Cloud platform system based on intelligent decision chain and resource availability method

Abstract

The invention provides a cloud platform system based on an intelligent decision chain and a resource availability method. The cloud platform system comprises a monitoring module, a feature library, a decision table and a decision tree, the monitoring module is used for collecting resource data, performing path matching detection in a decision according to a decision table and triggering a decision behavior of a decision chain when monitoring data is abnormal; the feature library is used for recording resource fault feature information; the decision table is used for recording a weight value of each feature point in the feature library, performing priority ranking according to the weight values, and constructing a decision chain in the decision tree; the decision-making tree is used for recording all decision-making links and decision-making selections corresponding to the decision-making links, each leaf node of the decision-making tree corresponds to one decision-making behavior, and when the decision-making path walks to the leaf node of the decision-making tree, the decision-making behavior corresponding to the leaf node is executed. The overall high availability of the cloud platform system is adjusted in time through state monitoring and feedback of each unit and the load condition.

Description

Technical Field

[0001] The present invention relates to the technical field of cloud platform systems, and in particular to a cloud platform system based on an intelligent decision chain and a resource availability method. Background Art

[0002] As cloud computing matures, the migration of core businesses to the cloud places very high demands on the reliability and availability of cloud virtual machine resources to ensure the continuity of core business system operations. However, the different types and forms of business cannot effectively detect the health status of business systems through unified standard monitoring. The critical state boundaries of physical machines and virtual machine operating systems are fuzzy, and there is a risk of false triggering of cloud virtual machine migration, which may bring catastrophic risks to the business.

[0003] In cloud scenarios, the industry usually uses the following methods to control resource high availability:

[0004] Method 1: Sense computing nodes through heartbeat

[0005] The cloud platform senses the health status of computing nodes through the heartbeat mechanism. If the heartbeat packet reception times out or an abnormal heartbeat packet is received from the node, the node is judged to be offline and the downtime migration task is started to pull up the cloud virtual machine from other nodes.

[0006] Disadvantages of method 1:

[0007] There are many reasons why cloud virtual machines are offline. A single heartbeat judgment through the cloud platform management network cannot fully cover the critical state judgment of cloud virtual machines. If there is a problem with the node management network, but the storage network is normal, the cloud virtual machine can still read and write normally. If the cloud platform determines that the node is offline at this time, and pulls up the cloud virtual machine again on other nodes, the storage volume will be read and written on multiple ends. If the operating system is too busy and temporarily unresponsive, or some system resources cannot be allocated, the network / disk io is stuck, and the cloud virtual machine is pulled up on other nodes at this time, the faulty node suddenly recovers, then both sides may read and write a storage device at the same time, causing device damage.

[0008] Method 2: Manual Migration

[0009] When a computing node goes offline or encounters other failures, manually check the status of the physical machine and decide to restart the node server or create a cloud virtual machine of the same specification on another computing node, manually migrate and restart the business.

[0010] Disadvantages of method 2:

[0011] Handling fault scenarios often relies on the level of on-site operation and maintenance personnel. Inexperienced operation and maintenance personnel may misjudge the current fault form and make wrong decisions when handling fault problems, bringing the risk of data corruption and expanding the impact. The disadvantage of this method is that it has low timeliness. If a network anomaly occurs without supervision, it is impossible to automatically achieve high availability of cloud virtual machines and ensure sustainable operation of the business. Summary of the invention

[0012] In view of the shortcomings of the prior art, the purpose of the present invention is to provide a cloud platform system and a usable method based on an intelligent decision chain, which aims to timely adjust the overall availability of the cloud platform system through status monitoring and feedback as well as load conditions, covering the vehicle-mounted cloud platform system and vehicle-mounted applications.

[0013] Another object of the present invention is to provide a usable method for the above-mentioned cloud platform system based on intelligent decision chain.

[0014] In order to achieve the above object, the present invention is implemented through the following technical solutions.

[0015] To this end, the present invention provides a cloud platform system based on an intelligent decision chain, including a monitoring module, a feature library, a decision table and a decision tree;

[0016] The monitoring module is used to collect resource data and match fault features according to feature points of the feature library, perform path matching detection in decision making according to the decision table, and trigger decision chain decision behavior when abnormalities occur in monitoring data;

[0017] The feature library is used to record resource fault feature information;

[0018] The decision table is used to record the weight value of each feature point in the feature library, and to construct a decision chain in the decision tree by prioritizing the weight value;

[0019] The decision tree is used to record all decision links and the decision choices of the corresponding decision links. Each leaf node of the decision tree corresponds to a decision behavior. When the decision path reaches a leaf node of the decision tree, the decision behavior corresponding to the leaf node will be executed.

[0020] According to the cloud platform system based on intelligent decision chain, the monitoring module includes a collector, a memory and a trigger, the collector is used to monitor the cloud platform service status, operating system and hardware status, cloud resource monitoring, and form monitoring data, and the memory is used to store the monitoring data;

[0021] The cloud platform service status monitoring includes service operation status, service log, service health status, etc.; the operating system includes system kernel (hardware driver, interrupt response), system operation status, system load (CPU, memory, disk), and global handle number;

[0022] The hardware status monitoring includes monitoring of physical CPU, memory, network card, PCI device, RAID card, disk, and motherboard;

[0023] The cloud resource monitoring includes monitoring of vcpu load information, virtual memory exception information, and disk IO information;

[0024] The trigger is used to monitor the monitoring data and trigger a decision-making behavior when the monitoring data is abnormal.

[0025] According to the cloud platform system based on the intelligent decision chain, the feature library is used to record feature information of cloud platform resources under a fault state, and the feature information includes computing, storage, and network information of the virtualization layer, computing, storage, and network information of the operating system layer, and computing, storage, network, and peripheral information of the physical hardware layer.

[0026] According to the cloud platform system based on intelligent decision chain, the working process of the decision tree includes three stages: construction of the decision tree, decision process, and decision feedback. The construction of the decision tree is carried out through the monitoring module and the feature library. When the monitoring module detects an abnormal indicator, the decision tree is reported, and the decision tree matches the fault feature information of the feature library according to the information uploaded by the monitoring module.

[0027] Find the next fault feature point based on the feature points and weight data of the matched feature library, and read the indicator information of the corresponding feature point of the monitoring module;

[0028] According to the returned indicator information, the feature library is screened again until the feature points that meet all the feature indicators are screened out, and finally the decision behavior for the feature point is executed.

[0029] According to the cloud platform system based on the intelligent decision chain, the decision-making behaviors executed for the feature point include: single virtual machine migration, whole machine migration, hot migration, cold migration, restarting the virtual machine operating system, abnormal virtual peripheral replacement (USB), creating a new virtual machine from a backup, and sending an alarm message, one or more combinations thereof.

[0030] According to the cloud platform system based on intelligent decision chain, the decision feedback stage includes:

[0031] During the decision-making process, the decision-making process will be continuously recorded, and statistics will be kept on each decision-making process, decision-making link, and decision-making plan;

[0032] In addition, statistical indicators are formed by counting the occurrence events of each fault indicator, the number of occurrences, and the relationship between the fault time, and the statistical indicators are used to improve the feature library and the decision tree.

[0033] The present invention also provides a resource availability method applied to the cloud platform system based on the intelligent decision chain, comprising the following steps:

[0034] S1: Establish multi-dimensional monitoring items. The collector of the monitoring module is used to collect monitoring resources of the cloud platform;

[0035] S2: Initialize the feature library by importing the feature library data file or manually entering the feature data by the user;

[0036] S3: The trigger of the monitoring module continuously monitors the system indicators and compares the characteristic points of the feature library. If the monitoring indicator meets at least one indicator of the characteristic points in the feature library, it enters the decision tree management;

[0037] S4: Decision chain identification finds the decision that meets the feature point and finds the next feature point information of the decision;

[0038] S5: Generate the next decision tree node according to the next feature point information, and continue to obtain whether the resource monitoring data meets the feature point characteristics according to the feature point of the decision tree node; if it meets the feature point characteristics, continue to generate the next decision node; if it does not meet the feature point characteristics, go back and try to find the feature point of the next feature until the feature data that meets all the feature points is found in the feature library, and finally execute the feature decision behavior to complete the construction of the decision tree.

[0039] According to the resource availability method of the cloud platform system based on the intelligent decision chain, in step S1, collecting monitoring resources of the cloud platform includes collecting cloud platform service status monitoring information, operating system monitoring information, hardware status monitoring information, and cloud resource monitoring information;

[0040] The cloud platform service monitoring information includes indicators such as service operation status, service logs, and service health status;

[0041] The operating system monitoring information includes system kernel (hardware driver, interrupt response), system operation status, system load (CPU, memory, disk), and global handle number;

[0042] The hardware status monitoring information includes physical CPU, memory, network card, PCI device, RAID card, disk, motherboard, etc. The cloud resource monitoring information includes vcpu load information, virtual memory exception information, and disk IO information;

[0043] After the data collection of the monitoring resources is completed, the data is classified and sorted and then collected into the memory for data storage. The storage data type includes the indicator name, indicator type, collection time, data unit, and indicator value data item.

[0044] According to the resource availability method of the cloud platform system based on the intelligent decision chain, the feature data manually entered by the user includes: the entry information includes the feature name, feature classification, feature detection index, and the execution strategy after the feature is met; the feature name is used to identify the unique feature; the feature classification is used to classify the resources to which the feature belongs, including a variety of different cloud platform resource types such as virtual machine resources, computing node resources, SDN services, cloud services, and storage services.

[0045] According to the resource availability method of the cloud platform system based on the intelligent decision chain, the execution strategy corresponds to the type of cloud resources:

[0046] For virtual machine type resources, the execution strategy includes configurable virtual machine migration, restarting the virtual machine operating system, abnormal virtual peripheral replacement (USB), creating a new virtual machine from a backup, and sending alarm information operations;

[0047] For cloud service resources, the execution strategy includes restarting the service to attempt to recover the service;

[0048] For node resources, the execution strategies include virtual machine downtime migration, network service restart, and system parameter tuning operations;

[0049] For SDN, the execution strategy includes cache cleaning, parameter tuning, and service restart operations;

[0050] For storage resources, the execution measurement includes restarting the ceph monitor service and reconnecting the iscsi service.

[0051] According to the resource availability method of the cloud platform system based on the intelligent decision chain, after completing the construction of the decision tree, it also includes:

[0052] The decision chain data is saved. When the same scenario occurs next time, the feature determination and decision execution are completed according to the decision chain saved last time.

[0053] The decision tree data is a tree structure data, including feature points constituting each tree node, each node pointing to a child feature node or a parent feature node; the leaves of the decision tree are decision nodes, representing execution actions that satisfy all feature points of the linked list;

[0054] The data structure of the tree node includes: the previous decision node, the next feature node, the node type and feature value identifier, and the feature point table identifier.

[0055] Compared with the prior art, the cloud platform system and resource availability method based on the intelligent decision chain have the following beneficial effects:

[0056] 1) Resource detectors can collect various indicators of different resources on the cloud platform (computing servers, virtual machines, applications) in multiple dimensions, and provide a large amount of multi-dimensional operation data of resources for judgment;

[0057] 2) Establish a feature library through faults, covering cloud resource fault status monitoring in various complex scenarios, and continuously update the feature library by adjusting feature parameters, so that it can get closer and closer to the fault feature information in specific scenarios and make correct actions to recover from the fault;

[0058] 3) The platform feature library data can be imported and exported; mature feature library data can be copied to other cloud platforms. Every update and improvement of the feature library can improve the robustness of each cloud platform using the feature library. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 It is a structural diagram of the cloud platform system based on the intelligent decision chain.

[0060] Figure 2 It is a flow chart of a resource high availability method for a cloud platform system based on an intelligent decision chain.

[0061] Figure 3 It is a flow chart for collecting and saving monitoring data.

[0062] Figure 4 It is a fault-triggered decision flow chart.

[0063] Figure 5 It is a failure scenario decision diagram. DETAILED DESCRIPTION

[0064] The present invention is further described in detail below through specific preferred embodiments in conjunction with the accompanying drawings, but the present invention is not limited to the following embodiments.

[0065] The present invention discloses a cloud platform system and a resource availability method based on an intelligent decision chain. The cloud platform system includes: a monitoring module, a feature library, a decision table, and a decision tree; the monitoring module is used to collect resource data, perform path matching detection in the decision according to the decision table, and trigger the decision chain decision behavior when the monitoring data is abnormal; the feature library is used to record resource fault feature information; the decision table is used to record the weight value of each feature point in the feature library, and construct a decision chain in the decision tree according to the priority sorting of the weight value; the decision tree is used to record all decision links and the decision selection of the corresponding decision links, and each leaf node of the decision tree corresponds to a decision behavior. When the decision path reaches the leaf node of the decision tree, the decision behavior corresponding to the leaf node will be executed. The overall high availability of the cloud platform system is adjusted in time through the status monitoring and feedback of each unit and the load situation.

[0066] Example 1

[0067] Figure 1 This is a schematic diagram of the structure of the cloud platform system based on the intelligent decision chain. Figure 1 As shown in the figure, the cloud platform system based on the intelligent decision chain includes four major components: monitoring module, feature library, decision table, and decision tree. The overall high availability of the cloud platform system is adjusted in time through the status monitoring and feedback of each unit and the load conditions, covering the vehicle-mounted cloud platform system and vehicle-mounted applications.

[0068] The monitoring module is used to collect and archive resource data, perform path matching detection in decision making according to the decision table, and trigger decision chain decision behavior when abnormal monitoring data occurs. Specifically, the monitoring module includes a collector, a memory, and a trigger. The collector is used to monitor the cloud platform service status, operating system and hardware status, and cloud resource monitoring to form monitoring data. The memory is used to store the monitoring data. The trigger is used to monitor the monitoring data and trigger the decision behavior when the monitoring data is abnormal.

[0069] Among them, the cloud platform service status monitoring indicators include service operation status, service log, service health status, etc.; the operating system includes the system kernel (hardware driver, interrupt response), system operation status, system load (CPU, memory, disk), global handle number, etc.; hardware status monitoring includes physical CPU, memory, network card, PCI device, RAID card, disk, motherboard, etc.; cloud resource monitoring information includes vcpu load information, virtual memory abnormality information, disk IO information, etc. In this way, through multi-level resource monitoring, the cloud platform system builds a multi-dimensional system monitoring indicator.

[0070] The feature library is used to record resource failure feature information, which is relied upon by monitoring triggers and decision chains. The feature library is used to record the feature information of various resources on the cloud platform under fault conditions. The feature information includes multi-dimensional information such as computing, storage, and network information at the virtualization layer, computing, storage, and network information at the operating system layer, and computing, storage, network, and peripheral information at the physical hardware layer.

[0071] In the early stage of use, developers or operation and maintenance personnel enter the platform feature library based on their own experience; the feature library data is a process of continuous improvement. If a system feature is found to be incomplete or inaccurate during daily operation and maintenance, the feature library can be improved by continuously adjusting parameter indicators; the feature library data can also be exported, and the exported data can be synchronized to other platforms, so that the mature experience accumulated on one platform can be copied to other platforms, and the experience of other platforms can be entered into the policy library to achieve continuous iteration.

[0072] The decision table is used to record the weight value of each feature point in the feature library, and to build a decision chain in the decision tree by prioritizing them according to the weight value.

[0073] The decision tree is used to record all decision links and the decision choices of the corresponding links. Each leaf node of the decision tree corresponds to a decision behavior. When the decision path reaches the leaf node of the decision tree, the decision behavior of the leaf node will be executed.

[0074] Specifically, the working process of the decision tree includes three stages: decision tree construction, decision process, and decision feedback. The construction of the decision tree depends on the monitoring module and the feature library, and is constructed through the monitoring module and the feature library. When the monitoring module detects an abnormal indicator, it reports to the decision tree. The decision tree matches the fault feature information of the feature library according to the information uploaded by the monitoring module, searches for the next fault feature point according to the matched feature library information and weight data, and reads the corresponding feature point indicator information of the monitoring module. According to the returned indicator information, it is screened again in the feature library until the feature point that meets all the feature indicators is screened out, and finally the decision behavior of the feature point is executed.

[0075] The execution strategy (i.e., the decision-making behavior for the feature point) includes: single virtual machine migration, whole machine migration, hot migration, cold migration, restarting the virtual machine operating system, abnormal virtual peripheral replacement (USB), creating a new virtual machine from a backup, sending an alarm message, or a combination of these to ensure the high availability of the virtual machine. This process is automatically controlled by the system without manual intervention.

[0076] The entire decision-making process can be traced back. In the decision feedback stage, the decision-making process will be continuously recorded during the decision-making process, and statistics will be kept on each decision-making process, decision link and decision plan; as well as statistics on the occurrence of each fault indicator, the number of occurrences and the relationship with the fault time. The feature library and decision tree will be continuously improved through statistical indicators.

[0077] Figure 2 This is a flow chart of a resource high availability method for a cloud platform system based on an intelligent decision chain. Figure 2 As shown, the resource availability method applied to the above-mentioned cloud platform system based on the intelligent decision chain includes the following steps:

[0078] S1: Establish multi-dimensional monitoring items. The collector of the monitoring module is responsible for collecting cloud platform monitoring resources, including multi-dimensional collection information, including cloud platform service status monitoring information, operating system monitoring information, hardware status monitoring information, and cloud resource monitoring information. Among them, cloud platform service monitoring information includes indicators such as service operation status, service log, and service health status; the operating system includes system kernel (hardware driver, interrupt response), system operation status, system load (CPU, memory, disk), global handle number, etc.

[0079] Hardware status monitoring includes physical CPU, memory, network card, PCI device, RAID card, disk, motherboard, etc. Cloud resource monitoring information includes vcpu load information, virtual memory exception information, disk IO information, etc. After data collection is completed, the data will be classified and sorted and then collected into the memory for data storage. The storage data type includes indicator name, indicator type, collection time, data unit, indicator value and other data items.

[0080] S2: The feature library can be initialized by importing the feature library data file or manually entering the feature data by the user. The feature library data file can be exported from other feature libraries that have been entered. Figure 3 shown.

[0081] If manual data entry is performed, the information that needs to be entered includes the feature name, feature classification, feature detection index, and execution strategy after the feature is met; the feature name is used to identify the unique feature; the feature classification is used to classify the resources to which the feature belongs, including a variety of different cloud platform resource types such as virtual machine resources, computing node resources, SDN services, cloud services, and storage services; different types of cloud resources have different executable strategies: for example, for virtual machine type resources, you can configure virtual machine migration (hot migration, cold migration), restart the virtual machine operating system, replace abnormal virtual peripherals (USB), create a new virtual machine from a backup, and send alarm information operations; for cloud service resources, you can restart the service to try to recover the service; for node resources, you can configure virtual machine downtime migration, network service restart, system parameter tuning and other operations; for SDN services, you can perform cache cleanup, parameter tuning, service restart and other operations; for storage resources, executable operations include ceph monitor service restart, iscsi service reconnection and other operations.

[0082] S3: The trigger in the monitoring module continuously monitors the system indicators, monitors and continuously compares the feature information of the feature points in the feature library. If the monitoring indicator meets a certain indicator in the feature library, it enters the decision tree management, as shown below: Figure 4 shown.

[0083] S4: Decision chain identification finds the decision that meets the feature point and finds the next feature point information of the decision.

[0084] S5: Generate the next decision tree node according to the next feature point information, and continue to obtain whether the resource monitoring data meets the feature point characteristics according to the feature point of the decision tree node: if it meets, continue to generate the next decision node; if not, backtrack and find the feature point of the next feature to try, until the feature data that meets all the feature points is found in the feature library, and finally execute the feature decision behavior to complete the construction of the decision tree.

[0085] Each time a decision tree is built, the decision chain data will be saved. Next time the same scenario occurs, the feature determination and decision execution can be quickly completed based on the decision chain saved last time.

[0086] The decision tree data structure is a tree structure. Each tree node is composed of feature points. Each node can point to a child feature node or a parent feature node. The leaves of the tree are decision nodes, which represent the execution actions that satisfy all feature points in the linked list. The data structure of the tree node includes the previous decision node, the next feature node, the node type and feature value identifier, and the feature point table identifier. The pseudo code can be expressed as follows:

[0087] {arentId / / Previous decision node

[0088] childId / / next feature node

[0089] type / / node type feature node or decision node

[0090] charaId / / feature id

[0091] charaPointId / / feature point id}.

[0092] Example 2

[0093] Figure 5 A feature detection decision diagram for a fault scenario provided in the embodiment of the present application is as follows: Figure 5 As shown in the figure, the platform computing resources fail, the trigger triggers the fault decision, and generates a decision tree, starting with feature point A as the root node to make decisions; in the query decision table, A has two reachable paths, namely feature point A->feature point B, with a weight value of 10; feature point A->feature point C, with a weight value of 2. According to the priority principle of large weight value, the feature point A->feature point B path is selected, and whether there is abnormal data of feature point B is detected. If the feature point B data is satisfied, the query decision table shows that feature point B has two reachable paths, namely feature point B->feature point and feature point B->feature point E, with weights of 3 and 6 respectively. Therefore, the feature point B->feature point E path is selected, and feature point E is detected. If feature point E satisfies, since feature point E only has a reachable path from feature point E->feature point F, it is detected whether feature point F is satisfied. If it is satisfied, the corresponding fault handling decision is executed.

[0094] The above-mentioned cloud platform system based on intelligent decision chain and resource availability method have the following beneficial effects:

[0095] 1) Resource detectors can collect various indicators of different resources on the cloud platform (computing servers, virtual machines, applications) in multiple dimensions, and provide a large amount of multi-dimensional operation data of resources for judgment;

[0096] 2) Establish a feature library through faults, covering cloud resource fault status monitoring in various complex scenarios, and continuously update the feature library by adjusting feature parameters, so that it can get closer and closer to the fault feature information in specific scenarios and make correct actions to recover from the fault;

[0097] 3) The platform feature library data can be imported and exported; mature feature library data can be copied to other cloud platforms. Every update and improvement of the feature library can improve the robustness of each cloud platform using the feature library.

[0098] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Therefore, any simple modification, equivalent change and modification made to the above embodiment according to the technical essence of the present invention without departing from the technical solution of the present invention still fall within the scope of the technical solution of the present invention.

Claims

1. A cloud platform system based on intelligent decision chain, characterized in that: Includes monitoring module, feature library, decision table and decision tree; The monitoring module is used to collect resource data and match fault features according to feature points of the feature library, perform path matching detection in decision making according to the decision table, and trigger decision chain decision behavior when abnormalities occur in monitoring data; The feature library is used to record resource fault feature information; The decision table is used to record the weight value of each feature point in the feature library, and to construct a decision chain in the decision tree by prioritizing the weight value; The decision tree is used to record all decision links and the decision choices of the corresponding decision links. Each leaf node of the decision tree corresponds to a decision behavior. When the decision path reaches a leaf node of the decision tree, the decision behavior corresponding to the leaf node will be executed. By collecting various resource indicators of the cloud platform in multiple dimensions through resource detectors and combining them with the establishment and update of a fault feature library, the cloud platform's fault monitoring and recovery capabilities can be continuously improved.

2. The cloud platform system based on intelligent decision chain according to claim 1, characterized in that: The monitoring module includes a collector, a memory and a trigger. The collector is used to monitor the cloud platform service status, the operating system and hardware status, and the cloud resource monitoring to form monitoring data. The memory is used to store the monitoring data. The cloud platform service status monitoring includes service operation status, service log, service health status, etc.; the operating system includes system kernel (hardware driver, interrupt response), system operation status, system load (CPU, memory, disk), and global handle number; The hardware status monitoring includes monitoring of physical CPU, memory, network card, PCI device, RAID card, disk, and motherboard; The cloud resource monitoring includes monitoring of vcpu load information, virtual memory exception information, and disk IO information; The trigger is used to monitor the monitoring data and trigger a decision-making behavior when the monitoring data is abnormal.

3. The cloud platform system based on intelligent decision chain according to claim 1, characterized in that: The feature library is used to record feature information of cloud platform resources under fault conditions, and the feature information includes computing, storage, and network information of the virtualization layer, computing, storage, and network information of the operating system layer, and computing, storage, network, and peripheral information of the physical hardware layer.

4. The cloud platform system based on intelligent decision chain according to claim 1, characterized in that: The working process of the decision tree includes three stages: construction of the decision tree, decision process, and decision feedback. The construction of the decision tree is carried out through the monitoring module and the feature library. When the monitoring module detects an abnormal indicator, it reports the decision tree, and the decision tree matches the fault feature information of the feature library according to the information uploaded by the monitoring module. Find the next fault feature point based on the feature points and weight data of the matched feature library, and read the indicator information of the corresponding feature point of the monitoring module; According to the returned indicator information, the feature library is screened again until the feature points that meet all the feature indicators are screened out, and finally the decision behavior for the feature point is executed.

5. The cloud platform system based on intelligent decision chain as described in claim 4 is characterized by: The decision-making actions executed for the feature point include: single virtual machine migration, whole machine migration, hot migration, cold migration, restarting the virtual machine operating system, abnormal virtual peripheral replacement (USB), creating a new virtual machine from a backup, sending an alarm message, or a combination thereof; The decision feedback stage includes: During the decision-making process, the decision-making process will be continuously recorded, and statistics will be kept on each decision-making process, decision-making link, and decision-making plan; In addition, statistical indicators are formed by counting the occurrence events of each fault indicator, the number of occurrences, and the relationship between the fault time, and the statistical indicators are used to improve the feature library and the decision tree.

6. A resource availability method for a cloud platform system based on an intelligent decision chain applied to any one of claims 1 to 5, characterized in that: The following steps are involved: S1: Establish multi-dimensional monitoring items. The collector of the monitoring module is used to collect monitoring resources of the cloud platform; S2: Initialize the feature library by importing the feature library data file or manually entering the feature data by the user; S3: The trigger of the monitoring module continuously monitors the system indicators and compares the characteristic points of the feature library. If the monitoring indicator meets at least one indicator of the characteristic points in the feature library, it enters the decision tree management; S4: Decision chain identification finds the decision that meets the feature point and finds the next feature point information of the decision; S5: Generate a next decision tree node according to the next feature point information, and continue to obtain whether the resource monitoring data meets the feature point characteristics according to the feature point of the decision tree node: if it meets the feature point characteristics, continue to generate the next decision node; If it is not satisfied, it will backtrack and find the feature point of the next feature to try until the feature data that satisfies all the feature points is found in the feature library. Finally, the feature decision behavior is executed to complete the construction of the decision tree.

7. The resource availability method of the cloud platform system based on the intelligent decision chain according to claim 6, characterized in that: In step S1, collecting monitoring resources of the cloud platform includes collecting cloud platform service status monitoring information, operating system monitoring information, hardware status monitoring information, and cloud resource monitoring information; The cloud platform service monitoring information includes indicators such as service operation status, service logs, and service health status; The operating system monitoring information includes system kernel (hardware driver, interrupt response), system operation status, system load (CPU, memory, disk), and global handle number; The hardware status monitoring information includes physical CPU, memory, network card, PCI device, RAID card, disk, motherboard, etc. The cloud resource monitoring information includes vcpu load information, virtual memory exception information, and disk IO information; After the data collection of the monitoring resources is completed, the data is classified and sorted and then collected into the memory for data storage. The storage data type includes the indicator name, indicator type, collection time, data unit, and indicator value data item.

8. The resource availability method of the cloud platform system based on the intelligent decision chain according to claim 6, characterized in that: The feature data manually entered by the user includes: the input information includes the feature name, feature classification, feature detection index, and the execution strategy after the feature is met; the feature name is used to identify the unique feature; the feature classification is used to classify the resources to which the feature belongs, including a variety of different cloud platform resource types such as virtual machine resources, computing node resources, SDN services, cloud services, and storage services.

9. The resource availability method of the cloud platform system based on the intelligent decision chain according to claim 8, characterized in that: The execution strategy corresponds to the type of cloud resource: For virtual machine type resources, the execution strategy includes configurable virtual machine migration, restarting the virtual machine operating system, abnormal virtual peripheral replacement (USB), creating a new virtual machine from a backup, and sending alarm information operations; For cloud service resources, the execution strategy includes restarting the service to attempt to recover the service; For node resources, the execution strategies include virtual machine downtime migration, network service restart, and system parameter tuning operations; For SDN, the execution strategy includes cache cleaning, parameter tuning, and service restart operations; For storage resources, the execution measurement includes restarting the ceph monitor service and reconnecting the iscsi service.

10. The resource availability method of the cloud platform system based on the intelligent decision chain according to claim 6, characterized in that: After completing the construction of the decision tree, it also includes: The decision chain data is saved. When the same scenario occurs next time, the feature determination and decision execution are completed according to the decision chain saved last time. The decision tree data is a tree structure data, including feature points constituting each tree node, each node pointing to a child feature node or a parent feature node; the leaves of the decision tree are decision nodes, representing execution actions that satisfy all feature points of the linked list; The data structure of the tree node includes: the previous decision node, the next feature node, the node type and feature value identifier, and the feature point table identifier.

Images