent agents, supported by the AgentRunner base class.

[0142] Based on a plug-in-based construction mechanism, a Standard Operating Procedure (SOP) plug-in and a memory abstraction plug-in are built. The SOP plug-in pre-defines standard operating procedures for the knowledge-aware agent, multi-path policy agent, task planning agent, tool-using agent, and data-decision agent. The memory abstraction plug-in abstracts four atomic operations—memory write, memory read, memory update, and memory delete—by masking the differences in usage scenarios, and extracts scenario differences into parameters, which are then passed to the memory service. Specifically, the implementation method for passing these parameters to the memory service is described.

[0143] In this embodiment, a full-device memory service model is constructed, establishing a layered usage scheme for memory abstraction and memory operations. In four scenarios—knowledge base construction, real-time data acquisition device, AIAgents, and SOP—a memory abstraction mechanism is built to shield usage differences. The underlying layer achieves unified data flow and manipulation within memory. This memory-based mechanism ensures the service timeliness of the entire grayscale device.

[0144] S200: Use a real-time data acquisition device to acquire the first running data and the second running data, and push them to the knowledge base module.

[0145] In this embodiment of the application, the real-time data acquisition module is used to acquire the first running data corresponding to the grayscale version and the second running data corresponding to the official version.

[0146] S300: The knowledge base module processes the first running data and the second running data respectively, thereby obtaining first real-time running data information corresponding to the first running data and second real-time running data information corresponding to the second running data; the first real-time running data information and the second real-time running data information are sent to the AI ​​Agents device, so that the AI ​​Agents device uses the knowledge base to determine the switched grayscale version and determines whether to perform large-scale expansion and promotion of the switched grayscale version; wherein the AI ​​Agents device includes a knowledge-aware intelligent agent, a multi-path strategy intelligent agent, a task planning intelligent agent, a tool-using intelligent agent, and a data-decision-making intelligent agent.

[0147] In this embodiment, the steps for processing the first and second running data include a data extraction module, a data processing and conversion unit, index construction, and the formation of real-time running data information using the RAG engine module. The real-time running data information formed using the first running data is called the first real-time running data information, and the real-time running data information formed using the second running data is called the second real-time running data information. Both the first and second real-time running data information are non-knowledge-based and are isolated from existing knowledge.

[0148] In some embodiments, Figure 6 A flowchart of yet another intelligent grayscale release method provided according to some embodiments is illustrated. The steps of the AIAgents device using a knowledge base to determine the grayscale version after switching and determining whether to perform a large-scale rollout of the grayscale version include S301-S306.

[0149] S301. Using the multi-path strategy agent, based on the first real-time running data information, the second real-time running data information, and the historical running data stored in the knowledge base, determine the covered and uncovered ranges corresponding to the historical running data.

[0150] In this embodiment, the knowledge base includes a business product construction and operation knowledge base and a canary release knowledge base. Historical operation data can be historical operation data corresponding to the four types of operation strategies of the corresponding business system, obtained from the knowledge base. The historical operation data has the same format as the first real-time operation data information and the second real-time operation data information.

[0151] For example, historical operation data belongs to the data related to the operation of a certain function in the business system. The first real-time operation data information and the second real-time operation data information are also related to the operation of the function. At this time, it is determined that the first real-time operation data information and the second real-time operation data information are covered by historical operation data. The first real-time operation data information and the second real-time operation data information belong to the covered range corresponding to historical operation data. The remaining first real-time operation data information and the second real-time operation data information belong to the uncovered range corresponding to historical operation data.

[0152] S302. Store the first and second real-time operational data information within the uncovered area in the business product construction and operation knowledge base. Thus, the first and second real-time operational data information within the uncovered area are stored as historical operational knowledge in the business product construction and operation knowledge base.

[0153] S303. The first real-time running data information and the second real-time running data information within the covered range are compared with the historical running data stored in the knowledge base to determine the pass rate.

[0154] In this embodiment, the first real-time running data information, the second real-time running data information, and the historical running data all have corresponding interfaces.

[0155] The step of comparing the first and second real-time operation data information within the covered range with historical operation data stored in the knowledge base to determine the passing ratio includes comparing the first real-time operation data information corresponding to the first interface with historical operation data, and comparing the second real-time operation data information corresponding to the second interface with historical operation data; calculating the sum of the number of first and second real-time operation data information that are the same as historical operation data, and the proportion of this sum to the total number of all first and second real-time operation data information. Here, the first interface and the second interface may be the same or different.

[0156] S304. If the pass rate is 100%, the knowledge sensory agent automatically forms a gray-scale release strategy and a verification strategy, and calls the task planning agent to form a complete release task and process arrangement result based on the gray-scale release strategy and the verification strategy.

[0157] In this embodiment, a 100% ratio is used to indicate that the first and second real-time operational data information within the currently covered range are identical to the historical operational data stored in the knowledge base. This means that the new function will not affect existing, normally functioning features. This requirement is also a core capability of the canary release system: controlling the scope of impact. The impact on existing, unmodified functions must be quantifiable and manageable. A 100% ratio is required to confirm that the canary function has zero impact on existing, unmodified functions. Only then can the canary release system take effect and drive the canary version switch.

[0158] If the pass rate is not 100%, the step of determining the covered and uncovered ranges corresponding to the historical running data based on the first real-time running data information, the second real-time running data information, and the historical running data stored in the knowledge base is re-executed until the pass rate is 100%.

[0159] S305. The task planning agent is used to call the tool using agent, so that the tool using agent schedules the version release tool, version update tool, application restart tool, health check tool, transaction verification tool and component verification tool based on the release task and process orchestration results planned by the task planning agent, so as to complete the switching and push of the gray version and obtain the switched gray version.

[0160] The version release tool, version update tool, application restart tool, health check tool, transaction verification tool, and component verification tool mentioned here are all existing tools and will not be described in detail here.

[0161] S306. Using the aforementioned tool, the intelligent agent invokes the data decision-making intelligent agent to enable the data decision-making intelligent agent to obtain real-time operating data of the business system after the switch to the gray-scale version, and calculate the system operation stability index data of the business system in conjunction with the operating status after the gray-scale release strategy takes effect; based on the system operation stability index data, determine whether to expand the scale of the gray-scale version after the switch.

[0162] In this embodiment, the real-time operational data of the business system is the data generated corresponding to the grayscale version after the switch. The operational status generated by the grayscale release strategy can include the risks and value after the operation, to support subsequent group strategies.

[0163] For example, system operation stability indicators can be functional response anomaly rate. If the functional response anomaly rate is less than 0.01%, it determines whether to perform a large-scale rollout of the gray-scale version after the switch, and can also switch to the final full version.

[0164] In some embodiments, the step of the data decision-making agent acquiring real-time operating data of the business system after the gray-scale version of the system is switched, and calculating the system operation stability index data of the business system in combination with the operating status after the gray-scale release strategy takes effect includes: acquiring real-time operating data of the business system after the gray-scale version of the system is switched, and determining the business system operation stability index data based on the verification strategy in combination with the operating status after the gray-scale release strategy takes effect.

[0165] The intelligent canary release method in this application embodiment includes an AI Agents device. The AI ​​Agents device includes five types of sensing agents: a knowledge-aware agent, a multi-path strategy agent, a task planning agent, a tool-using agent, and a data-decision agent. It utilizes knowledge services provided in a knowledge base to determine the specific steps and processes for canary release implementation without relying on human experience.

[0166] The method in this application embodiment has a complete knowledge base method for data extraction, index construction and RAG engine building, multiple knowledge bases are integrated and coordinated, and real-time running data collection and use are used to ensure the timeliness of knowledge. Agent invocation combined with SOP enables multi-Agent decision-making to improve version verification quality and complete workflow.

[0167] In this embodiment, five intelligent agent modules are constructed: a knowledge-aware intelligent agent, a multi-path strategy intelligent agent, a task planning intelligent agent, a tool-using intelligent agent, and a data-decision intelligent agent. These modules transform the knowledge services provided by the aforementioned domain knowledge base into specific steps and processes for canary release implementation. Furthermore, to enhance the effectiveness of canary verification, a multi-path verification strategy is implemented, consisting of a combined behavioral group pattern forward verification strategy, a group pattern reverse verification strategy, an individual pattern forward verification strategy, and an individual pattern reverse verification strategy. Data-driven decision services based on the strategy execution results further strengthen the capabilities of the canary release system. To ensure the operational stability of the canary release system in the financial field, standard operating procedures (SOPs) are pre-defined in the design of each intelligent agent. This addresses issues such as insufficient planning capabilities leading to logical discontinuities when large models handle complex tasks, and unstable tool calls due to poor API interface compatibility. Ultimately, this solves the problems of excessive manual operations and insufficient exception handling capabilities.

[0168] The method in this embodiment adopts a multi-agent architecture of AI Agents. First, it effectively integrates historical data generated from the construction and operation of the business system, as well as data generated from the online operation of canary services, with the financial business domain attributes of the business products to construct a domain-level knowledge base corresponding to the business product construction. This provides intelligent foundational services for subsequent product construction and canary operation. Various intelligent agents in the canary release domain are constructed. By building the data perception capabilities, operational planning capabilities, enhanced memory capabilities, tool usage capabilities, and autonomous action capabilities of these agents, the autonomous intelligence level of canary release is strengthened based on the capabilities supported by the business product domain knowledge base. Simultaneously, to ensure the operational quality of AI Agents in the financial field and effectively address issues such as insufficient planning capabilities leading to logical discontinuities and unstable tool calls due to poor API interface compatibility when large models handle complex tasks, a standard operating procedure (SOP) for the canary release domain is built based on existing data mining and expert experience, ensuring the efficiency and stability of the canary device itself. In the digital age, AI has opened up new possibilities for building canary release capabilities in the financial field.

[0169] The method in this application embodiment can comprehensively enhance the intelligent capabilities of canary releases in the financial field. Through data extraction, index construction, and a knowledge base efficiently built using the RAG engine, it leverages the knowledge service capabilities of system construction, operation, and canary release-related data, ensuring the timeliness of knowledge through real-time data acquisition and processing. It establishes an intelligent agent operation mechanism with group decision-making AI Agents and standard operating procedures (SOPs) as a safety net. Supported by the product construction and operation knowledge base and the canary release knowledge base, it realizes five types of intelligent agents: knowledge perception, strategy planning, task planning, tool invocation, and data-driven decision-making, addressing the previous problems of reliance on manual labor and operational complexity. The overall system operating efficiency is improved through in-memory computing devices.

[0170] This application also provides an intelligent canary release system for performing:

[0171] The grayscale version and the official version of the business system are controlled to enter the running state respectively, and the streaming mirror function is started to generate the first running data corresponding to the grayscale version and the second running data corresponding to the official version.

[0172] The first and second running data are acquired using a real-time data acquisition device and then pushed to the knowledge base module.

[0173] The knowledge base module processes the first running data and the second running data respectively, thereby obtaining first real-time running data information corresponding to the first running data and second real-time running data information corresponding to the second running data. The first real-time running data information and the second real-time running data information are sent to the AIAgents device, so that the AIAgents device uses the knowledge base to determine the switched grayscale version and whether to perform large-scale expansion and promotion of the switched grayscale version. The AI ​​Agents device includes a knowledge-aware intelligent agent, a multi-path strategy intelligent agent, a task planning intelligent agent, a tool-using intelligent agent, and a data-decision-making intelligent agent.

[0174] In some embodiments, Figure 7 An example is shown in the overall architecture diagram of an intelligent canary release system.

[0175] It should be noted that other corresponding descriptions of the functional units involved in the intelligent grayscale release system provided in the embodiments of this application can be found in the corresponding descriptions in the intelligent grayscale release method provided in the above embodiments, and will not be repeated here.

[0176] This application also provides a computer device, specifically a personal computer, server, network device, etc. The computer device includes a bus, processor, memory, and communication interface, and may also include input / output interfaces and a display device. The processor of the computer device provides computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database of the computer device stores location information. The network interface of the computer device is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements the steps in the various method embodiments.

[0177] Those skilled in the art will understand that the structure of the computer device described above is only a partial structure related to the solution of this application, and does not constitute a limitation on the computer device to which the solution of this application is applied. A specific computer device may include more or fewer components, or combine certain components, or have different component arrangements.

[0178] In one embodiment, a computer-readable storage medium is also provided, which may be non-volatile or volatile, and a computer program is stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0179] In one embodiment, a computer program product is also provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0180] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0181] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods.

[0182] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0183] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present application. It should be noted that a person of ordinary skill in the art may make various modifications and improvements without departing from the spirit of the present application, and these modifications and improvements fall within the scope of protection of the present application. Therefore, the scope of protection of the present application shall be determined by the appended claims.

Claims

1. A method for intelligent grayscale deployment, characterized in that, include: The grayscale version and the official version of the business system are controlled to enter the running state respectively, and the streaming mirror function is started to generate the first running data corresponding to the grayscale version and the second running data corresponding to the official version. The first and second running data are acquired using a real-time data acquisition device and then pushed to the knowledge base module. The knowledge base module is used to process the first running data and the second running data respectively, and correspondingly, the first real-time running data information corresponding to the first running data and the second real-time running data information corresponding to the second running data are obtained. The first real-time running data information and the second real-time running data information are sent to the AIAgents device so that the AIAgents device can use the knowledge base to determine the switched grayscale version and determine whether to implement the scale expansion and promotion of the switched grayscale version. The AI ​​Agents device includes a knowledge-aware agent, a multi-path policy agent, a task planning agent, a tool-using agent, and a data-driven decision-making agent.

2. The method according to claim 1, characterized in that, The steps of the AIAgents device using a knowledge base to determine the switched grayscale version and whether to proceed with the large-scale rollout of the switched grayscale version include: The multi-path strategy agent uses the first real-time running data information, the second real-time running data information, and the historical running data stored in the knowledge base to determine the covered and uncovered ranges corresponding to the historical running data; The first and second real-time operating data information within the uncovered area are stored in the business product construction and operation knowledge base; The first real-time operation data information and the second real-time operation data information within the covered area are compared with the historical operation data stored in the knowledge base to determine the pass rate; If the pass rate is 100%, the knowledge-sensing intelligence agent will automatically form a gray-scale release strategy and a verification strategy, and call the task planning intelligence agent to form a complete release task and process arrangement result based on the gray-scale release strategy and the verification strategy. The task planning agent is used to call the tool using agent, so that the tool using agent, based on the release task and process orchestration results planned by the task planning agent, schedules the version release tool, version update tool, application restart tool, health check tool, transaction verification tool and component verification tool to complete the switching and push of the gray version, and obtain the switched gray version; The tool is used to invoke a data decision-making agent, which in turn obtains real-time operational data of the business system after the switch to the gray-scale version. Combined with the operational status after the gray-scale release strategy takes effect, the system operation stability index data of the business system is calculated. Based on the system operation stability index data, it is determined whether to expand the scale of the switch to the gray-scale version.

3. The method according to claim 1, characterized in that, Before the steps of separately controlling the grayscale version and the official version of the business system to enter the running state, and starting the streaming mirror function to generate the first running data corresponding to the grayscale version and the second running data corresponding to the official version, the method further includes: A knowledge base construction module is used to build a knowledge base. The knowledge base construction includes building a business product construction and operation knowledge base based on the construction data and operation history data of the business system, and building a gray release knowledge base based on the enterprise's historical gray release data. The knowledge base includes a business product construction and operation knowledge base and a gray release knowledge base. Based on the business product construction and operation knowledge base and the gray release knowledge base, as well as the prompt word project, a fusion and collaboration device for multiple knowledge bases is constructed. Construct real-time data acquisition devices; Standard operating procedure (SOP) documents are prepared based on pre-existing experience. Construct an AI Agents device for group decision-making, the AI ​​Agents device including a knowledge-aware agent, a multi-path policy agent, a task planning agent, a tool-using agent, and a data-decision agent.

4. The method according to claim 3, characterized in that, The steps for building a business product construction and operation knowledge base based on the construction data and historical operation data of the business system include: The data extraction module is used to receive construction data and historical operation data from the business system. SID annotation is performed based on the construction data and operation history data of the aforementioned business system to obtain the original data; Convert the raw data into a corresponding file stream; The data processing and transformation unit uses the file stream to form a graph, which includes graph data nodes, and the graph data nodes include feature data, abstract data, fragmented data and raw data; An indexing mechanism is established based on the graph using the index building module; Based on the indexing mechanism, the RAG engine module is used to build a query engine and a dialogue engine.

5. The method according to claim 4, characterized in that, The step of using data processing to transform the file stream into a graph includes: The file stream is fragmented to obtain fragmented data, and a CID identifier is set on the fragmented data to obtain identifiable fragmented data; The identified fragmented data is abstracted to obtain abstract data, and an AID identifier is set on the abstract data to obtain identified abstract data. The abstract data includes data blocks, data modalities, and data features. Currently, the fields of the data features are empty. Extract feature data from the abstract data after identification, and fill the feature data into the data features to form new abstract data after identification; The newly identified abstract data is used to construct a graph based on the subject information. The graph includes graph data nodes, which include feature data, abstract data, fragmented data, and original data.

6. The method according to claim 4, characterized in that, The step of converting the raw data into a corresponding file stream includes: When the original data includes text data, the text data is converted into a first file stream; When the original data includes audio data, the audio data is converted into text format data using ASR, and the text format data is converted into a second file stream; When the original data includes image data, the image data is converted, processed by a convolutional neural network, and then encoded using Base64 to become a third file stream. When the original data includes video data, the audio portion of the video data is converted into a fourth file stream, and each image frame in the video data is processed by a convolutional neural network and then Base64 encoded to be converted into a fifth file stream, wherein the fourth file stream and the fifth file stream are associated by a time series identifier. When the original data includes web page data, the text portion of the web page data is converted into a sixth file stream, and the image portion of the web page data is processed by a convolutional neural network and then Base64 encoded into a seventh file stream.

7. The method according to claim 4, characterized in that, The steps for establishing an indexing mechanism based on the map include: Based on the graph, a knowledge graph index is constructed and stored in a graph database; Using a large model, keywords are intelligently extracted from the feature data in the graph, and a keyword index is constructed and stored in Elasticsearch based on the keywords; Based on the abstract data, fragmented data, and original data in the graph, a vectorized index is constructed using an embedding model and stored in a vectorized database; A tree structure index is formed, wherein the tree structure index is rooted at the original data and sequentially concatenates the fragmented data, abstract data, and feature data. Weights are assigned to the knowledge graph index, keyword index, vectorized index, and tree structure index respectively to construct an indexing mechanism, wherein the weights of the knowledge graph index, keyword index, vectorized index, and tree structure index decrease sequentially.

8. The method according to claim 3, characterized in that, The steps for constructing the AIAgents device for group decision-making include: Supported by the AgentRunner base class, AgentWorker is expanded to form knowledge-aware intelligent agents, multi-path strategy intelligent agents, task planning intelligent agents, tool-using intelligent agents, and data-driven decision-making intelligent agents; Based on a plug-in construction mechanism, SOP plug-ins and memory abstraction plug-ins are constructed. The SOP plug-ins are used to pre-configure the standard operation steps SOP files for the knowledge-aware agent, multi-path policy agent, task planning agent, tool-using agent, and data-decision agent, respectively. The memory abstraction plug-in abstracts four atomic operations—memory write, memory read, memory update, and memory delete—by shielding the usage differences of scene modes, and extracts the scene differences into parameters, which are then passed to the memory service.

9. An intelligent grayscale publishing system, characterized in that, Used to execute: The grayscale version and the official version of the business system are controlled to enter the running state respectively, and the streaming mirror function is started to generate the first running data corresponding to the grayscale version and the second running data corresponding to the official version. The first and second running data are acquired using a real-time data acquisition device and then pushed to the knowledge base module. The knowledge base module is used to process the first running data and the second running data respectively, and correspondingly, the first real-time running data information corresponding to the first running data and the second real-time running data information corresponding to the second running data are obtained. The first real-time running data information and the second real-time running data information are sent to the AI ​​Agents device so that the AI ​​Agents device can use the knowledge base to determine the switched grayscale version and determine whether to implement the scale expansion and promotion of the switched grayscale version. The AI ​​Agents device includes a knowledge-aware agent, a multi-path policy agent, a task planning agent, a tool-using agent, and a data-driven decision-making agent.

10. A computer device, characterized in that, The system includes a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, wherein the processor executes the program to implement the above-described intelligent grayscale release method.

Images