Multi-dimensional predictable load balancing fault switching method and system

Abstract

The invention provides a multidimensional predictable load balancing fault switching method and a multidimensional predictable load balancing fault switching system. The method comprises the following steps: constructing a machine learning model based on time sequence data of historical fault probabilities, and predicting the fault probabilities of fault granularities in a preset period of time in the future by adopting the machine learning model; and when the predicted fault probability exceeds a corresponding standard threshold value, triggering a preheating mechanism for disaster recovery switching, including operations of hot standby starting, state synchronization and standby cluster traffic undertaking. According to the technical scheme, through the multi-dimensional prediction and flexible switching mechanism, various faults in the load balancing system can be sensed and dealt with more timely and more accurately, the high availability and reliability of the load balancing system are remarkably improved, and stable operation of services and continuity of user experience are ensured.

Description

Technical Field

[0001] The present invention relates to the field of load balancing network technology, and in particular to a multi-dimensionally predictable load balancing fault switching method and system. Background Art

[0002] As enterprise business grows and architecture evolves, the load balancing deployment model of a single data center is insufficient to cope with risks caused by network failures, hardware failures and other problems.

[0003] Therefore, the load balancing disaster recovery mode of multiple data centers can be used to improve system availability and business continuity.

[0004] Generally speaking, disaster recovery can be divided into different concepts based on the mode, such as cold standby and hot standby, or active-active standby and active-active.

[0005] The following describes the active-standby and active-active modes.

[0006] Active / standby means that only the active node processes all traffic at the same time, and the standby node takes over all its traffic only after the active node fails.

[0007] Active-active means that both the primary node and the backup node are processing traffic at the same time. When a node fails, the other node actively takes over its traffic.

[0008] Research has found that how to quickly perceive failures and quickly perform disaster recovery switching in load balancing scenarios, thereby improving the high availability of the system, is an issue that technical personnel in this field urgently need to solve.

[0009] In the load balancing architecture of existing products, there are several main steps for disaster recovery switching: perception of failure, which is mainly achieved through active detection on the control plane or mutual detection on the data plane between nodes.

[0010] Set the conditions for triggering disaster recovery, such as determining whether the load balancing virtual machine has failed, determining whether the cluster it is in has failed, etc.

[0011] When the failover conditions are met, the global control plane and the local control plane are combined to take over the services of the failed node.

[0012] Further analysis reveals that current technologies generally have the following shortcomings: First, the real-time performance is not high: neither the control plane nor the data plane detection mode can detect the occurrence of faults in a timely manner.

[0013] The second aspect is low accuracy: the accuracy of fault perception is low. For example, a cluster failure does not mean a business failure, and unnecessary switching may occur.

[0014] Therefore, the above two shortcomings are particularly important in load balancing scenarios, because they often carry important traffic and have extremely high requirements for stability and availability. Summary of the Invention

[0015] How to quickly perceive failures and quickly perform disaster recovery switching in a load balancing scenario, thereby improving the high availability of the system, is the problem solved by the present invention.

[0016] The present invention provides a multi-dimensional predictable load balancing fault switching method, comprising the following steps: Implementing the current system status collection step: Based on the attribute characteristic information of the load balancing system, collecting multi-dimensional fault indicators and the impact factors of different granularity faults corresponding to the fault indicators to form current system status data, and summarizing the current system status data to form a data set; Normalization and historical data analysis steps: normalize the collected current system status data; and also include establishing a mapping relationship table based on the data set formed from the aggregated current system status data, wherein the vertical items in the mapping relationship table are fault indicators, and are labeled F1, F2, ... Fn from top to bottom; and the horizontal items in the mapping relationship table are the impact factors of faults of different granularities corresponding to the fault indicators, and the granularities are labeled L1, L2, L3, L4 from left to right; According to the preset formula, the failure probability corresponding to different granularities is calculated: failure probability of L1, failure probability of L2, failure probability of L3 and failure probability of L4; The failure probability corresponding to each current fault granularity is calculated according to the above preset formula and finally saved in the database. As time goes by, the failure probability corresponding to a set of time series is calculated in the database to obtain time series data of historical failure probabilities. A machine learning model is constructed based on the time series data of historical failure probabilities, and the machine learning model is used to predict the failure probability of each fault granularity within a preset period of time in the future. When the predicted failure probability exceeds the corresponding standard threshold, the preheating mechanism for disaster recovery switching is triggered.

[0017] Preferably, as an implementable method, the load balancing failover method supports the active-standby mode at the LBGroup granularity and the active-active mode at the LB virtual machine granularity.

[0018] Preferably, as an implementable embodiment, the attribute characteristic information of the load balancing system includes network connectivity information, switching frequency information, network status information and virtual machine information.

[0019] Preferably, as an implementable embodiment; the multi-dimensional types of fault indicators include network connectivity between LB virtual machines in the LB virtual machine group, network connectivity between the LB global controller and the LB virtual machine group, the number or frequency of virtual services that undergo HA switching at the LB virtual machine group level, the number or frequency of virtual services that undergo HA switching at the VM level, the number or frequency of virtual services that undergo HA switching at the network card level, the number or frequency of virtual services that undergo HA switching at the virtual service level, the packet loss rate or retransmission rate of business-related network cards, the packet loss rate or retransmission rate of all network cards in the VM, the packet loss rate or retransmission rate of all network cards in the LB virtual machine group, a significant increase or instability in business network latency, a sudden increase or decrease in network card throughput, the health status between virtual machines, virtual machine monitoring indicator data, virtual machine internal system log data, the health status between virtual services and back-end servers, the packet loss rate or retransmission rate between virtual services and back-end servers, and the number of new connections or changes in the total number of connections of virtual services.

[0020] Preferably, as an implementable method, in the step of normalizing the collected current system status data, the normalization process uses a preset formula that normalizes the data of the influencing factors of different granularity faults to the range of [0,1]; the preset formula is .

[0021] Preferably, as an implementable embodiment, the machine learning model is selected from one of an LSTM model or a Transformer model.

[0022] Preferably, as an implementable method, the preheating mechanism includes starting hot standby in advance, synchronizing the virtual service status to the standby cluster, and taking over part of the virtual service traffic in the standby cluster in advance.

[0023] Preferably, as an implementable method, the multi-dimensional types perform fault switching in sequence according to the priority relationship of LBGroup granularity, LB virtual machine granularity, virtual service group granularity and virtual service granularity.

[0024] The present invention provides a multi-dimensional predictable load balancing fault switching system, including an acquisition module, a normalization processing module, a historical data analysis module, and a model processing module: The acquisition module is configured to collect multi-dimensional fault indicators and influencing factors of faults of different granularities corresponding to the fault indicators based on the attribute characteristic information of the load balancing system, form current system status data, and aggregate the current system status data to form a data set; The normalization processing module is configured to normalize the collected current system status data; establish a mapping relationship table based on a data set formed from the aggregated current system status data, wherein the vertical items in the mapping relationship table are fault indicators and are labeled F1, F2, ..., Fn from top to bottom; the horizontal items in the mapping relationship table are influencing factors of faults of different granularities corresponding to the fault indicators, and the granularities are labeled L1, L2, L3, and L4 from left to right; and calculate the fault probabilities corresponding to the different granularities according to a preset formula: i.e., the fault probability of L1, the fault probability of L2, the fault probability of L3, and the fault probability of L4; The historical data analysis module is used to calculate the failure probability corresponding to each current failure granularity according to the above preset formula and finally save it in the database; as time goes by, the failure probability corresponding to a set of time series is calculated in the database to obtain the time series data of historical failure probabilities; The model processing module is used to build a machine learning model based on the time series data of the historical failure probability, and use the machine learning model to predict the failure probability of each failure granularity within a preset period of time in the future; when the predicted failure probability exceeds the corresponding standard threshold, the preheating mechanism of disaster recovery switching is triggered.

[0025] Compared with the prior art, the embodiments of the present invention have at least the following technical effects: In summary, the multi-dimensionally predictable load balancing fault switching solution provided by the above technical solution of the present invention includes: a current system status collection step: collecting multi-dimensional fault indicators and influencing factors of faults of different granularities corresponding to the fault indicators based on the attribute characteristic information of the load balancing system to form current system status data, and summarizing the current system status data to form a data set; a normalization processing and historical data analysis step: normalizing the collected current system status data; and also including establishing a mapping relationship table based on the data set formed by summarizing the current system status data, wherein the vertical items of the mapping relationship table are fault indicators, and are marked as F1, F2...Fn from top to bottom; the horizontal items of the mapping relationship table are influencing factors of faults of different granularities corresponding to the fault indicators, and the granularities are marked as L1, L2, L3, L4 from left to right; According to a preset formula, the failure probabilities corresponding to different granularities are calculated: namely, the failure probability of L1, the failure probability of L2, the failure probability of L3, and the failure probability of L4. The failure probability corresponding to each current fault granularity is calculated according to the preset formula and finally saved in the database. As time goes by, the failure probabilities corresponding to a set of time series are calculated in the database to obtain time series data of historical failure probabilities. A machine learning model is constructed based on this time series data of historical failure probabilities, and the machine learning model is used to predict the failure probability of each fault granularity within a preset period of time in the future. When the predicted failure probability exceeds the corresponding standard threshold, the disaster recovery switch preheating mechanism is triggered, including hot standby startup, status synchronization, and backup cluster traffic taking over.

[0026] The present invention provides a multi-dimensional predictable load balancing fault switching method, which achieves significant technical effects through the following technical means: By collecting and analyzing load balancing indicators in multiple dimensions (such as network connectivity, HA switching frequency, packet loss rate, retransmission rate, CPU utilization, etc.), the operating status of the load balancing system is monitored comprehensively and meticulously, effectively covering potential failure points at different levels and granularities.

[0027] Leveraging historical data analysis and advanced machine learning models (such as LSTM and Transformer), we accurately predict the probability of failure at various fault granularities within a certain period of time. This predictive capability enables the system to identify potential failures in advance and mitigate the impact of sudden failures on services.

[0028] Based on the prediction results of different fault granularities (such as LBGroup granularity, LB virtual machine granularity, virtual service group granularity and virtual service granularity), the corresponding level of disaster recovery switching strategy (active-standby mode or active-active mode) is dynamically selected to ensure efficient response in various fault scenarios and guarantee business continuity.

[0029] At the same time, perform disaster recovery switch preheating preparation: When a possible failure is predicted, start the disaster recovery switch preheating mechanism in advance, including hot standby activation, status synchronization, and backup cluster traffic acceptance.

[0030] This warm-up process significantly improves the response speed and switching success rate of disaster recovery, and reduces the business interruption time when an actual failure occurs.

[0031] The technical solution adopted in the embodiment of the present invention uses the above-mentioned multi-dimensional prediction and flexible switching mechanism. The present invention can perceive and respond to various faults in the load balancing system more promptly and accurately, significantly improve the high availability and reliability of the load balancing system, ensure the stable operation of the business and the continuity of user experience, and has obvious technical advantages and practical application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the structures shown in these drawings without paying any creative work.

[0033] Figure 1 This is a flow chart of a multi-dimensional predictable load balancing failover method provided by an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of a multi-dimensional predictable load balancing fault switching system provided by an embodiment of the present invention; Figure 3 This is a schematic diagram of the mechanism architecture principle of a multi-dimensional predictable load balancing failover system; Figure 4 It is a flowchart of a multi-dimensional predictable load balancing failover method; Figure 5 It is a schematic diagram of a set of time-series related failure probabilities that a multi-dimensional predictable load balancing failover system will obtain in the database; Figure 6 This is a schematic diagram of the effect of the warm-up mechanism for triggering disaster recovery switching in a multi-dimensional predictable load balancing fault switching system.

[0034] Label: acquisition module 10; normalization processing module 20; historical data analysis module 30; model processing module 40.

[0035] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0036] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0037] In order to solve the problems in the prior art, the following embodiments of the present invention provide a multi-dimensional predictable load balancing fault switching solution, which aims to solve the problem of being able to quickly perceive faults and quickly perform disaster recovery switching in the above-mentioned load balancing scenario, thereby improving the high availability of the system.

[0038] Example 1 To achieve the above purpose, see Figure 1 , Figure 1 A flowchart of a multi-dimensional predictable load balancing fault switching method provided by an embodiment of the present invention.

[0039] like Figure 1 As shown, a multi-dimensional predictable load balancing fault switching method provided by the first embodiment of the present invention includes the following steps: S110: Implementing a current system status collection step: Based on the attribute characteristic information of the load balancing system, collecting multi-dimensional fault indicators and the impact factors of faults of different granularities corresponding to the fault indicators to form current system status data, and summarizing the current system status data to form a data set; S120, normalization processing and historical data analysis step: normalizing the collected current system status data; also including establishing a mapping relationship table based on the data set formed from the aggregated current system status data, wherein the vertical items in the mapping relationship table are fault indicators, and are labeled F1, F2 ... Fn from top to bottom; and the horizontal items in the mapping relationship table are influencing factors of faults of different granularities corresponding to the fault indicators, and the granularities are labeled L1, L2, L3, L4 from left to right; S130. Calculate the failure probabilities corresponding to different granularities according to a preset formula: failure probability of L1, failure probability of L2, failure probability of L3, and failure probability of L4; S140. Calculate the current failure probability corresponding to each fault granularity according to the above-mentioned preset formula and save it in a database. As time goes by, calculate the failure probability corresponding to a set of time series in the database to obtain time series data of historical failure probabilities. Build a machine learning model based on the time series data of historical failure probabilities, and use the machine learning model to predict the failure probability of each fault granularity within a preset period of time in the future. S150: When the predicted failure probability exceeds the corresponding standard threshold, the preheating mechanism for disaster recovery switching is triggered, including hot standby start-up, state synchronization, and backup cluster traffic taking over.

[0040] Preferably, as an implementable method, the load balancing failover method supports the active-standby mode at the LBGroup granularity and the active-active mode at the LB virtual machine granularity.

[0041] Preferably, as an implementable embodiment, the attribute characteristic information of the load balancing system includes network connectivity information, switching frequency information, network status information and virtual machine information.

[0042] Preferably, as an implementable method; in the system status collection step, the multi-dimensional types of fault indicators collected include the network connectivity between LB VMs (i.e., LB virtual machines) in the LB virtual machine group (i.e., LBGroup), the network connectivity between the LB global controller (i.e., Controller) and the LB virtual machine group, the number or frequency of virtual services that undergo HA switching at the LB virtual machine group level, the number or frequency of virtual services that undergo HA switching at the VM level, the number or frequency of virtual services that undergo HA switching at the network card level, the number or frequency of virtual services that undergo HA switching at the virtual service level, the packet loss rate or retransmission rate of business-related network cards, the packet loss rate or retransmission rate of all network cards in the VM, the packet loss rate or retransmission rate of all network cards in the LB virtual machine group, a significant increase or instability in business network latency, a sudden increase or decrease in network card throughput, the health status between virtual machines, virtual machine monitoring indicator data, virtual machine internal system log data, the health status between the virtual service and the back-end server, the packet loss rate or retransmission rate between the virtual service and the back-end server, and the number of new connections or changes in the total number of connections of the virtual service.

[0043] Preferably, as an implementation method, in the step of normalizing the collected current system status data, the normalization process uses a preset formula to normalize the data of the influencing factors of faults of different granularities to the range of [0, 1] (i.e., according to the preset formula, the fault probabilities corresponding to different granularities are calculated); the preset formula is .

[0044] Preferably, as an implementable embodiment; in the future failure probability prediction step, the machine learning model is selected from one of an LSTM model or a Transformer model.

[0045] Preferably, as an implementable method, in the disaster recovery switching preheating preparation step, the preheating mechanism includes starting hot standby in advance, synchronizing the virtual service status to the standby cluster, and taking over part of the virtual service traffic in the standby cluster in advance.

[0046] Preferably, as an implementable method, the multiple dimensions perform fault switching in sequence according to the priority relationship of LBGroup granularity, LB virtual machine granularity, virtual service group granularity and virtual service granularity.

[0047] Leveraging historical failure probability time series data, appropriate machine learning models (such as LSTM and Transformer) are employed to predict the failure probability at various fault granularities over a period of time. Based on predefined thresholds (e.g., 80% for LBGroup granularity, 70% for VM granularity, and 50% for virtual service group and virtual service granularity), a corresponding disaster recovery switchover is triggered when the predicted probability exceeds the threshold.

[0048] After predicting a potential failure, the system preheats for disaster recovery switching, including: Hot standby is enabled: The status of all virtual services on the active cluster is synchronized to the standby cluster to ensure that the standby cluster is capable of taking over.

[0049] Traffic transfer: The standby cluster begins to transfer some virtual service traffic to ensure a quick response when a switch occurs, minimizing service interruption.

[0050] The technical solution implemented in this embodiment of the present invention achieves highly accurate and timely detection of load balancing system failures through real-time collection of multi-dimensional data and historical analysis and prediction. It triggers disaster recovery switchover based on the granularity of the failure, significantly improving the high availability and business continuity of the load balancing system. It is applicable to load balancing systems in various complex network environments.

[0051] Example 2 In addition, based on the same concept of the above-mentioned method embodiment, the embodiment of the present invention also provides a multi-dimensional predictable load balancing fault switching system for implementing the above-mentioned method of the present invention. Since the principles and methods of solving the problems in the system embodiment are similar, it has at least all the beneficial effects brought by the technical solutions of the above-mentioned embodiments, and will not be repeated here one by one.

[0052] Combine Figure 2 The multi-dimensional predictable load balancing failover system shown is as follows: See also Figure 2 A multi-dimensional predictable load balancing fault switching system provided by the second embodiment of the present invention includes a collection module 10, a normalization processing module 20, a historical data analysis module 30, and a model processing module 40: The collection module 10 is used to collect multi-dimensional fault indicators and the impact factors of faults of different granularities corresponding to the fault indicators based on the attribute characteristic information of the load balancing system to form current system status data, and to summarize the current system status data to form a data set; Normalization processing module 20 is configured to normalize the collected current system status data; establish a mapping relationship table based on a data set formed from the aggregated current system status data, wherein the vertical items in the mapping relationship table are fault indicators, labeled F1, F2, ..., Fn from top to bottom; and the horizontal items in the mapping relationship table are the impact factors of faults of different granularities corresponding to the fault indicators, labeled L1, L2, L3, L4 from left to right; and calculate the fault probabilities corresponding to the different granularities according to a preset formula: i.e., the fault probability of L1, the fault probability of L2, the fault probability of L3, and the fault probability of L4. The historical data analysis module 30 is used to calculate the current failure probability corresponding to each fault granularity according to the above preset formula and finally save it in the database; as time goes by, the failure probability corresponding to a set of time series is calculated in the database to obtain the time series data of historical failure probability; The model processing module 40 is used to build a machine learning model based on the time series data of the historical failure probability, and use the machine learning model to predict the failure probability of each failure granularity within a preset period of time in the future; when the predicted failure probability exceeds the corresponding standard threshold, the preheating mechanism of disaster recovery switching is triggered, including the operations of hot standby startup, status synchronization and backup cluster traffic taking over.

[0053] The following first describes the system architecture on which the invention mechanism is based, wherein lbgroup represents a set of load-balancing virtual machines.

[0054] superior Figure 3 Two disaster recovery modes are described: Mode 1, the LBGroup granularity is the active-standby mode; Mode 2, the LB virtual machine granularity is the active-active mode.

[0055] Both modes are triggered by sensing failures in the graph.

[0056] superior Figure 3 A represents an active virtual service, and S represents an inactive virtual service.

[0057] A' represents the deployment of an active virtual service on another LBGroup. Similarly, S' represents the deployment of an inactive virtual service on another LBGroup.

[0058] When a cluster-level or disaster recovery switch occurs, lbgroup-2 takes over the services in lbgroup-1, and A' corresponds to A, and S' corresponds to S. When a virtual machine-level switch occurs, services are switched between virtual machines in the lbgroup, and A corresponds to S, and A' corresponds to S'.

[0059] Fault detection, based on the load balancing system's control plane, collects real-time data from various dimensions within the load balancing system. By analyzing this multi-dimensional data, it determines the probability of a fault occurring now or within a certain period of time, triggering the system's internal disaster recovery switchover logic. This allows for preemptive disaster recovery switchover before the system actually fails or services become completely unavailable.

[0060] The second embodiment of the present invention provides a multi-dimensional predictable load balancing fault switching method, which mainly implements the processing steps of current system status collection, historical data analysis, future failure probability prediction, disaster recovery switching preheating preparation, etc. Figure 4 .

[0061] When executing the current system status collection and processing, based on the characteristics of the load balancing itself and the various levels of disaster recovery it can support, multi-dimensional types of fault indicators are collected to form the current status set.

[0062] First, the present invention supports failover at LBGroup granularity, LB virtual machine granularity, virtual service group, and virtual service granularity.

[0063] Among them, their priority relationship is: LBGroup granularity > LB virtual machine granularity > virtual service group granularity > virtual service granularity; secondly, for failover modes of different granularities, corresponding indicator data is collected.

[0064] Below, we list all possible indicators that may reflect faults and their impact factors on faults of different granularities. The impact factors are divided into three levels from high to low: 1, 0.5, and 0: ;

[0065] Historical data analysis and processing: Based on the data obtained by the acquisition module 10, the historical data analysis module will first normalize the data and normalize the collected data to the range of [0,1]. The formula used is: ; Then calculate the current failure probability based on the indicators and the corresponding influencing factors.

[0066] The various indicators in the above table are marked as F1, F2...Fn from top to bottom, and the granularity is marked as L1, L2, L3, L4 from left to right.

[0067] Then the current failure probability of L1 is F1*1+F2*0.5+F3*1+F4*0.5+....+Fn*0; The failure probability of L2 is F1*1+F2*0.5+F3*1+F4*1+....+Fn*0; The failure probability of L3 is F1*1+F2*0.5+F3*1+F4*1+....+Fn*0; The failure probability of L4 is F1*1+F2*0.5+F3*1+F4*1+....+Fn*1; Then, the fault probability corresponding to each fault granularity is calculated according to the above formula and finally saved in the database. As time goes by, a set of time series related fault probabilities will be obtained in the database, as follows Figure 5 shown.

[0068] Realize the prediction of future failure probability; based on the historical data analysis module, you can see the failure rate corresponding to each disaster recovery granularity at each previous moment.

[0069] Based on empirical values ​​or user-defined thresholds, disaster recovery is triggered when the failure rate reaches a certain threshold. For example, the failure threshold at the LBGroup level is 80, the failure threshold at the VM level is 70, and the failure thresholds at both the virtual service group and virtual service levels are 50.

[0070] Through time series, you can see all the moments in historical data when the threshold was reached and the disaster recovery switch was triggered. Here, you can predict the probability of failure occurring in a certain period of time in the future based on the historical curve.

[0071] You can use basic machine learning methods or models such as LSTM, transformer, etc.

[0072] Finally, disaster recovery switching preheating preparation: once a possible failure is predicted, the system will promptly trigger the disaster recovery switching preheating mechanism, that is, before the failure actually occurs, disaster recovery preparation work will be started. The main highlights are: 1. Start hot standby in advance and synchronize the status of all virtual services on the primary cluster to the backup cluster, so as to improve timeliness when switching occurs.

[0073] 2. The standby cluster begins to take over some virtual service traffic. That is, newly created virtual services take effect on the standby cluster first.

[0074] Display effects, such as Figure 6 As shown, Figure 6 Medium gray indicates the faulty area, and white indicates the normal area. This describes the following: after disaster recovery is predicted at the LBGroup and VM granularity, the active LB VM group synchronizes the status of all its virtual services with the standby LB VM group, and the active LB VM synchronizes the status of all its virtual services with the standby LB VM.

[0075] At the same time, the new virtual service will be deployed in a certain proportion and take effect in the standby LB virtual machine and the standby LB virtual machine group.

[0076] Similarly, if it is at the virtual service group level or virtual service level, the above two disaster recovery switching preparations are also adopted.

[0077] In summary, the present invention provides a multi-dimensional and predictable load balancing failover method, which achieves significant technical effects through the following technical means: By collecting and analyzing load balancing indicators in multiple dimensions (such as network connectivity, HA switching frequency, packet loss rate, retransmission rate, CPU utilization, etc.), the operating status of the load balancing system is monitored comprehensively and meticulously, effectively covering potential failure points at different levels and granularities.

[0078] Leveraging historical data analysis and advanced machine learning models (such as LSTM and Transformer), we accurately predict the probability of failure at various fault granularities within a certain period of time. This predictive capability enables the system to identify potential failures in advance and mitigate the impact of sudden failures on services.

[0079] Based on the prediction results of different fault granularities (such as LBGroup granularity, LB virtual machine granularity, virtual service group granularity and virtual service granularity), the corresponding level of disaster recovery switching strategy (active-standby mode or active-active mode) is dynamically selected to ensure efficient response in various fault scenarios and guarantee business continuity.

[0080] At the same time, perform disaster recovery switch preheating preparation: When a possible failure is predicted, start the disaster recovery switch preheating mechanism in advance, including hot standby activation, status synchronization, and backup cluster traffic acceptance.

[0081] This warm-up process significantly improves the response speed and switching success rate of disaster recovery, and reduces the business interruption time when an actual failure occurs.

[0082] The technical solution adopted in the embodiment of the present invention uses the above-mentioned multi-dimensional prediction and flexible switching mechanism. The present invention can perceive and respond to various faults in the load balancing system more promptly and accurately, significantly improve the high availability and reliability of the load balancing system, ensure the stable operation of the business and the continuity of user experience, and has obvious technical advantages and practical application value.

[0083] Those skilled in the art should appreciate that the embodiments of the present invention may be provided as a method, a system or a computer program product.

[0084] Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects.

[0085] Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, and optical storage) containing computer-usable program codes.

[0086] The present invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the invention.

[0087] It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0088] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0089] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0090] It should be noted that, in the claims, any reference signs placed between parentheses shall not be construed as limiting the claim.The word "comprising" does not exclude the presence of elements or steps not listed in a claim.

[0091] The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements.

[0092] The invention can be implemented by means of hardware comprising several distinct elements, and by means of a suitably programmed computer. In a unit claim enumerating several means, several of these means may be embodied by one and the same item of hardware.

[0093] The usage of the words first, second and third etc. do not indicate any order.

[0094] These words can be interpreted as names.

[0095] Although preferred embodiments of the present invention have been described, additional changes and modifications to these embodiments may occur to those skilled in the art once the basic inventive concepts become known.

[0096] Therefore, it is intended that the appended claims be interpreted as including the preferred embodiment and all changes and modifications that fall within the scope of the present invention.

[0097] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the invention.

[0098] Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is intended to include these modifications and variations.

Claims

1. A multi-dimensional predictable load balancing fault switching method, characterized in that: The following steps are involved: According to the attribute characteristic information of the load balancing system, multi-dimensional fault indicators and the influencing factors of faults of different granularities corresponding to the fault indicators are collected to form current system status data, and the current system status data is aggregated to form a data set; The collected current system status data is normalized; and a mapping relationship table is established according to the data set formed by the aggregated current system status data, wherein the vertical items of the mapping relationship table are fault indicators, and are marked as F1, F2 ... Fn from top to bottom; the horizontal items of the mapping relationship table are influencing factors of faults of different granularities corresponding to the fault indicators, and the granularities are marked as L1, L2, L3, L4 from left to right; According to the preset formula, the failure probabilities corresponding to different granularities are calculated: namely, the failure probability of L1, the failure probability of L2, the failure probability of L3, and the failure probability of L4; The fault probability corresponding to each current fault granularity is calculated according to the above preset formula and finally saved in the database; As time goes by, a set of failure probabilities corresponding to a time series is calculated in the database to obtain time series data of historical failure probabilities; a machine learning model is constructed based on the time series data of historical failure probabilities, and the machine learning model is used to predict the failure probabilities of each fault granularity within a preset period of time in the future; When the predicted failure probability exceeds the corresponding standard threshold, the disaster recovery switch preheating mechanism is triggered.

2. The method according to claim 1, characterized in that: The load balancing failover method supports the active-standby mode at the LBGroup granularity and the active-active mode at the LB virtual machine granularity.

3. The method according to claim 1, characterized in that The attribute characteristic information of the load balancing system includes network connectivity information, switching frequency information, network status information and virtual machine information.

4. The method according to claim 3, characterized in that: The multi-dimensional types of fault indicators include network connectivity between LB virtual machines in the LB virtual machine group, network connectivity between the LB global controller and the LB virtual machine group, the number or frequency of virtual services that undergo HA switching at the LB virtual machine group level, the number or frequency of virtual services that undergo HA switching at the VM level, the number or frequency of virtual services that undergo HA switching at the network card level, the number or frequency of virtual services that undergo HA switching at the virtual service level, the packet loss rate or retransmission rate of business-related network cards, the packet loss rate or retransmission rate of all network cards in the VM, the packet loss rate or retransmission rate of all network cards in the LB virtual machine group, a significant increase or instability in business network latency, a sudden increase or decrease in network card throughput, the health status between virtual machines, virtual machine monitoring indicator data, virtual machine internal system log data, the health status between virtual services and backend servers, the packet loss rate or retransmission rate between virtual services and backend servers, and the number of new connections or changes in the total number of connections for virtual services.

5. The method according to claim 4, characterized in that In the step of normalizing the collected current system status data, the normalization process uses a preset formula that normalizes the data of the influencing factors of faults of different granularities to a range of [0,1]; the preset formula is .

6. The method according to claim 1, characterized in that The machine learning model is selected from an LSTM model or a Transformer model.

7. The method according to claim 1, characterized in that The preheating mechanism includes enabling hot standby in advance, synchronizing the virtual service status to the standby cluster, and taking over part of the virtual service traffic in the standby cluster in advance.

8. The method according to claim 1, characterized in that: The multi-dimensional types perform fault switching in sequence according to the priority relationship of LBGroup granularity, LB virtual machine granularity, virtual service group granularity and virtual service granularity.

9. A multi-dimensional predictable load balancing fault switching system, characterized in that: It includes acquisition module, normalization processing module, historical data analysis module and model processing module: The acquisition module is used to collect multi-dimensional fault indicators and influencing factors of faults of different granularities corresponding to the fault indicators according to the attribute characteristic information of the load balancing system, form current system status data, and summarize the current system status data to form a data set; The normalization processing module is used to perform normalization processing on the collected current system state data; A mapping relationship table is established based on the data set formed by the aggregated current system status data, wherein the vertical items of the mapping relationship table are fault indicators, and are marked as F1, F2 ... Fn from top to bottom; the horizontal items of the mapping relationship table are influencing factors of faults of different granularities corresponding to the fault indicators, and the granularities are marked as L1, L2, L3, L4 from left to right; and the fault probabilities corresponding to different granularities are calculated according to a preset formula: namely, the fault probability of L1, the fault probability of L2, the fault probability of L3, and the fault probability of L4; The historical data analysis module is used to calculate the fault probability corresponding to each current fault granularity according to the above preset formula, and finally save it in the database; As time goes by, the failure probability corresponding to a set of time series is calculated in the database to obtain the time series data of historical failure probability; The model processing module is used to build a machine learning model based on the time series data of the historical failure probability, and use the machine learning model to predict the failure probability of each failure granularity within a preset period of time in the future; when the predicted failure probability exceeds the corresponding standard threshold, the preheating mechanism of disaster recovery switching is triggered.

Images