Multi-mode cloud disaster recovery adaptive method based on dynamic service awareness and cooperative scheduling

Abstract

The invention relates to a multi-mode cloud disaster recovery self-adaptive method based on dynamic service awareness and collaborative scheduling, which comprises the following steps: acquiring KPIs of a plurality of service systems in real time, generating current actual disaster recovery demand factors of each service system through a time sequence enhanced machine learning model in combination with preset service influence analysis information, and generating a disaster recovery self-adaptive model of each service system; determining a current disaster recovery demand of each service system; monitoring and quantifying the available disaster recovery capacity of all disaster recovery resource pools in real time, matching the current disaster recovery demand with the available disaster recovery capacity, and selecting an optimal disaster recovery mode and generating a final disaster recovery strategy by performing collaborative scheduling decision of global resource optimization on each service system; and according to the final disaster recovery strategy, automatically arranging corresponding disaster recovery resources for each service system, so that the disaster recovery environment of each service system enters a ready state. The business continuity, the multi-mode cooperation efficiency and the cost effectiveness are remarkably enhanced, and the method is suitable for high-availability scenes such as finance and e-commerce.

Description

Technical Field

[0001] This invention belongs to the field of cloud computing and disaster recovery technology, and in particular relates to a multi-mode cloud disaster recovery adaptive method based on dynamic business awareness and collaborative scheduling. Background Technology

[0002] With the widespread adoption of cloud computing, enterprises are increasingly demanding business continuity and data protection. However, traditional disaster recovery systems have revealed significant limitations in cloud environments. Cloud computing business systems are diverse, and their Recovery Time Objective (RTO) and Recovery Point Objective (RPO) change dynamically due to business type and load fluctuations. Static Business Impact Analysis (BIA) and manual configuration are difficult to match actual needs in real time, leading to insufficient support or resource waste during peak periods. Existing disaster recovery technologies (such as same-city active-active, off-site disaster recovery, and edge node clusters) each have their advantages and disadvantages in terms of RTO / RPO, cost, and applicable scenarios, but they lack a unified mechanism for dynamic selection and coordinated scheduling based on real-time business risks and resource status, resulting in low resource utilization and difficulty in cost control. Furthermore, disaster recovery switching usually relies on manual intervention, which is highly complex when involving multiple system dependencies and cross-regional data synchronization, making it difficult to consistently achieve the expected RTO / RPO, and the delayed recovery verification increases the risk of business interruption. In recent years, machine learning and automated operation and maintenance technologies have provided possibilities for solving the above problems, but existing solutions still fail to achieve closed-loop management of dynamic demand perception, multi-mode collaboration, automated switching, and verification.

[0003] Therefore, there is an urgent need for an adaptive cloud disaster recovery method that can dynamically perceive business needs, intelligently coordinate multiple disaster recovery modes, and efficiently perform switching and verification to improve business continuity, resource efficiency, and cost-effectiveness. Summary of the Invention

[0004] The purpose of this invention is to provide a multi-mode cloud disaster recovery adaptive method based on dynamic business awareness and collaborative scheduling, so as to solve the problems of low matching degree between business disaster recovery level and actual needs, weak collaborative management capability of multiple disaster recovery modes, low utilization rate of disaster recovery resources, and insufficient automation of disaster recovery switching and verification in existing cloud disaster recovery solutions.

[0005] To achieve one of the above-mentioned objectives, an embodiment of the present invention provides a multi-mode cloud disaster recovery adaptive method based on dynamic service awareness and collaborative scheduling, characterized in that it includes:

[0006] Real-time collection of KPIs from multiple business systems, combined with pre-set business impact analysis information, and generation of current actual disaster recovery demand factors for each business system through time-series augmented machine learning models to determine the current disaster recovery needs of each business system.

[0007] Real-time monitoring and quantification of the available disaster recovery capabilities of all disaster recovery resource pools, matching current disaster recovery needs with available disaster recovery capabilities, and selecting the optimal disaster recovery mode and generating the final disaster recovery strategy through collaborative scheduling decisions for global resource optimization of various business systems.

[0008] Based on the final disaster recovery strategy, corresponding disaster recovery resources are automatically orchestrated for each business system, so that the disaster recovery environment of each business system is ready.

[0009] As a further improvement to one embodiment of the present invention, the method further includes, in which the step of generating the current actual disaster recovery demand factor through a time-series augmented machine learning model includes,

[0010] Construct composite features for the time-series augmented machine learning model, including the volatility features of the KPIs, the periodic features associated with the business calendar, and the cross features of multiple KPIs;

[0011] The KPIs include API call success rate, response time, and number of concurrent users;

[0012] Based on the composite features, a basic time-series augmented machine learning model is generated through initial training using historical data and expert annotations.

[0013] Based on the aforementioned basic time-series augmented machine learning model, the current actual disaster recovery requirement factors and their confidence scores for each business system are generated through reasoning. Predictive samples with confidence scores below a preset threshold are identified through uncertainty sampling.

[0014] The low-confidence prediction samples are forwarded to experts for review, and the disaster recovery requirement factors after review are obtained. The review results are used to incrementally train the basic time-series augmented machine learning model through an online learning algorithm to generate an optimized time-series augmented machine learning model.

[0015] Based on the optimized time-series augmented machine learning model, combined with real-time KPIs and preset business impact analysis information, the current actual disaster recovery requirement factor is output.

[0016] As a further improvement to one embodiment of the present invention, the method further includes, wherein the current disaster recovery requirement includes a dynamic recovery time target and a dynamic recovery point target, calculated using the following formula:

[0017]

[0018]

[0019] in, For dynamic recovery time target, To dynamically restore point targets, and As a preset basic recovery target, This represents the actual disaster recovery demand factor.

[0020] As a further improvement to one embodiment of the present invention, the method further includes, in which the real-time monitoring and quantification of the available disaster recovery capabilities of all disaster recovery resource pools includes,

[0021] For all disaster recovery resource pools, including intra-city active-active, off-site disaster recovery and edge node clusters, collect resource capacity indicators of each resource pool in real time, including computing capacity, storage capacity, network bandwidth, recovery time target support capability and recovery point target support capability;

[0022] The resource capability indicators are standardized into unified capability values ​​to generate quantitative capability data for each resource pool;

[0023] Based on the quantified capability data, a global resource view is constructed and updated to record the available disaster recovery capabilities of each resource instance;

[0024] The global resource view is dynamically updated based on resource allocation, and resource conflicts are detected to support subsequent degradation mechanisms.

[0025] As a further improvement to one embodiment of the present invention, the method further includes, in the step of selecting the optimal disaster recovery mode and generating the final disaster recovery strategy through collaborative scheduling decision-making for global resource optimization of each business system, the method includes,

[0026] Dynamic disaster recovery levels are mapped based on the actual disaster recovery needs of each business system, and multiple business systems are prioritized based on their inherent importance.

[0027] Based on the aforementioned priority ranking, the disaster recovery modes and resource instances are traversed to generate candidate solutions for each business system, and scores are calculated using a weighted scoring formula:

[0028]

[0029] in, The weighted score for assigning the j-th disaster recovery mode instance to the i-th business system is used to quantitatively evaluate the merits of the assignment scheme. and These represent the differences between the available resource tolerance for business interruption and the tolerance for data loss, respectively, and the business requirements. It is a monotonically increasing function. To standardize the costs of adopting this solution, For cost mapping function, This refers to the capacity redundancy of the resource instance after resource allocation. It is a nonlinear function. For pre-configured weighting factors;

[0030] Based on the priority ranking, select the highest-scoring scheme for each business system, reserve resources, and update the global resource view;

[0031] Based on the selected disaster recovery mode instance, configuration parameters are generated for each business system, including data synchronization frequency and load balancing strategy;

[0032] Integrate the disaster recovery modes, resource allocations, and configuration parameters of various business systems to generate the final disaster recovery strategy;

[0033] If there are conflicts in resource allocation or no solution that meets the requirements, the degradation mechanism is triggered, selecting the suboptimal disaster recovery mode or reducing resource allocation, and generating an alarm.

[0034] As a further improvement to one embodiment of the present invention, the method further includes, in part, automatically orchestrating corresponding disaster recovery resources for each business system according to the final disaster recovery strategy, including...

[0035] Obtain the deployment topology of the business application and its dependent components;

[0036] Based on the deployment topology, use IaC tools to configure infrastructure on selected resource instances;

[0037] Using the configuration results, deploy application services and dependent components, and configure a data synchronization mechanism to meet the current recovery point objectives;

[0038] Based on the deployment results, perform service health checks and warm-up operations to ensure that the disaster recovery environment of each business system is ready.

[0039] As a further improvement of one embodiment of the present invention, the method further includes, upon receiving a disaster trigger signal or drill instruction, performing automated switching and scenario-based verification for multiple business systems, evaluating the recovery status of each business system, and providing feedback to optimize the time-series augmented machine learning model and disaster recovery strategy.

[0040] The automated switching and scenario-based verification include,

[0041] Upon receiving a disaster trigger signal or drill instruction, a directed acyclic graph containing multiple switching tasks is dynamically generated based on the call relationships between multiple business systems, and the switching tasks are concurrently scheduled through critical path analysis.

[0042] Based on the aforementioned concurrent scheduling, atomic and idempotent operations are performed, and checkpoints are set to support failure recovery;

[0043] After the switchover is completed, a scenario-based verification process combining black-box and white-box verification is used to assess the recovery status of each business system and generate a quantitative recovery health score.

[0044] The black-box verification includes simulating user operations to check the end-to-end business path;

[0045] The white-box verification includes random sampling to compare data consistency and performance baseline, and calculating a health score.

[0046] As a further improvement to one embodiment of the present invention, the method further includes, in which the feedback optimization of the time-series augmented machine learning model and the disaster recovery strategy includes,

[0047] Based on the recovery health score and the actual recovery time and actual recovery point of each business system recorded during the switchover process, feedback data is collected;

[0048] Using the feedback data, the parameters of the time-series augmented machine learning model are adjusted through an online learning algorithm;

[0049] Based on the feedback data, optimize the weighting factors of the preset basic recovery target or collaborative scheduling decision.

[0050] To achieve one of the above-mentioned objectives, an embodiment of the present invention also provides a multi-mode cloud disaster recovery adaptive system based on dynamic business awareness and collaborative scheduling. The system includes a business data acquisition module, a demand quantification module, and a resource scheduling and strategy generation module.

[0051] The business data acquisition module is used to collect KPIs of multiple business systems in real time, combine them with preset business impact analysis information, and generate the current actual disaster recovery requirement factors of each business system through a time-series augmented machine learning model to determine the current disaster recovery requirements of each business system.

[0052] The demand quantification module is used to monitor and quantify the available disaster recovery capabilities of all disaster recovery resource pools in real time, match the current disaster recovery needs with the available disaster recovery capabilities, and select the optimal disaster recovery mode and generate the final disaster recovery strategy through collaborative scheduling decisions for global resource optimization of each business system.

[0053] The resource scheduling and strategy generation module is used to automatically orchestrate corresponding disaster recovery resources for each business system according to the final disaster recovery strategy, so that the disaster recovery environment of each business system enters a ready state.

[0054] As a further improvement of one embodiment of the present invention, the system further includes a business data acquisition module comprising a cloud computing business system monitoring and perception module and a business impact analysis and disaster recovery baseline management module; the cloud computing business system monitoring and perception module is used to collect key performance indicators and operating status data of multiple cloud computing business systems in real time, providing input for disaster recovery requirement analysis; the business impact analysis and disaster recovery baseline management module is used to manage and maintain the business impact analysis knowledge base and basic recovery target data of the business system.

[0055] The demand quantification module includes a dynamic disaster recovery demand analysis and quantification module; the dynamic disaster recovery demand analysis and quantification module is used to dynamically generate disaster recovery demand factors and recovery targets for each business system based on real-time KPIs and business impact analysis information.

[0056] The resource scheduling and strategy generation module includes a disaster recovery resource status and capability assessment module, a disaster recovery strategy intelligent decision-making and collaboration engine, a disaster recovery resource orchestration and execution module, and a disaster recovery status verification and feedback module. The disaster recovery resource status and capability assessment module monitors and quantifies the health status and capabilities of the disaster recovery resource pool in real time, constructing a global resource view to support resource allocation. The disaster recovery strategy intelligent decision-making and collaboration engine intelligently selects the optimal disaster recovery mode and collaboratively schedules resources based on dynamic disaster recovery needs and resource capabilities, generating the final disaster recovery strategy. The disaster recovery resource orchestration and execution module automatically configures infrastructure, deploys applications, adjusts data synchronization, and performs switching during disasters or drills, based on the disaster recovery strategy. The disaster recovery status verification and feedback module performs disaster recovery environment verification, assesses the recovery status, generates reports with recovery time targets or recovery point targets, and provides feedback on optimization models and strategies.

[0057] Compared with existing technologies, this invention provides a multi-mode cloud disaster recovery adaptive method based on dynamic business awareness and collaborative scheduling. It dynamically senses the disaster recovery needs of multiple business systems, combines time-series augmented machine learning to generate demand factors, and achieves adaptive adjustment of recovery targets. Through multi-objective optimized scheduling, it intelligently selects the optimal disaster recovery mode, improving resource utilization efficiency. Automated orchestration and switching ensure rapid and reliable business recovery, reducing manual intervention. Closed-loop feedback optimizes models and strategies, improving disaster recovery accuracy. This invention significantly enhances business continuity, multi-mode collaborative efficiency, and cost-effectiveness, and is suitable for high-availability scenarios such as finance and e-commerce. Attached Figure Description

[0058] Figure 1 This is an overall flowchart of the multi-mode cloud disaster recovery adaptive method based on dynamic service awareness and collaborative scheduling described in this invention.

[0059] Figure 2 This is a schematic diagram of the architecture of the multi-mode cloud disaster recovery adaptive system based on dynamic service awareness and collaborative scheduling as described in this invention.

[0060] Figure 3 This is a schematic diagram of the multi-mode cloud disaster recovery adaptive system based on dynamic service awareness and collaborative scheduling described in this invention. Detailed Implementation

[0061] The present invention will now be described in detail with reference to the specific embodiments shown in the accompanying drawings. However, these embodiments do not limit the present invention, and any structural, methodological, or functional modifications made by those skilled in the art based on these embodiments are included within the scope of protection of the present invention.

[0062] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0063] In Embodiment 1 of the present invention, the present invention provides a multi-mode cloud disaster recovery adaptive method based on dynamic service awareness and collaborative scheduling, such as... Figure 1 As shown, the method includes,

[0064] S1: Real-time collection of KPIs from multiple business systems, combined with preset business impact analysis information, and generation of current actual disaster recovery demand factors for each business system through a time-series augmented machine learning model to determine the current disaster recovery needs of each business system.

[0065] S2: Monitor and quantify the available disaster recovery capabilities of all disaster recovery resource pools in real time, match the current disaster recovery needs with the available disaster recovery capabilities, and select the optimal disaster recovery mode and generate the final disaster recovery strategy through collaborative scheduling decisions for global resource optimization of various business systems.

[0066] S3: Based on the final disaster recovery strategy, automatically orchestrate the corresponding disaster recovery resources for each business system, so that the disaster recovery environment of each business system enters a ready state.

[0067] In one specific embodiment of the present invention, a time-series augmented machine learning model is used to generate the current actual disaster recovery demand factor, specifically,

[0068] Construct composite features for the time-series augmented machine learning model, including the volatility features of the KPIs, the periodic features associated with the business calendar, and the cross features of multiple KPIs;

[0069] The KPIs include API call success rate, response time, and number of concurrent users;

[0070] Based on the composite features, a basic time-series augmented machine learning model is generated through initial training using historical data and expert annotations.

[0071] Based on the aforementioned basic time-series augmented machine learning model, the current actual disaster recovery requirement factors and their confidence scores for each business system are generated through reasoning. Predictive samples with confidence scores below a preset threshold are identified through uncertainty sampling.

[0072] The low-confidence prediction samples are forwarded to experts for review, and the disaster recovery requirement factors after review are obtained. The review results are used to incrementally train the basic time-series augmented machine learning model through an online learning algorithm to generate an optimized time-series augmented machine learning model.

[0073] Based on the optimized time-series augmented machine learning model, combined with real-time KPIs and preset business impact analysis information, the current actual disaster recovery requirement factor is output.

[0074] It should be noted that, in order to achieve dynamic perception of the disaster recovery needs of multiple cloud computing business systems, this invention generates the current actual disaster recovery demand factor (CDRDF) of each business system through a time-series augmented machine learning model (TSE-ML) to quantify its real-time disaster recovery needs.

[0075] Furthermore, to ensure that the TSE-ML model can fully capture the dynamic operational characteristics of business systems, a composite feature set is constructed for each business system, including the following three types of features:

[0076] KPI volatility characteristics: By analyzing the time-series data of key performance indicators (KPIs), their volatility characteristics can be extracted, such as the short-term volatility of API call success rate (e.g., standard deviation or coefficient of variation) and the frequency of abnormal peaks in response time. These characteristics reflect the stability of the business system under different loads; for example, the response time volatility of a trading system may increase during peak periods.

[0077] Periodic characteristics associated with the business calendar: By combining the business calendar (such as holidays and promotional activities), time-related periodic characteristics can be extracted. For example, the number of concurrent users on e-commerce platforms surges periodically during "Double Eleven," or there are regular changes in traffic on weekdays. These characteristics capture the periodic patterns of business load.

[0078] Cross-features of KPIs: By combining multiple KPIs (such as API call success rate, response time, and concurrent users), cross-features are generated. For example, the correlation coefficient between response time and concurrent users, or the joint distribution of API call success rate and response time. These features reveal the interaction relationships between KPIs, improving the model's ability to understand complex business scenarios.

[0079] The aforementioned KPIs include, but are not limited to, API call success rate (reflecting service availability), response time (reflecting service performance), and concurrent users (reflecting business load). These features are transformed into an input format recognizable by the model through preprocessing (such as normalization and noise reduction), providing a foundation for subsequent training and inference.

[0080] Furthermore, based on the aforementioned composite feature set, initial training is performed using historical data and expert annotations to generate a basic TSE-ML model. Historical data includes KPI records for each business system at different time periods (e.g., API call success rate sequences over the past 30 days) and their corresponding disaster recovery requirement labels (e.g., CDRDF values ​​during high load). Expert annotations are provided by business domain experts; for example, CDRDF of a transaction system during peak periods is annotated based on Business Impact Analysis (BIA). The training process employs temporal augmentation algorithms (e.g., LSTM or Transformer variants) to capture the temporal dependencies of KPIs, generating a basic model capable of predicting CDRDF. Initial training ensures the model can preliminarily identify the mapping relationship between the operating modes of business systems and disaster recovery requirements.

[0081] Furthermore, to improve the model's prediction accuracy, uncertainty sampling is employed to identify low-confidence prediction samples output by the base TSE-ML model. For example, by calculating the entropy or variance of the prediction output, samples with CDRDF prediction confidence below a threshold are identified. These samples typically correspond to complex scenarios (such as sudden traffic surges) that the model struggles to accurately identify. Low-confidence samples are forwarded to experts for review, who provide corrected CDRDF values ​​based on real-time business context (such as promotional activities or system failures). Using the review results, the model is incrementally trained using online learning algorithms (such as gradient updates) to optimize model parameters and improve its adaptability to dynamic scenarios. This incremental training process ensures the model continuously learns new patterns, reducing prediction errors.

[0082] Furthermore, after incremental training, the TSE-ML model combines real-time KPI data (such as the response time series for the current hour) and pre-defined BIA information (such as the inherent importance level of the business system) to infer the current actual disaster recovery requirement factor for each business system. The CDRDF value reflects the real-time risk and recovery requirements of the business system and is used for subsequent RTO / RPO calculations and disaster recovery strategy decisions. The inference process is implemented through efficient model deployment (such as a containerized inference engine) to ensure real-time performance and low latency.

[0083] In one specific embodiment of the present invention, the current disaster recovery requirements include a dynamic recovery time target and a dynamic recovery point target, calculated using the following formula:

[0084]

[0085]

[0086] in, For dynamic recovery time target, To dynamically restore point targets, and As a preset basic recovery target, This represents the actual disaster recovery demand factor.

[0087] It should be noted that, in order to accurately quantify the disaster recovery needs of multiple cloud computing business systems, this invention generates adaptive disaster recovery requirement indicators for each business system by calculating dynamic recovery time targets and dynamic recovery point targets, so as to guide the subsequent selection of disaster recovery modes and resource allocation.

[0088] Furthermore, and The basic recovery targets are based on Business Impact Analysis (BIA) and represent the maximum allowable business interruption time and data loss for each business system under ideal conditions. This reflects the longest downtime required for the business system to return to normal operation. Reflecting the maximum acceptable data loss for a business system, these targets are pre-configured by business domain experts based on the functional importance of the business system, user scale, and the consequences of outages, and stored in the system as baseline values. Each business system has its own... and This reflects the inherent importance of disaster recovery and the differences in disaster recovery needs.

[0089] Furthermore, for multiple business systems, and The calculations are performed independently, with each business system based on its specific CDRDF. and Generate dedicated dynamic recovery goals. These dynamic goals provide a quantitative basis for subsequent matching of disaster recovery resource pools and selection of the optimal disaster recovery mode, ensuring that high-priority business systems receive resources that meet their stringent recovery requirements first.

[0090] In one specific embodiment of the present invention, the available disaster recovery capabilities of all disaster recovery resource pools are monitored and quantified in real time, specifically as follows:

[0091] For all disaster recovery resource pools, including intra-city active-active, off-site disaster recovery and edge node clusters, collect resource capacity indicators of each resource pool in real time, including computing capacity, storage capacity, network bandwidth, recovery time target support capability and recovery point target support capability;

[0092] The resource capability indicators are standardized into unified capability values ​​to generate quantitative capability data for each resource pool;

[0093] Based on the quantified capability data, a global resource view is constructed and updated to record the available disaster recovery capabilities of each resource instance;

[0094] The global resource view is dynamically updated based on resource allocation, and resource conflicts are detected to support subsequent degradation mechanisms.

[0095] It should be noted that, in order to achieve efficient allocation of disaster recovery resources for multiple cloud computing business systems, this invention generates a global resource view by real-time monitoring and quantification of the available disaster recovery capabilities of all disaster recovery resource pools, providing a precise basis for subsequent collaborative scheduling decisions and automated resource orchestration.

[0096] Furthermore, for all disaster recovery resource pools, including intra-city active-active data centers, off-site disaster recovery sites, and edge node clusters, resource capability metrics are collected in real time to comprehensively assess their available disaster recovery capabilities. The collection process is implemented through a distributed monitoring agent or cloud platform API, periodically collecting real-time status data from each resource pool to ensure the timeliness and accuracy of the metrics. For different types of resource pools (such as the low latency of edge nodes and the high reliability of off-site disaster recovery), the metric collection method is customized according to their technical characteristics to cover all disaster recovery scenarios.

[0097] Furthermore, to facilitate comparison and matching across resource pools, the collected resource capability indicators are standardized into unified capability values, generating quantitative capability data for each resource pool. The standardization process includes: normalizing each indicator, converting RTO and RPO support capabilities into a format comparable to business requirements, and converting network bandwidth and storage capacity into standardized values ​​through linear mapping (e.g., bandwidth divided by maximum bandwidth). Standardization ensures that indicators from different resource pools can be directly used for subsequent disaster recovery mode selection, improving matching efficiency.

[0098] Furthermore, based on the standardized quantified capability data, a global resource view is constructed to record the current available disaster recovery capability of each resource instance. The global resource view uses a structured data format (such as key-value pairs or database tables) and contains the identifier of each resource pool and its quantified capability data. The view is maintained in real time through a centralized management platform to ensure data consistency. The construction process considers the different needs of multiple business systems to ensure that the view can reflect the resource pool's capabilities for different... and Support capabilities.

[0099] Furthermore, based on resource allocation, the global resource view is dynamically updated to reflect the latest availability of the resource pool. The update process is implemented through an event-driven mechanism, such as triggering a view refresh after resource allocation is completed. Simultaneously, the system detects resource conflicts; for example, if multiple business systems simultaneously request bandwidth exceeding their limits, a conflict is identified. The detection results are used to trigger degradation mechanisms (such as selecting a suboptimal resource pool or reducing the allocation amount) and generate alarms to notify the administrator. Dynamic updates and conflict detection ensure the fairness and feasibility of resource allocation and support collaborative scheduling across multiple business systems.

[0100] In one specific embodiment of the present invention, by making collaborative scheduling decisions for global resource optimization of various business systems, the optimal disaster recovery mode is selected, and a final disaster recovery strategy is generated. Specifically,

[0101] Dynamic disaster recovery levels are mapped based on the actual disaster recovery needs of each business system, and multiple business systems are prioritized based on their inherent importance.

[0102] Based on the aforementioned priority ranking, the disaster recovery modes and resource instances are traversed to generate candidate solutions for each business system, and scores are calculated using a weighted scoring formula:

[0103]

[0104] in, The weighted score for assigning the j-th disaster recovery mode instance to the i-th business system is used to quantitatively evaluate the merits of the assignment scheme. and These represent the differences between the available resource tolerance for business interruption and the tolerance for data loss, respectively, and the business requirements. It is a monotonically increasing function. To standardize the costs of adopting this solution, For cost mapping function, This refers to the capacity redundancy of the resource instance after resource allocation. It is a nonlinear function. For pre-configured weighting factors;

[0105] Based on the priority ranking, select the highest-scoring scheme for each business system, reserve resources, and update the global resource view;

[0106] Based on the selected disaster recovery mode instance, configuration parameters are generated for each business system, including data synchronization frequency and load balancing strategy;

[0107] Integrate the disaster recovery modes, resource allocations, and configuration parameters of various business systems to generate the final disaster recovery strategy;

[0108] If there are conflicts in resource allocation or no solution that meets the requirements, the degradation mechanism is triggered, selecting the suboptimal disaster recovery mode or reducing resource allocation, and generating an alarm.

[0109] It should be noted that, in order to achieve efficient allocation of disaster recovery resources and optimization of strategies for multiple cloud computing business systems, this invention selects the optimal disaster recovery mode for each business system and generates the final disaster recovery strategy through collaborative scheduling decision-making of global resource optimization, ensuring that resource allocation is accurately matched with business needs, while optimizing resource utilization efficiency and cost-effectiveness.

[0110] Furthermore, for each business system, a Dynamic Disaster Recovery Level (DDRL) is mapped based on its current Actual Disaster Recovery Requirement Factor (CDRDF), and multiple business systems are prioritized based on their inherent importance. DDRLs are generated using predefined mapping rules; for example, CDRDF ≥ 1.8 maps to Level 1 (high priority), and 1.5 ≤ CDRDF < 1.8 maps to Level 2. Inherent importance is determined based on Business Impact Analysis (BIA), reflecting the functional criticality of the business system. For example, the core transaction system has higher inherent importance than the log analysis system. The prioritization rule first arranges systems by DDRL from high to low. If DDRLs are the same, they are then prioritized by inherent importance. For example, the transaction system (DDRL Level 1) takes precedence over the payment system (DDRL Level 2, high importance), which in turn takes precedence over the log system (DDRL Level 2, low importance). The priority ranking results form an ordered list to guide subsequent resource allocation, ensuring that high-priority business systems receive the disaster recovery resources that meet their needs first.

[0111] Furthermore, based on priority ranking, the system iterates through all disaster recovery modes and their resource instances, generating candidate solutions for each business system. Each candidate solution consists of business system i and resource instance j, and its merits are calculated using a weighted scoring formula. The scoring process ensures that high-priority business systems are evaluated first for high-performance resource instances, generating a list of candidate solutions.

[0112] Furthermore, based on priority ranking, the highest-scoring disaster recovery mode instance is selected for each business system from the candidate solutions. After selection, the system allocates corresponding resources to the business system and updates the global resource view to record the allocated resources. The update process ensures the real-time nature and consistency of resource allocation, avoiding over-allocation.

[0113] Furthermore, based on the selected disaster recovery mode instance, configuration parameters are generated for each business system, including data synchronization frequency and load balancing strategy. The data synchronization frequency is determined according to... Confirmed. The load balancing strategy is optimized based on business load and resource characteristics. For example, dynamic load balancing based on request volume is configured for the transaction system to ensure stability in high-concurrency scenarios. Configuration parameters are generated using a template-based approach to ensure compatibility with the technical characteristics of the resource instances.

[0114] Furthermore, the disaster recovery modes, resource allocations, and configuration parameters of various business systems are integrated to generate a final disaster recovery policy in a structured format. The integration process ensures that the policy covers all business systems and is consistent with priority ranking, with sub-policies for high-priority systems executed first. The final policy is output to the automated orchestration module for resource configuration and failover execution.

[0115] Furthermore, during resource allocation, the system detects potential conflicts, such as multiple business systems requesting bandwidth exceeding their limits. Conflict detection is based on real-time data from the global resource view. If over-allocation is detected, a degradation mechanism is triggered, selecting a suboptimal disaster recovery mode or reducing resource allocation (e.g., reducing the number of CPU cores). Simultaneously, an alarm is generated to notify the administrator, recording conflict details and the degradation decision. The degradation mechanism ensures the feasibility of resource allocation and maintains fairness among multiple business systems.

[0116] In one specific embodiment of the present invention, based on the final disaster recovery strategy, corresponding disaster recovery resources are automatically orchestrated for each business system. Specifically,

[0117] Obtain the deployment topology of the business application and its dependent components;

[0118] Based on the deployment topology, use IaC tools to configure infrastructure on selected resource instances;

[0119] Using the configuration results, deploy application services and dependent components, and configure a data synchronization mechanism to meet the current recovery point objectives;

[0120] Based on the deployment results, perform service health checks and warm-up operations to ensure that the disaster recovery environment of each business system is ready.

[0121] It should be noted that, in order to achieve rapid deployment and efficient operation of disaster recovery environments for multiple cloud computing business systems, this invention, based on the final disaster recovery strategy, uses automated means to orchestrate corresponding disaster recovery resources for each business system, ensuring that the disaster recovery environment can meet the dynamic recovery point objectives and enter a ready state, thus providing support for subsequent automated switching and verification.

[0122] Furthermore, to ensure accurate allocation of disaster recovery resources, the system first acquires the deployment topology of each business system and its dependent components. The deployment topology describes the service architecture and dependencies of the business applications, including application services (such as front-end API services and back-end databases), middleware (such as message queues), and infrastructure dependencies (such as load balancers and storage volumes). For example, the topology of a transaction system might include a front-end API server, payment processing service, and database cluster, dependent on Redis caching and network security groups. Topology information is automatically collected through service discovery tools or configuration management systems, generating a structured description (as shown in the diagram or YAML file) reflecting the component hierarchy and call chain of the business systems. The collection process ensures that the topology covers all business systems and their dependencies, supporting parallel processing of multiple business systems.

[0123] Furthermore, based on the acquired deployment topology, the system uses Infrastructure as Code (IaC) tools to configure the infrastructure on the selected disaster recovery mode instance. The configuration process is executed according to the resource allocation and configuration parameters (such as load balancing strategies) specified in the final disaster recovery strategy. The configuration script is based on the topology definition, ensuring that the infrastructure matches the service architecture of the business system and reducing configuration errors. The configuration process is implemented through a template-based approach, improving automation efficiency and consistency.

[0124] Furthermore, leveraging the infrastructure configuration results, the system deploys application services and their dependent components for each business system, and configures a data synchronization mechanism to meet the requirements. The data synchronization mechanism is based on... Configuration options include, for example, using asynchronous replication to ensure the database synchronizes once per second, or using log streaming to achieve near real-time data consistency. Deployment is automated through container orchestration tools or scripts, ensuring proper initialization of services and dependent components. Data synchronization configurations are aligned with disaster recovery features.

[0125] Furthermore, based on the deployment results, the system performs service health checks and preheating operations to ensure the disaster recovery environment of each business system is ready. Health checks include: verifying the running status of application services (e.g., normal API interface response), checking the availability of dependent components (e.g., successful database connection), and confirming whether the data synchronization mechanism meets requirements. Health checks are performed via automated scripts or monitoring tools, generating health status reports. Warm-up operations include initializing service caches, preloading data, or simulating low-load requests to ensure the disaster recovery environment can respond quickly during failover. For example, a transaction system's disaster recovery environment can immediately process API requests after warming up, reducing performance fluctuations after failover. Health checks and warm-up ensure the reliability and immediate availability of the disaster recovery environment.

[0126] In one specific embodiment of the present invention, upon receiving a disaster trigger signal or drill instruction, automated switching and scenario-based verification are performed on multiple business systems to assess the recovery status of each business system and provide feedback to optimize the time-series augmented machine learning model and disaster recovery strategy.

[0127] The automated switching and scenario-based verification include,

[0128] Upon receiving a disaster trigger signal or drill instruction, a directed acyclic graph containing multiple switching tasks is dynamically generated based on the call relationships between multiple business systems, and the switching tasks are concurrently scheduled through critical path analysis.

[0129] Based on the aforementioned concurrent scheduling, atomic and idempotent operations are performed, and checkpoints are set to support failure recovery;

[0130] After the switchover is completed, a scenario-based verification process combining black-box and white-box verification is used to assess the recovery status of each business system and generate a quantitative recovery health score.

[0131] The black-box verification includes simulating user operations to check the end-to-end business path;

[0132] The white-box verification includes random sampling to compare data consistency and performance baseline, and calculating a health score.

[0133] It should be noted that, in order to ensure that multiple cloud computing business systems can be quickly restored and their availability verified under disaster triggering or drill scenarios, this invention performs efficient disaster recovery switching for each business system through automated switching and scenario-based verification processes, assesses its recovery status, and optimizes the time-series augmented machine learning model (TSE-ML) and disaster recovery strategy through a feedback mechanism.

[0134] Furthermore, upon receiving a disaster trigger signal (such as a primary data center failure) or a drill instruction (such as a scheduled disaster recovery test), the system dynamically generates a directed acyclic graph (DAG) containing multiple switching tasks based on the call relationships between multiple business systems. The call relationships are obtained by analyzing the deployment topology of the business systems; for example, a transaction system depends on a payment service and a database, and the payment service depends on a message queue. Nodes in the DAG represent switching tasks, and edges represent dependencies between tasks. Through critical path analysis (such as the longest path algorithm), the system determines the concurrent order of task execution and optimizes the switching time. For example, tasks from high-priority business systems are scheduled first to ensure their needs are met. DAG generation and scheduling are implemented through automated scripts, supporting parallel switching of multiple service systems.

[0135] Furthermore, based on DAG-based concurrent scheduling, the system executes switching tasks, ensuring the atomicity and idempotency of operations. Atomicity operations guarantee that each switching task (such as service migration or DNS redirection) either completes entirely or rolls back to its initial state, preventing inconsistencies caused by partial switching. For example, the switching task for the payment system database includes data synchronization confirmation and connection redirection, which must be executed as a whole. Idempotency operations ensure that repeated execution does not produce side effects; for example, multiple DNS updates only apply the latest configuration. Checkpoints are set at critical stages of each task (such as data synchronization completion or service startup) to record the task status. If a task fails (such as due to network interruption), the system recovers from the most recent checkpoint and re-executes the unfinished tasks. Checkpoint data is stored in highly available storage, ensuring the reliability of multi-service system switching.

[0136] Furthermore, after the switchover is completed, the system uses a scenario-based verification process that combines black-box and white-box verification to assess the recovery status of each business system and generate a quantitative recovery health score.

[0137] Black-box verification: This involves simulating user actions to check the availability of the end-to-end business path. For example, simulating a user initiating a payment request on a transaction system verifies whether the API response is normal and whether the payment is successful. Verification covers key business scenarios (such as high-concurrency payments and data queries) to ensure a consistent user experience.

[0138] White-box verification: This involves randomly sampling and comparing data consistency (e.g., record consistency between primary and backup database nodes) and performance baselines (e.g., deviation of response time from historical averages) to assess the internal state of the system. For example, sampling and inspecting 100 records in a payment system database to confirm that the data loss level meets the requirements. Compare API response time with baseline to ensure performance meets standards.

[0139] Preferably, the recovery health score is calculated using a weighted formula, which combines the results of black-box and white-box verification.

[0140] Furthermore, based on the recovery health score and actual recovery data during the switchover process (such as actual RTO / RPO), the system generates feedback data to optimize the TSE-ML model and disaster recovery strategy. For example, if the actual RTO of the trading system exceeds... This indicates that CDRDF predictions may be underestimated, and the feedback data is used to adjust model parameters. Disaster recovery strategy optimization includes updating priority ranking rules (such as adjusting DDRL thresholds) or resource allocation strategies. Feedback is implemented through an online learning mechanism to ensure that the model and strategies continuously adapt to business changes. Feedback data also records switchover failure cases (such as network conflicts) to improve the DAG generation logic.

[0141] In one specific embodiment of the present invention, feedback optimization of the time-series augmented machine learning model and the disaster recovery strategy specifically includes:

[0142] Based on the recovery health score and the actual recovery time and actual recovery point of each business system recorded during the switchover process, feedback data is collected;

[0143] Using the feedback data, the parameters of the time-series augmented machine learning model are adjusted through an online learning algorithm;

[0144] Based on the feedback data, optimize the weighting factors of the preset basic recovery target or collaborative scheduling decision.

[0145] It should be noted that, in order to ensure that the disaster recovery strategies of multiple business systems in the cloud computing environment can continuously adapt to the dynamically changing needs, this invention uses a feedback optimization mechanism to continuously improve the temporal augmented machine learning model (TSE-ML) and disaster recovery strategies by utilizing the switching verification results, so as to improve the model prediction accuracy and resource allocation efficiency.

[0146] Furthermore, the feedback data is based on the recovery health score generated from scenario-based verification, as well as the actual recovery time and actual recovery point recorded for each business system during the switchover process. The recovery health score reflects the availability of the business system's disaster recovery environment. The actual recovery time records the actual switchover completion time, and the actual recovery point records the actual amount of data loss. In addition, the feedback data includes abnormal events during the switchover process (such as task failures and network latency) and their context (such as high-concurrency scenarios). This data is collected through automated monitoring tools, stored as structured logs, and compared with the CDRDF of each business system. / Correlation is used to analyze and predict deviations. Feedback data covers all business systems, ensuring comprehensive optimization across multiple system scenarios.

[0147] Furthermore, utilizing the collected feedback data, the system adjusts the parameters of the TSE-ML model through an online learning algorithm to improve its prediction accuracy for CDRDF. The online learning employs an incremental update strategy, such as gradient descent-based online optimization, comparing the actual recovery data in the feedback data with the predicted values ​​to calculate the prediction error. Samples with larger errors are prioritized for updating model weights, optimizing the model's adaptability to high-risk scenarios (such as sudden traffic surges). For example, if the actual recovery point of the payment system exceeds... The model adjusts its weights on KPI volatility characteristics to enhance its predictive ability for similar scenarios. This adjustment process is implemented through a lightweight inference engine, ensuring low computational overhead and supporting parallel optimization across multiple business systems.

[0148] Furthermore, based on feedback data, the system optimizes the preset basic recovery targets and the weighting factors for collaborative scheduling decisions. and The optimization is based on a comparison of actual recovery data and Business Impact Analysis (BIA). For example, if the actual recovery point of the log system is consistently lower than... This indicates that its basic target is too conservative and can be adjusted to a more lenient value to reduce resource costs. The optimization of the weighting factors is based on the recovery health score and business scenario requirements. For example, if the Health_Score of a financial business system is low due to RTO not meeting the target, it can be increased... Prioritizing recovery time. Optimization is achieved through a rule engine or heuristic algorithm, combining the inherent importance of the business system and DDRL to generate new weight configurations. The optimization results are stored in the policy database for subsequent scheduling decisions.

[0149] In a second embodiment of the present invention, the present invention provides a multi-mode cloud disaster recovery adaptive system based on dynamic service awareness and collaborative scheduling, such as... Figure 2As shown, the system includes a business data acquisition module, a demand quantification module, and a resource scheduling and strategy generation module;

[0150] The business data acquisition module is used to collect KPIs of multiple business systems in real time, combine them with preset business impact analysis information, and generate the current actual disaster recovery requirement factors of each business system through a time-series augmented machine learning model to determine the current disaster recovery requirements of each business system.

[0151] The demand quantification module is used to monitor and quantify the available disaster recovery capabilities of all disaster recovery resource pools in real time, match the current disaster recovery needs with the available disaster recovery capabilities, and select the optimal disaster recovery mode and generate the final disaster recovery strategy through collaborative scheduling decisions for global resource optimization of each business system.

[0152] The resource scheduling and strategy generation module is used to automatically orchestrate corresponding disaster recovery resources for each business system according to the final disaster recovery strategy, so that the disaster recovery environment of each business system enters a ready state.

[0153] In one specific embodiment of the present invention, such as Figure 3 As shown,

[0154] The business data acquisition module includes a cloud computing business system monitoring and perception module and a business impact analysis and disaster recovery baseline management module. The cloud computing business system monitoring and perception module is used to collect key performance indicators and operating status data of multiple cloud computing business systems in real time, providing input for disaster recovery requirements analysis. The business impact analysis and disaster recovery baseline management module is used to manage and maintain the business impact analysis knowledge base and basic recovery target data of the business systems.

[0155] The demand quantification module includes a dynamic disaster recovery demand analysis and quantification module; the dynamic disaster recovery demand analysis and quantification module is used to dynamically generate disaster recovery demand factors and recovery targets for each business system based on real-time KPIs and business impact analysis information.

[0156] The resource scheduling and strategy generation module includes a disaster recovery resource status and capability assessment module, a disaster recovery strategy intelligent decision-making and collaboration engine, a disaster recovery resource orchestration and execution module, and a disaster recovery status verification and feedback module. The disaster recovery resource status and capability assessment module monitors and quantifies the health status and capabilities of the disaster recovery resource pool in real time, constructing a global resource view to support resource allocation. The disaster recovery strategy intelligent decision-making and collaboration engine intelligently selects the optimal disaster recovery mode and collaboratively schedules resources based on dynamic disaster recovery needs and resource capabilities, generating the final disaster recovery strategy. The disaster recovery resource orchestration and execution module automatically configures infrastructure, deploys applications, adjusts data synchronization, and performs switching during disasters or drills, based on the disaster recovery strategy. The disaster recovery status verification and feedback module performs disaster recovery environment verification, assesses the recovery status, generates reports with recovery time targets or recovery point targets, and provides feedback on optimization models and strategies.

[0157] In summary, this invention provides a multi-mode cloud disaster recovery adaptive method based on dynamic business awareness and collaborative scheduling. It dynamically senses the disaster recovery needs of multiple business systems, combines time-series augmented machine learning to generate demand factors, and achieves adaptive adjustment of recovery targets. Through multi-objective optimized scheduling, it intelligently selects the optimal disaster recovery mode, improving resource utilization efficiency. Automated orchestration and switching ensure rapid and reliable business recovery, reducing manual intervention. Closed-loop feedback optimizes models and strategies, improving disaster recovery accuracy. This invention significantly enhances business continuity, multi-mode collaborative efficiency, and cost-effectiveness, and is suitable for high-availability scenarios such as finance and e-commerce.

[0158] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the modules described above can be referred to the corresponding process in the aforementioned method implementation, and will not be repeated here.

[0159] The modules described as separate components may or may not be physically separate. Similarly, the components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0160] Furthermore, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated module can be implemented in hardware or in a combination of hardware and software functional modules.

[0161] The integrated modules implemented as software functional modules described above can be stored in a computer-readable storage medium. These software functional modules, stored in a storage medium, include several instructions to cause a computer system (which may be a personal computer, server, or network system, etc.) or processor to execute some steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0162] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A multi-mode cloud disaster recovery adaptive method based on dynamic service awareness and collaborative scheduling, characterized in that: include, Real-time collection of KPIs from multiple business systems, combined with pre-set business impact analysis information, and generation of current actual disaster recovery demand factors for each business system through time-series augmented machine learning models to determine the current disaster recovery needs of each business system. Real-time monitoring and quantification of the available disaster recovery capabilities of all disaster recovery resource pools, matching current disaster recovery needs with available disaster recovery capabilities, and selecting the optimal disaster recovery mode and generating the final disaster recovery strategy through collaborative scheduling decisions for global resource optimization of various business systems. Based on the final disaster recovery strategy, corresponding disaster recovery resources are automatically orchestrated for each business system, so that the disaster recovery environment of each business system is ready.

2. The multi-mode cloud disaster recovery adaptive method based on dynamic service awareness and collaborative scheduling according to claim 1, characterized in that: The process of generating current actual disaster recovery demand factors through a time-series augmented machine learning model includes... Construct composite features for the time-series augmented machine learning model, including the volatility features of the KPIs, the periodic features associated with the business calendar, and the cross features of multiple KPIs; The KPIs include API call success rate, response time, and number of concurrent users; Based on the composite features, a basic time-series augmented machine learning model is generated through initial training using historical data and expert annotations. Based on the aforementioned basic time-series augmented machine learning model, the current actual disaster recovery requirement factors and their confidence scores for each business system are generated through reasoning. Predictive samples with confidence scores below a preset threshold are identified through uncertainty sampling. The low-confidence prediction samples are forwarded to experts for review, and the disaster recovery requirement factors after review are obtained. The review results are used to incrementally train the basic time-series augmented machine learning model through an online learning algorithm to generate an optimized time-series augmented machine learning model. Based on the optimized time-series augmented machine learning model, combined with real-time KPIs and preset business impact analysis information, the current actual disaster recovery requirement factor is output.

3. The multi-mode cloud disaster recovery adaptive method based on dynamic service awareness and collaborative scheduling according to claim 2, characterized in that: The current disaster recovery requirements include dynamic recovery time targets and dynamic recovery point targets, calculated using the following formula: ； ； in, For dynamic recovery time target, To dynamically restore point targets, and As a preset basic recovery target, This represents the actual disaster recovery needs factor.

4. The multi-mode cloud disaster recovery adaptive method based on dynamic service awareness and collaborative scheduling according to claim 3, characterized in that: The real-time monitoring and quantification of the available disaster recovery capabilities of all disaster recovery resource pools includes, For all disaster recovery resource pools, including intra-city active-active, off-site disaster recovery and edge node clusters, collect resource capacity indicators of each resource pool in real time, including computing capacity, storage capacity, network bandwidth, recovery time target support capability and recovery point target support capability; The resource capability indicators are standardized into unified capability values ​​to generate quantitative capability data for each resource pool; Based on the quantified capability data, a global resource view is constructed and updated to record the available disaster recovery capabilities of each resource instance; The global resource view is dynamically updated based on resource allocation, and resource conflicts are detected to support subsequent degradation mechanisms.

5. The multi-mode cloud disaster recovery adaptive method based on dynamic service awareness and collaborative scheduling according to claim 4, characterized in that: The process of collaborative scheduling decisions involving global resource optimization across various business systems, selecting the optimal disaster recovery mode, and generating the final disaster recovery strategy includes: Dynamic disaster recovery levels are mapped based on the actual disaster recovery needs of each business system, and multiple business systems are prioritized based on their inherent importance. Based on the aforementioned priority ranking, the disaster recovery modes and resource instances are traversed to generate candidate solutions for each business system, and scores are calculated using a weighted scoring formula: ； in, The weighted score for assigning the j-th disaster recovery mode instance to the i-th business system is used to quantitatively evaluate the merits of the assignment scheme. and These represent the differences between the available resource tolerance for business interruption and the tolerance for data loss, respectively, and the business requirements. It is a monotonically increasing function. To standardize the costs of adopting this solution, For cost mapping function, This refers to the capacity redundancy of the resource instance after resource allocation. It is a nonlinear function. For pre-configured weighting factors; Based on the priority ranking, select the highest-scoring scheme for each business system, reserve resources, and update the global resource view; Based on the selected disaster recovery mode instance, configuration parameters are generated for each business system, including data synchronization frequency and load balancing strategy; Integrate the disaster recovery modes, resource allocations, and configuration parameters of various business systems to generate the final disaster recovery strategy; If there are conflicts in resource allocation or no solution that meets the requirements, the degradation mechanism is triggered, selecting the suboptimal disaster recovery mode or reducing resource allocation, and generating an alarm.

6. The multi-mode cloud disaster recovery adaptive method based on dynamic service awareness and collaborative scheduling according to claim 5, characterized in that: The automatic orchestration of corresponding disaster recovery resources for each business system according to the final disaster recovery strategy includes, Obtain the deployment topology of the business application and its dependent components; Based on the deployment topology, use IaC tools to configure infrastructure on selected resource instances; Using the configuration results, deploy application services and dependent components, and configure a data synchronization mechanism to meet the current recovery point objectives; Based on the deployment results, perform service health checks and warm-up operations to ensure that the disaster recovery environment of each business system is ready.

7. The multi-mode cloud disaster recovery adaptive method based on dynamic service awareness and collaborative scheduling according to claim 1, characterized in that: It also includes, Upon receiving a disaster trigger signal or drill instruction, the system performs automated switching and scenario-based verification for multiple business systems, assesses the recovery status of each business system, and provides feedback to optimize the time-series augmented machine learning model and disaster recovery strategy. The automated switching and scenario-based verification include, Upon receiving a disaster trigger signal or drill instruction, a directed acyclic graph containing multiple switching tasks is dynamically generated based on the call relationships between multiple business systems, and the switching tasks are concurrently scheduled through critical path analysis. Based on the aforementioned concurrent scheduling, atomic and idempotent operations are performed, and checkpoints are set to support failure recovery; After the switchover is completed, a scenario-based verification process combining black-box and white-box verification is used to assess the recovery status of each business system and generate a quantitative recovery health score. The black-box verification includes simulating user operations to check the end-to-end business path; The white-box verification includes random sampling to compare data consistency and performance baseline, and calculating a health score.

8. The multi-mode cloud disaster recovery adaptive method based on dynamic service awareness and collaborative scheduling according to claim 7, characterized in that: The feedback optimization of the time-series augmented machine learning model and disaster recovery strategy includes... Based on the recovery health score and the actual recovery time and actual recovery point of each business system recorded during the switchover process, feedback data is collected; Using the feedback data, the parameters of the time-series augmented machine learning model are adjusted through an online learning algorithm; Based on the feedback data, optimize the weighting factors of the preset basic recovery target or collaborative scheduling decision.

9. A multi-mode cloud disaster recovery adaptive system based on dynamic service awareness and collaborative scheduling, characterized in that: This includes a business data acquisition module, a demand quantification module, and a resource scheduling and strategy generation module; The business data acquisition module is used to collect KPIs of multiple business systems in real time, combine them with preset business impact analysis information, and generate the current actual disaster recovery requirement factors of each business system through a time-series augmented machine learning model to determine the current disaster recovery requirements of each business system. The demand quantification module is used to monitor and quantify the available disaster recovery capabilities of all disaster recovery resource pools in real time, match the current disaster recovery needs with the available disaster recovery capabilities, and select the optimal disaster recovery mode and generate the final disaster recovery strategy through collaborative scheduling decisions for global resource optimization of each business system. The resource scheduling and strategy generation module is used to automatically orchestrate corresponding disaster recovery resources for each business system according to the final disaster recovery strategy, so that the disaster recovery environment of each business system enters a ready state.

10. The multi-mode cloud disaster recovery adaptive system based on dynamic service awareness and collaborative scheduling according to claim 9, characterized in that: It also includes, The business data acquisition module includes a cloud computing business system monitoring and perception module and a business impact analysis and disaster recovery baseline management module. The cloud computing business system monitoring and perception module is used to collect key performance indicators and operating status data of multiple cloud computing business systems in real time, providing input for disaster recovery requirements analysis. The business impact analysis and disaster recovery baseline management module is used to manage and maintain the business impact analysis knowledge base and basic recovery target data of the business systems. The demand quantification module includes a dynamic disaster recovery demand analysis and quantification module; the dynamic disaster recovery demand analysis and quantification module is used to dynamically generate disaster recovery demand factors and recovery targets for each business system based on real-time KPIs and business impact analysis information. The resource scheduling and strategy generation module includes a disaster recovery resource status and capability assessment module, a disaster recovery strategy intelligent decision-making and collaboration engine, a disaster recovery resource orchestration and execution module, and a disaster recovery status verification and feedback module. The disaster recovery resource status and capability assessment module monitors and quantifies the health status and capabilities of the disaster recovery resource pool in real time, constructing a global resource view to support resource allocation. The disaster recovery strategy intelligent decision-making and collaboration engine intelligently selects the optimal disaster recovery mode and collaboratively schedules resources based on dynamic disaster recovery needs and resource capabilities, generating the final disaster recovery strategy. The disaster recovery resource orchestration and execution module automatically configures infrastructure, deploys applications, adjusts data synchronization, and performs switching during disasters or drills, based on the disaster recovery strategy. The disaster recovery status verification and feedback module performs disaster recovery environment verification, assesses the recovery status, generates reports with recovery time targets or recovery point targets, and provides feedback on optimization models and strategies.

Images