Cloud service full-life-cycle intelligent operation and maintenance method and system based on dynamic resource arrangement

Abstract

The invention relates to the technical field of cloud computing and mobile communication, in particular to a cloud service full-life-cycle intelligent operation and maintenance method and system based on dynamic resource orchestration, and the method comprises the following steps: resource portrait construction, network demand matching, dynamic resource orchestration, and intelligent operation and maintenance guarantee. The method has the beneficial effects that a cloud resource access and integrated scheduling mechanism is continuously improved by constructing a computing power perception-fusion scheduling-intelligent operation and maintenance full-process method system, the computing power use cost and threshold are reduced, the agile supply of the computing power is realized, and the core problems of insufficient computing network collaboration, low operation and maintenance automation and the like in the existing scheme are effectively solved.

Description

Technical Field

[0001] This invention relates to the fields of cloud computing and mobile communication technology, specifically to a method and system for intelligent operation and maintenance of cloud services throughout their entire lifecycle based on dynamic resource orchestration. Background Technology

[0002] With the widespread adoption of 5G networks and the accelerating trend of the Internet of Things, the volume of business carried by mobile cloud services is growing exponentially, placing higher demands on the dynamic allocation and full lifecycle management of cloud resources. Traditional static resource allocation models have significant drawbacks due to their inability to perceive business changes in real time: First, resource scheduling is lagging, leading to frequent service interruptions under sudden traffic surges; second, maintenance costs are high, with manual intervention accounting for over 60%, insufficient automation, and simple, repetitive resource configuration and troubleshooting consuming significant manpower; third, lifecycle management is fragmented, with access, maintenance, and operational data belonging to different systems, failing to form a closed loop, resulting in low efficiency in problem localization and a lack of data support for resource optimization.

[0003] Furthermore, existing cloud operations and maintenance (O&M) solutions often focus on optimizing single aspects, lacking integrated collaborative scheduling of computing and network resources, multi-element joint orchestration, and intelligent prediction capabilities. For example, some solutions only achieve independent scheduling of intelligent computing resources without considering network bandwidth and latency requirements, resulting in AI training tasks failing to run efficiently due to network bottlenecks. While some solutions possess basic O&M automation capabilities, they lack time-series prediction for early warning and still rely on manual handling of sudden failures. These shortcomings make it difficult for existing solutions to support the flexible deployment of computing power hubs in national strategic scenarios such as "Eastern Data, Western Computing," and fail to meet the demands of industrial digital transformation for efficient and low-cost computing power services.

[0004] Therefore, there is an urgent need for a cloud service intelligent operation and maintenance solution that covers the entire lifecycle and focuses on methodological innovation to achieve agile resource provisioning, cost reduction, and full-process automation. Summary of the Invention

[0005] The purpose of this invention is to provide a cloud service full lifecycle intelligent operation and maintenance method and system based on dynamic resource orchestration, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a cloud service full lifecycle intelligent operation and maintenance method based on dynamic resource orchestration, comprising the following steps: Resource profiling: Lightweight data collection logic is embedded in the intelligent computing GPU cluster and general computing CPU cluster to collect intelligent computing indicators and general computing indicators in real time. Based on the data, a computing power resource profile including "real-time resource load, historical utilization trend, and task type proportion" is constructed. The data update frequency is consistent with the indicator collection frequency, and it supports multi-dimensional display by "cluster dimension, business dimension, and resource type dimension". Network demand matching: A pre-defined "business type - network demand" mapping rule base is available, which supports dynamic updates. Temporary rules can be added for new business types, and they will be included in the official rule base after verification. After the rules are updated, they are automatically synchronized to the "business - network" matching logic without restarting the service. Intelligent computing tasks are associated with high-bandwidth policies, and general computing tasks are associated with low-latency policies. When a business initiates a computing power request, the network demand tag is automatically matched through the "business type identifier". Dynamic resource orchestration: Monitors the utilization rate of the intelligent computing cluster; when it is less than 30%, it adapts idle GPU resources to intelligent computing resources using Kubernetes containerization technology; when the QPS of the general computing cluster exceeds the threshold, it calls upon idle computing power of the intelligent computing cluster; it also has a general computing task resource release mechanism; it converts network demand tags into SDN configuration instructions and issues them, verifies the effectiveness status through the SDN controller status feedback interface, and reissues them if they do not take effect within 5 seconds, and records the exception log after 3 failed retries; it ensures the exclusive use of computing power for intelligent computing tasks. Intelligent Operation and Maintenance Support: Automatically migrates faulty intelligent computing tasks and expands general computing resources; outputs predicted metric values ​​for the next 10 minutes; anomaly warning thresholds include GPU memory usage exceeding 90%, CPU load exceeding 85%, and QPS exceeding the threshold by 1.5 times, with warning information synchronized to operation and maintenance personnel and including suggested handling solutions; uses AES-256 algorithm to encrypt and store scheduling logs and user information, employing a "master key + data key" hierarchical management system, with a default key rotation cycle of 90 days, and the rotation process does not affect data access; scheduling logs are stored in JSON format, with fields including resource call time, computing power type, business ID, scheduling instruction, anomaly description, and recovery steps, supporting multi-dimensional retrieval, and a default retention time of 365 days; and controls computing power access permissions through the RBAC permission model.

[0007] Preferably, it also includes standardized computing power invocation steps: building a unified computing power interface in RESTful style, with the intelligent computing interface compatible with TensorFlow2.x / PyTorch1.8+ framework, and the universal computing interface adapted to general business systems; the interface has built-in protocol conversion logic, supporting "businesses can invoke heterogeneous computing power without modifying code".

[0008] Preferably, the standardized computing power call process also includes an interface testing tool that supports business verification of compatibility and provides detailed suggestions on output parameter formats, permissions, and protocol adaptation dimensions.

[0009] Preferably, in the dynamic resource orchestration step, when dynamically allocating computing resources, priority is given to ensuring the resource needs of critical businesses. Based on business priority and resource occupancy, the resource allocation strategy is intelligently adjusted to ensure the high availability and high performance of critical businesses.

[0010] Preferably, in the intelligent operation and maintenance support steps, for abnormal early warning information, in addition to synchronizing it to operation and maintenance personnel and including suggested handling solutions, the corresponding emergency handling process is automatically triggered according to the type and severity of the abnormality, automatically switching to backup resources and starting degraded services to minimize the impact on business.

[0011] A cloud service lifecycle intelligent operation and maintenance system based on dynamic resource orchestration includes the following modules: Data Acquisition and Profile Building Module: Lightweight acquisition logic is embedded in the intelligent computing GPU cluster and general computing CPU cluster to collect intelligent computing indicators and general computing indicators in real time. Based on the collected data, a computing power resource profile is built. This profile includes real-time resource load, historical utilization trend, and task type proportion information. The data update frequency is consistent with the indicator acquisition frequency, which is second-level update, and supports multi-dimensional display by cluster dimension, business dimension, and resource type dimension. Network demand matching module: It has a preset "business type-network demand" mapping rule library. The mapping rule library supports dynamic updates. When a new business type is added, temporary rules can be added. After verification, the rules are included in the formal rule library. After the rules are updated, they are automatically synchronized to the "business-network" matching logic without restarting the service. Intelligent computing tasks are associated with high bandwidth policies and general computing tasks are associated with low latency policies. When a business initiates a computing power request, the network demand tag is automatically matched through the business type identifier. Dynamic resource orchestration module: Executes dynamic resource orchestration logic, monitors the utilization rate of the intelligent computing cluster, and adapts idle GPU resources to intelligent computing available resources through Kubernetes containerization technology when the utilization rate is less than 30%. When the QPS of the general computing cluster exceeds the threshold, it calls upon idle intelligent computing power and also has a general computing task resource release mechanism. It converts network demand tags into SDN configuration instructions and issues them, verifies the effectiveness status through the SDN controller status feedback interface, and reissues them if they do not take effect within 5 seconds. After 3 failed retries, it records an exception log containing the instruction content, failure reason, and timestamp. It ensures the exclusive use of computing power for intelligent computing tasks. The intelligent operation and maintenance support module automatically migrates faulty intelligent computing tasks and expands general computing resources; outputs predicted metric values ​​for the next 10 minutes; sets anomaly warning thresholds, including GPU memory usage exceeding 90%, CPU load exceeding 85%, and QPS exceeding the threshold by 1.5 times, with warning information synchronized to operation and maintenance personnel and including suggested handling solutions; uses AES-256 algorithm to encrypt and store scheduling logs and user information, with AES-256 encryption using a "master key + data key" hierarchical management, and a default key rotation cycle of 90 days, during which data access is not affected; scheduling logs are stored in JSON format, with fields including resource call time, computing power type, business ID, scheduling instruction, anomaly description, and recovery steps, supporting multi-dimensional retrieval, and a default retention time of 365 days; and controls computing power access permissions through the RBAC permission model.

[0012] Preferably, it also includes a standardized computing power invocation module, which builds a unified computing power interface in RESTful style. The intelligent computing interface is compatible with TensorFlow 2.x / PyTorch 1.8+ framework, and the universal computing interface is adapted to general business systems. The interface has built-in protocol conversion logic, which allows businesses to invoke heterogeneous computing power without modifying the code.

[0013] Preferably, the standardized computing power call module also includes an interface testing tool, which supports business verification of compatibility and provides detailed suggestions on output parameter format, permissions, and protocol adaptation dimensions.

[0014] Preferably, the data acquisition and profiling module adopts a distributed acquisition architecture when collecting data to improve the efficiency and stability of data acquisition, ensuring that various indicator data can be obtained in real time and accurately in a large-scale cluster environment; at the same time, it has an automatic identification and repair mechanism for the collected abnormal data to ensure the accuracy and reliability of the computing power resource profiling.

[0015] Preferably, when dealing with abnormal situations, the intelligent operation and maintenance support module, in addition to issuing warnings and handling them according to preset rules, also has intelligent decision-making capabilities. It can automatically generate the optimal handling strategy based on historical data and the current business scenario. At the same time, for frequently occurring abnormal situations, it conducts in-depth analysis to uncover potential root causes of problems and provides targeted optimization suggestions to continuously improve the stability and performance of cloud services.

[0016] Compared with the prior art, the beneficial effects of the present invention are: The cloud service full lifecycle intelligent operation and maintenance method and system proposed in this invention, based on dynamic resource orchestration, continuously improves the cloud resource access and integrated scheduling mechanism by constructing a full-process method system of "computing power perception-integrated scheduling-intelligent operation and maintenance", reduces the cost and threshold of computing power use, realizes agile computing power supply, and effectively solves the core problems of insufficient computing network collaboration and low operation and maintenance automation in existing solutions. Attached Figure Description

[0017] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the present invention clear and complete, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only some, not all, embodiments of the present invention, and are merely illustrative of the embodiments of the present invention. They are not intended to limit the embodiments of the present invention. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Example 1: This invention provides a technical solution: an intelligent operation and maintenance method for the entire lifecycle of cloud services based on dynamic resource orchestration. It constructs a three-layer architecture of "computing power awareness - fusion scheduling - intelligent operation and maintenance," with each layer's methods interconnected and working synergistically to form an intelligent operation and maintenance closed loop covering the entire lifecycle of cloud services. The specific technical solution is as follows: 1. Methods of perceiving intelligence: Data collection logic deployment: Lightweight data collection logic is embedded in the intelligent computing GPU cluster and the general computing CPU cluster, requiring no additional hardware devices and achieving real-time data collection of indicators only through software scripts; Key metrics are defined as follows: Intelligent computing resources include GPU utilization (accuracy up to 1%), video memory usage (distinguishing between used / cache / idle), and training task progress (statistics by epoch / step); General computing resources include CPU load (average over 1 minute / 5 minutes / 15 minutes), memory usage (including swap partition), and business request QPS (statistics by business type). Data reporting and profile building: The profile includes three core dimensions: real-time resource load, utilization trend over the past 24 hours, and the proportion of current task types. The profile data is updated synchronously with the collection frequency to ensure data timeliness.

[0020] Mapping rule base construction: Pre-set "business type-network requirement" mapping rules, which are stored in the form of a configurable form and support subsequent dynamic updates; Among them, intelligent computing tasks (such as AI training and image recognition) are marked with "high bandwidth" requirements and associated with SDN network bandwidth guarantee parameters (default allocation of 10Gbps bandwidth, with the highest priority); general computing tasks (such as web services and database queries) are marked with "low latency" requirements and associated with SDN network latency optimization parameters (default control latency <50ms, using a dynamic routing optimization strategy); Automatic matching logic: When a business initiates a computing power request, the business attributes are first identified by the "business type identifier" in the request header, and then the rule base is called for matching to automatically tag the business with network demand. If a new business type is encountered, temporary rules can be added manually, and temporary rules can be included in the formal rule base after verification to ensure matching flexibility.

[0021] 2. Fusion Scheduling Method Resource balancing judgment logic: Real-time monitoring of the overall utilization rate of the intelligent computing cluster. When the utilization rate is less than 30%, the idle GPU resource adaptation process is triggered. Idle GPU resources are added to the "Intelligent Computing Available Resource Pool" through K8s containerization technology (using a custom resource scheduler) and the computing power resource profile is updated. At the same time, the QPS of each service in the general computing cluster is monitored in real time. When the QPS of a certain service exceeds the preset threshold (the threshold is set at 1.2 times the historical peak value of the service and supports dynamic adjustment), resources are automatically allocated from the "Intelligent Computing Available Resource Pool" to the service until the QPS falls back to within the threshold. The computing network collaborative scheduling logic is as follows: It receives the service network requirement tags generated during the computing power perception phase and converts them into configuration instructions that can be executed by the SDN controller. For intelligent computing tasks, the instructions include bandwidth allocation values, traffic priorities, and port mapping rules. For general computing tasks, the instructions include latency limits, routing optimization strategies, and packet loss rate control targets. After the instructions are issued, the configuration effectiveness status is verified through the status feedback interface of the SDN controller. If the configuration does not take effect within 5 seconds, it is reissued and retried 3 times. If it fails, an exception log is recorded. Scheduling priority control logic: Two-level priority rules are set. Intelligent computing training tasks are of the first priority. After starting, the allocated GPU resources are locked immediately and cannot be preempted by any general computing tasks until the task ends normally or is manually terminated. General computing tasks are of the second priority and can only use the idle intelligent computing resources in the "general computing available resource pool".

[0022] Unified Interface Specification Definition: A unified interface for heterogeneous computing power is constructed, adopting a RESTful style and supporting HTTP / HTTPS protocols. Among them, the intelligent computing interface is compatible with TensorFlow (2.x and above) and PyTorch (1.8 and above) frameworks, and business applications can directly call it through the framework's native "computing power call SDK" without modifying the code; the universal computing interface is adapted to general business systems (such as Java / Python applications) and supports calls through API key authentication. Interface compatibility assurance: The interface has built-in protocol conversion logic. When the business request protocol does not match the interface specification, it will automatically perform protocol conversion (such as converting SOAP protocol to RESTful protocol). At the same time, it provides interface testing tools to support business to verify compatibility before formal calls and output detailed adaptation suggestions (such as parameter format errors, missing permissions, etc.).

[0023] 3. Intelligent Operation and Maintenance Methods Anomaly prediction logic: Based on resource indicators collected during the computing power awareness phase, anomaly thresholds are set (GPU memory usage exceeds 90%, CPU load exceeds 85%, QPS exceeds the threshold by 1.5 times). When the predicted value exceeds any threshold, an alert is immediately triggered. The alert information is synchronized to the operation and maintenance personnel via SMS and email, and includes the anomaly type, associated business ID, expected occurrence time, and suggested handling solution. Automatic recovery logic: For intelligent computing tasks, the GPU running status is monitored in real time (checked every 2 seconds via a GPU health check script). When a GPU failure is detected (such as memory overflow, hardware error, or temperature exceeding the threshold), the task migration process is automatically triggered—first, the current intermediate data of the task is saved (stored to a distributed file system), then the node with the lowest load is selected from the list of standby GPU nodes to restore the task and continue running. The total migration time is less than 30 seconds. For general computing tasks, when the CPU load is detected to exceed 85%, the intelligent computing idle resource expansion process is automatically triggered. Log tracing logic: Records full-process operation logs, stored in JSON format. Fields include resource call time (accurate to milliseconds), computing power type, business ID, scheduling instruction content, abnormal event description, recovery operation steps, and operator (if manual intervention is required). Logs are stored in a distributed log system, supporting multi-dimensional retrieval by business ID, time range, operation type, and exception level. The retrieval response time is less than 1 second, and the log retention period is 365 days by default, meeting audit requirements.

[0024] Computing power access control logic: An RBAC (Role-Based Access Control) permission model is adopted, defining three core roles: "Intelligent Computing Administrator" (can call and configure intelligent computing resources, and view all logs), "General Computing Operator" (can only call general computing resources and view related business logs), and "Auditor" (can only view logs, without resource operation permissions). Users must pass two-factor authentication using "account password + dynamic token." After successful authentication, permissions are assigned according to the role. When a user initiates a computing power call request, permissions are verified first; if no permissions are granted, a 403 error is returned, and the access log is recorded. Data encryption logic: For business scheduling logs and user permission information, AES-256 encryption algorithm is used for storage encryption. The key is generated based on the hardware encryption module (HSM). A unique encryption key is generated for each data block (default 1MB), and the key is managed in a hierarchical manner of "master key + data key". The master key is stored in the HSM, and the data key is encrypted and stored in the database. At the same time, log transmission encryption is supported (using TLS1.3 protocol) to prevent data leakage during transmission. The key is rotated regularly, with a default rotation period of 90 days. Data access is not affected during the rotation.

[0025] Example 2: A cloud service full lifecycle intelligent operation and maintenance system based on dynamic resource orchestration, comprising the following modules: Data Acquisition and Profile Building Module: Lightweight acquisition logic is embedded in the intelligent computing GPU cluster and general computing CPU cluster to collect intelligent computing indicators and general computing indicators in real time. Based on the collected data, a computing power resource profile is built. This profile includes real-time resource load, historical utilization trend, and task type proportion information. The data update frequency is consistent with the indicator acquisition frequency, which is second-level update, and supports multi-dimensional display by cluster dimension, business dimension, and resource type dimension. Network demand matching module: It has a preset "business type-network demand" mapping rule library. The mapping rule library supports dynamic updates. When a new business type is added, temporary rules can be added. After verification, the rules are included in the formal rule library. After the rules are updated, they are automatically synchronized to the "business-network" matching logic without restarting the service. Intelligent computing tasks are associated with high bandwidth policies and general computing tasks are associated with low latency policies. When a business initiates a computing power request, the network demand tag is automatically matched through the business type identifier. Dynamic resource orchestration module: Executes dynamic resource orchestration logic, monitors the utilization rate of the intelligent computing cluster, and adapts idle GPU resources to intelligent computing available resources through Kubernetes containerization technology when the utilization rate is less than 30%. When the QPS of the general computing cluster exceeds the threshold, it calls upon idle intelligent computing power and also has a general computing task resource release mechanism. It converts network demand tags into SDN configuration instructions and issues them, verifies the effectiveness status through the SDN controller status feedback interface, and reissues them if they do not take effect within 5 seconds. After 3 failed retries, it records an exception log containing the instruction content, failure reason, and timestamp. It ensures the exclusive use of computing power for intelligent computing tasks. The intelligent operation and maintenance support module automatically migrates faulty intelligent computing tasks and expands general computing resources; outputs predicted metric values ​​for the next 10 minutes; sets anomaly warning thresholds, including GPU memory usage exceeding 90%, CPU load exceeding 85%, and QPS exceeding the threshold by 1.5 times, with warning information synchronized to operation and maintenance personnel and including suggested handling solutions; uses AES-256 algorithm to encrypt and store scheduling logs and user information, with AES-256 encryption using a "master key + data key" hierarchical management, and a default key rotation cycle of 90 days, during which data access is not affected; scheduling logs are stored in JSON format, with fields including resource call time, computing power type, business ID, scheduling instruction, anomaly description, and recovery steps, supporting multi-dimensional retrieval, and a default retention time of 365 days; and controls computing power access permissions through the RBAC permission model.

[0026] It also includes a standardized computing power invocation module, which builds a unified computing power interface in RESTful style. The intelligent computing interface is compatible with TensorFlow 2.x / PyTorch 1.8+ framework, and the universal computing interface is adapted to general business systems. The interface has built-in protocol conversion logic, which allows businesses to invoke heterogeneous computing power without modifying the code.

[0027] The standardized computing power call module also includes an interface testing tool, which supports business verification of compatibility and provides detailed suggestions on output parameter formats, permissions, and protocol adaptation dimensions.

[0028] The data acquisition and profiling module adopts a distributed acquisition architecture to improve the efficiency and stability of data acquisition, ensuring that various indicator data can be obtained in real time and accurately in a large-scale cluster environment. At the same time, it has an automatic identification and repair mechanism for abnormal data, ensuring the accuracy and reliability of computing resource profiling.

[0029] When handling abnormal situations, the intelligent operation and maintenance support module not only provides early warnings and handles them according to preset rules, but also has intelligent decision-making capabilities. It can automatically generate the optimal handling strategy based on historical data and the current business scenario. At the same time, it conducts in-depth analysis of frequently occurring abnormal situations, digs out the root causes of potential problems, and provides targeted optimization suggestions to continuously improve the stability and performance of cloud services.

[0030] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A cloud service full lifecycle intelligent operation and maintenance method based on dynamic resource orchestration, characterized in that: Includes the following steps: Resource profiling: Lightweight data collection logic is embedded in the intelligent computing GPU cluster and general computing CPU cluster to collect intelligent computing indicators and general computing indicators in real time. Based on the data, a computing power resource profile including "real-time resource load, historical utilization trend, and task type proportion" is constructed. The data update frequency is consistent with the indicator collection frequency, and it supports multi-dimensional display by "cluster dimension, business dimension, and resource type dimension". Network demand matching: A pre-defined "business type - network demand" mapping rule base is available, which supports dynamic updates. Temporary rules can be added for new business types, and they will be included in the official rule base after verification. After the rules are updated, they are automatically synchronized to the "business - network" matching logic without restarting the service. Intelligent computing tasks are associated with high-bandwidth policies, and general computing tasks are associated with low-latency policies. When a business initiates a computing power request, the network demand tag is automatically matched through the "business type identifier". Dynamic resource orchestration: Monitors the utilization rate of the intelligent computing cluster; when it is less than 30%, it adapts idle GPU resources to intelligent computing resources using Kubernetes containerization technology; when the QPS of the general computing cluster exceeds the threshold, it calls upon idle computing power of the intelligent computing cluster; it also has a general computing task resource release mechanism; it converts network demand tags into SDN configuration instructions and issues them, verifies the effectiveness status through the SDN controller status feedback interface, and reissues them if they do not take effect within 5 seconds, and records the exception log after 3 failed retries; it ensures the exclusive use of computing power for intelligent computing tasks. Intelligent O&M Support: Automatically migrates faulty intelligent computing tasks and expands general computing resources; outputs predicted metric values ​​for the next 10 minutes; anomaly warning thresholds include GPU memory usage exceeding 90%, CPU load exceeding 85%, and QPS exceeding the threshold by 1.5 times, with warning information synchronized to O&M personnel and including suggested handling solutions; uses AES-256 algorithm to encrypt and store scheduling logs and user information, employing a "master key + data key" hierarchical management system, with a default key rotation cycle of 90 days, and the rotation process does not affect data access; scheduling logs are stored in JSON format, with fields including resource call time, computing power type, business ID, scheduling instruction, anomaly description, and recovery steps, supporting multi-dimensional retrieval, and a default retention time of 365 days; and controls computing power access permissions through the RBAC permission model.

2. The intelligent operation and maintenance method for the entire lifecycle of cloud services based on dynamic resource orchestration as described in claim 1, characterized in that: It also includes standardized computing power invocation steps: building a unified computing power interface in RESTful style, with the intelligent computing interface compatible with TensorFlow 2.x / PyTorch 1.8+ framework, and the universal computing interface adapted to general business systems; the interface has built-in protocol conversion logic, supporting "businesses can invoke heterogeneous computing power without modifying code".

3. The intelligent operation and maintenance method for the entire lifecycle of cloud services based on dynamic resource orchestration as described in claim 2, characterized in that: The standardized computing power call process also includes interface testing tools that support business verification of compatibility and provide detailed suggestions on output parameter formats, permissions, and protocol adaptation dimensions.

4. The intelligent operation and maintenance method for the entire lifecycle of cloud services based on dynamic resource orchestration as described in claim 3, characterized in that: In the dynamic resource orchestration process, when dynamically allocating computing resources, priority is given to ensuring the resource needs of critical services. Based on service priority and resource occupancy, the resource allocation strategy is intelligently adjusted to ensure the high availability and high performance of critical services.

5. The intelligent operation and maintenance method for the entire lifecycle of cloud services based on dynamic resource orchestration according to claim 4, characterized in that: In the intelligent operation and maintenance support process, in addition to synchronizing abnormal early warning information with operation and maintenance personnel and including suggested handling solutions, the corresponding emergency handling process is automatically triggered according to the type and severity of the abnormality, automatically switching to backup resources and starting degraded services to minimize the impact on business.

6. A cloud service full lifecycle intelligent operation and maintenance system based on dynamic resource orchestration, applied to the method described in claim 5, characterized in that: Includes the following modules: Data Acquisition and Profile Building Module: Lightweight acquisition logic is embedded in the intelligent computing GPU cluster and general computing CPU cluster to collect intelligent computing indicators and general computing indicators in real time. Based on the collected data, a computing power resource profile is built. This profile includes real-time resource load, historical utilization trend, and task type proportion information. The data update frequency is consistent with the indicator acquisition frequency, which is second-level update, and supports multi-dimensional display by cluster dimension, business dimension, and resource type dimension. Network demand matching module: It has a preset "business type-network demand" mapping rule library. The mapping rule library supports dynamic updates. When a new business type is added, temporary rules can be added. After verification, the rules are included in the official rule library. After the rules are updated, they are automatically synchronized to the "business-network" matching logic without restarting the service. Intelligent computing tasks are associated with high bandwidth policies and general computing tasks are associated with low latency policies. When a business initiates a computing power request, the network demand tag is automatically matched through the business type identifier. Dynamic resource orchestration module: Executes dynamic resource orchestration logic, monitors the utilization rate of the intelligent computing cluster, and adapts idle GPU resources to intelligent computing available resources through Kubernetes containerization technology when the utilization rate is less than 30%. When the QPS of the general computing cluster exceeds the threshold, it calls upon idle intelligent computing power and also has a general computing task resource release mechanism. It converts network demand tags into SDN configuration instructions and issues them, verifies the effectiveness status through the SDN controller status feedback interface, and reissues them if they do not take effect within 5 seconds. After 3 failed retries, it records an exception log containing the instruction content, failure reason, and timestamp. It ensures the exclusive use of computing power for intelligent computing tasks. The intelligent operation and maintenance support module automatically migrates faulty intelligent computing tasks and expands general computing resources; outputs predicted metric values ​​for the next 10 minutes; sets anomaly warning thresholds, including GPU memory usage exceeding 90%, CPU load exceeding 85%, and QPS exceeding the threshold by 1.5 times, with warning information synchronized to operation and maintenance personnel and including suggested handling solutions; uses AES-256 algorithm to encrypt and store scheduling logs and user information, with AES-256 encryption using a "master key + data key" hierarchical management, and a default key rotation cycle of 90 days, during which data access is not affected; scheduling logs are stored in JSON format, with fields including resource call time, computing power type, business ID, scheduling instruction, anomaly description, and recovery steps, supporting multi-dimensional retrieval, and a default retention time of 365 days; and controls computing power access permissions through the RBAC permission model.

7. A cloud service full lifecycle intelligent operation and maintenance system based on dynamic resource orchestration as described in claim 6, characterized in that: It also includes a standardized computing power invocation module, which builds a unified computing power interface in RESTful style. The intelligent computing interface is compatible with TensorFlow 2.x / PyTorch 1.8+ framework, and the universal computing interface is adapted to general business systems. The interface has built-in protocol conversion logic, which allows businesses to invoke heterogeneous computing power without modifying the code.

8. The cloud service full lifecycle intelligent operation and maintenance system based on dynamic resource orchestration according to claim 7, characterized in that: The standardized computing power call module also includes an interface testing tool, which supports business verification of compatibility and provides detailed suggestions on output parameter formats, permissions, and protocol adaptation dimensions.

9. A cloud service full lifecycle intelligent operation and maintenance system based on dynamic resource orchestration as described in claim 8, characterized in that: The data acquisition and profiling module adopts a distributed acquisition architecture to improve the efficiency and stability of data acquisition, ensuring that various indicator data can be obtained in real time and accurately in a large-scale cluster environment. At the same time, it has an automatic identification and repair mechanism for abnormal data, ensuring the accuracy and reliability of computing resource profiling.

10. A cloud service full lifecycle intelligent operation and maintenance system based on dynamic resource orchestration according to claim 9, characterized in that: When dealing with abnormal situations, the intelligent operation and maintenance support module, in addition to issuing warnings and handling them according to preset rules, also has intelligent decision-making capabilities and can automatically generate the optimal handling strategy based on historical data and the current business scenario. Meanwhile, for frequently occurring anomalies, we conduct in-depth analysis to uncover the root causes of potential problems and provide targeted optimization suggestions to continuously improve the stability and performance of cloud services.

Images