Multi-node cooperative control task scheduling optimization and fault-tolerant processing method

Abstract

The invention relates to the technical field of task scheduling optimization and fault-tolerant processing. The invention relates to a multi-node cooperative control task scheduling optimization and fault-tolerant processing method. The method comprises the following steps: S1, acquiring a running node list, and extracting resource configuration of running nodes, historical running tasks of the running nodes and corresponding running data; s2, performing data segment division on the operation data of the historical operation task according to the fluctuation amplitude to obtain a plurality of data segments corresponding to the historical operation task; through the scheduling and fault-tolerant integrated design, the node fault probability and the fault-tolerant resource overhead are fused into the scheduling decision, and meanwhile, the differential fault-tolerant strategy is arranged in advance in the scheduling stage, so that the bidirectional linkage of scheduling adaptive fault tolerance and fault-tolerant supporting scheduling is realized; the problems of high fault recovery time delay and resource waste caused by scheduling and fault-tolerant splitting in the prior art are effectively solved, and the pre-judgment capability and the processing efficiency of a system for coping with node faults are greatly improved.

Description

Technical Field

[0001] This invention relates to the field of task scheduling optimization and fault tolerance technology, and more specifically, to a task scheduling optimization and fault tolerance method for multi-node collaborative control. Background Technology

[0002] Multi-node collaborative control of task scheduling and fault tolerance is a core supporting technology in fields such as distributed computing, industrial internet, and cloud-edge-device collaboration. Its role is to achieve reasonable allocation and efficient execution of tasks in a system composed of multiple heterogeneous nodes, while dealing with emergencies such as node failures, resource fluctuations, and network anomalies, ensuring the stable operation of the system and the timely completion of tasks.

[0003] In practical applications, existing technologies often suffer from disconnects between scheduling and fault tolerance design. Most scheduling algorithms focus solely on single objectives such as resource utilization and latency, failing to incorporate node failure probability and fault tolerance resource overhead into scheduling decisions. Fault tolerance mechanisms are often reactive, leading to scheduling scheme failures, excessively long recovery delays, and significant waste of redundant resources. Furthermore, the threshold settings are poorly adapted to node states. Existing technologies often employ globally uniform fault judgment thresholds without considering the differences in remaining resource configurations across different nodes. This results in delayed fault tolerance triggering for nodes with sufficient remaining resources, or excessive fault tolerance for nodes with scarce remaining resources leading to resource exhaustion. Therefore, this paper proposes a task scheduling optimization and fault tolerance processing method based on multi-node collaborative control. Summary of the Invention

[0004] The purpose of this invention is to provide a task scheduling optimization and fault-tolerant processing method for multi-node collaborative control, so as to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, a task scheduling optimization and fault-tolerant processing method for multi-node collaborative control is provided, comprising the following steps:

[0006] S1. Obtain the list of running nodes, extract the resource configuration of the running nodes, and the historical running tasks and corresponding running data of the running nodes;

[0007] S2. Divide the historical running task's running data into data segments based on the fluctuation range to obtain multiple data segments corresponding to the historical running task. Then, associate the data segments with similar data to form a data segment set. Generate new data segments based on the running data and data segment length in the data segment set.

[0008] S3. Set the data truncation length according to the data segment length of historical running tasks. Truncate the latest running data at the running node according to the truncation length, extract the data segment set corresponding to the running node, and perform similarity analysis between it and the latest running data. Select the most similar target data segment, and perform fault coefficient analysis on the target data segment in combination with the resource configuration of the running node to obtain the fault coefficient corresponding to the running node.

[0009] S4. Set the fault optimization threshold according to the remaining resource configuration of each running node, compare the fault coefficient of each running node with the corresponding optimization threshold for risk, and determine that the running node is a node to be protected against fault tolerance when the fault coefficient is greater than the optimization threshold.

[0010] S5. For the node to be protected against faults, select the remaining resource configuration of other running nodes based on the corresponding target data segment, and coordinate the node to be protected against faults with the selected other running nodes to perform the same running tasks until the fault coefficient of the node to be protected against faults is less than the optimization threshold.

[0011] As a further improvement to this technical solution, in S1, a multi-node operating system is connected, and the online status of all nodes is detected through the multi-node system. The operating nodes that are in the online operating state are selected, and each operating node is given a unique identifier. Then, the unique identifiers of the operating nodes are summarized to obtain a list of operating nodes.

[0012] The current configuration values ​​and remaining available values ​​of computing power, storage, network bandwidth, and hardware load of each running node are statistically analyzed and quantified to obtain the resource configuration of the running node;

[0013] Retrieve task execution logs and data collection logs from each running node according to the time dimension, filter out completed and incomplete historical running tasks and classify them according to task number, and extract the running data of each historical running task during the execution process.

[0014] As a further improvement to this technical solution, in S2, a threshold for the fluctuation range of the running data is set, and the time series running data of the historical running tasks is segmented using the sliding window method. When the fluctuation range of the running data in the sliding window exceeds the threshold, segmentation is performed until the division of all historical running task running data is completed, resulting in multiple data segments corresponding to the historical running tasks.

[0015] A data feature matching algorithm is used to calculate the similarity value of running data between any two data segments. An association threshold is set. When the similarity value of two data segments exceeds the association threshold, the two data segments are associated. Conversely, when the similarity value of two data segments does not exceed the association threshold, the two data segments are not associated.

[0016] Among them, the same data segment can establish a relationship with multiple data segments that meet the similarity standard;

[0017] The data segment set is a dedicated data segment set that is independently divided according to the running node. The historical running task data segments of each running node are only included in its own dedicated data segment set.

[0018] As a further improvement to this technical solution, in step S2, at least two data segments are selected from the data segment set, and feature fusion and interpolation are performed on the running data of the selected data segments. At the same time, the selected data segments are spliced ​​together according to their lengths to generate a new data segment that conforms to the characteristics of the running data. The length of the new data segment is the average length of the selected data segments. After adding corresponding identification information, the generated new data segment is added to the original data segment set to complete the expansion of the data segment set.

[0019] As a further improvement to this technical solution, in step S3, the average length of all data segments of the historical running tasks corresponding to each running node is calculated, and the average length is used as the data truncation length of the latest running data of that running node.

[0020] The latest running data, matching the data truncation length, is extracted from the real-time data acquisition terminals of each running node in chronological order from newest to oldest.

[0021] Extract the dedicated data segment set corresponding to each running node, perform similarity analysis between the latest running data and each data segment in the data segment set, obtain the similarity value between the latest running data and each data segment, and then select the data segment with the largest similarity value as the target data segment.

[0022] As a further improvement to this technical solution, in step S3, the running data of the target data segment is subjected to multi-dimensional fitting analysis with the resource configuration of the running node, the fault weight coefficients of each dimension are set, the initial fault coefficient of the running node is obtained by weighted calculation, and then the initial fault coefficient is corrected according to the length of the target data segment. The corrected value is the final fault coefficient of the running node.

[0023] The higher the final failure coefficient value, the higher the risk of node failure;

[0024] The smaller the final failure coefficient, the lower the risk of node failure.

[0025] As a further improvement to this technical solution, in step S4, the remaining resource configuration is analyzed based on the resource configuration corresponding to the running node to obtain the remaining resource configuration of each running node;

[0026] Resource allocation consists of used resource allocation and remaining resource allocation;

[0027] Summarize the remaining resource configuration of each running node and set the fault optimization threshold;

[0028] The more remaining resources are allocated, the lower the fault optimization threshold will be;

[0029] The less remaining resources are allocated, the higher the fault optimization threshold.

[0030] Compare the failure coefficient of each running node with the corresponding optimization threshold for risk assessment;

[0031] When the fault coefficient is greater than the optimization threshold, the running node is determined to be a node to be protected against fault tolerance.

[0032] If the fault coefficient is less than the optimization threshold, the running node is determined not to be a node to be protected against faults.

[0033] As a further improvement to this technical solution, in step S5, for the node to be protected against fault, the resource requirement parameters of the target data segment of the node to be protected against fault are calculated.

[0034] Based on the resource requirement parameter, other running nodes in the multi-node system whose remaining resource configuration meets the resource requirement parameter are selected. Priority is given to the normally operating nodes whose remaining resource configuration value is higher than the benchmark parameter and whose node load rate is the lowest. Among them, other running nodes can be one or more.

[0035] As a further improvement to this technical solution, in step S5, all the running tasks currently being executed by the fault-tolerant protection node are transmitted to other selected running nodes according to the task synchronization method, and the fault-tolerant protection node and the other selected nodes are coordinated and scheduled to execute the same running tasks.

[0036] During the collaborative scheduling process, the fault coefficient of the node to be protected is monitored in real time. When the fault coefficient is lower than the optimization threshold, the fault protection of other selected running nodes is stopped, and the independent task scheduling of each node is restored.

[0037] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0038] 1. In this multi-node collaborative control task scheduling optimization and fault-tolerant processing method, the scheduling and fault tolerance are integrated into the scheduling decision by designing scheduling and fault tolerance in an integrated manner. At the same time, differentiated fault tolerance strategies are deployed in advance during the scheduling phase, realizing the two-way linkage of scheduling adapting to fault tolerance and fault tolerance supporting scheduling. This effectively solves the problems of high fault recovery latency and resource waste caused by the separation of scheduling and fault tolerance in the existing technology, and greatly improves the system's ability to predict and process node faults.

[0039] 2. In this multi-node collaborative control task scheduling optimization and fault tolerance processing method, a dedicated fault optimization threshold is dynamically set based on the remaining resource configuration of the node. Combined with the length and characteristics of historical data segments, a structured processing and expansion is performed to construct a set of dedicated data segments for each node. This ensures that the threshold for fault judgment and the benchmark for data matching can be adapted to the individual characteristics of the node, avoiding the judgment bias caused by uniform settings across the entire domain. This significantly improves the accuracy of fault risk judgment and ensures that tolerance is provided where necessary and not required where not necessary.

[0040] 3. In this multi-node collaborative control task scheduling optimization and fault tolerance processing method, the cosine similarity algorithm is used to achieve accurate matching between the latest running data and historical data segments. Combined with multi-dimensional fitting and length correction, the fault coefficient is calculated, making the fault risk quantification more in line with the actual running status of the nodes. Compared with the existing fuzzy qualitative judgment or single-dimensional quantification methods, the accuracy and reference value of the fault coefficient are greatly improved, providing a reliable basis for fault tolerance decision-making. Attached Figure Description

[0041] Figure 1 This is a flowchart illustrating a task scheduling optimization and fault-tolerant processing method for multi-node collaborative control according to the present invention.

[0042] Figure 2 This is a flowchart of S1 of the present invention;

[0043] Figure 3 This is a flowchart of S2 of the present invention;

[0044] Figure 4 This is a flowchart of S3 of the present invention;

[0045] Figure 5 This is a flowchart of S4 of the present invention;

[0046] Figure 6 This is a flowchart of S5 of the present invention. Detailed Implementation

[0047] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on 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.

[0048] Please see Figures 1-6 As shown, the purpose of this embodiment is to provide a task scheduling optimization and fault-tolerant processing method for multi-node collaborative control, including the following steps:

[0049] S1. Obtain the list of running nodes, extract the resource configuration of the running nodes, as well as the historical running tasks and corresponding running data of the running nodes; complete the full collection, quantification and structured processing of the three types of information: node identity, resource configuration and historical task data, to ensure that the data source for subsequent data processing and fault calculation is real, complete and matchable;

[0050] In step S1, a multi-node operating system is connected, and the online status of all nodes in the domain is detected through the multi-node system. The operating nodes that are in the online operating state are selected and each operating node is given a unique identifier. Then, the unique identifiers of the operating nodes are summarized to obtain a list of operating nodes.

[0051] The current configuration values ​​and remaining available values ​​of computing power, storage, network bandwidth, and hardware load of each running node are statistically analyzed and quantified to obtain the resource configuration of the running node;

[0052] Retrieve task execution logs and data collection logs from each running node according to the time dimension, filter out completed and incomplete historical running tasks and classify them according to task number, and extract the running data of each historical running task during the execution process.

[0053] S2. Divide the historical running task's running data into data segments based on fluctuation amplitude to obtain multiple data segments corresponding to the historical running task. Then, perform similarity association on each data segment to form a data segment set. Generate new data segments based on the running data and data segment length in the data segment set. Perform three-step processing of the historical running data collected in S1: segmentation, association, and expansion. Transform the original disordered time series data into a structured, specialized, and enriched data segment set to solve the problem of subsequent similarity matching deviation caused by insufficient historical data samples and unclear data features.

[0054] In step S2, a threshold for the fluctuation range of the running data is set. The sliding window method is used to segment the time series running data of historical running tasks. When the fluctuation range of the running data within the sliding window exceeds the threshold, segmentation is performed until the running data of all historical running tasks is divided, resulting in multiple data segments corresponding to the historical tasks. The formula is as follows:

[0055] ;

[0056] in, The fluctuation range of the running data within the k-th sliding window of the j-th historical running task at the i-th node. This is the set of running data within the k-th sliding window of the j-th historical running task at the i-th node. for The maximum value of the running data, for The minimum value of the running data, The threshold for the fluctuation range of the running data;

[0057] when At that time, segmentation is performed at the k-th sliding window;

[0058] A data feature matching algorithm is used to calculate the similarity value of running data between any two data segments. An association threshold is set. When the similarity value of two data segments exceeds the association threshold, the two data segments are associated; otherwise, when the similarity value of two data segments does not exceed the association threshold, the two data segments are not associated. The formula is as follows:

[0059] ;

[0060] in, Let X be the cosine similarity (i.e., similarity value) between data segment X and data segment Y. Let m be the m-th running data feature value of data segment X. Let m be the m-th running data feature value of data segment Y. The total number of runtime data feature dimensions for a single data segment. This is the correlation threshold;

[0061] when At that time, data segment X and data segment Y establish a relationship;

[0062] Among them, the same data segment can establish a relationship with multiple data segments that meet the similarity standard;

[0063] The data segment set is a dedicated data segment set that is independently divided according to the running node. The historical running task data segments of each running node are only included in its own dedicated data segment set.

[0064] In step S2, at least two data segments are selected from the data segment set. Feature fusion and interpolation are performed on the running data of the selected data segments. Simultaneously, the selected data segments are concatenated based on their lengths to generate new data segments that conform to the characteristics of the running data. The length of the new data segment is the average length of the selected data segments. After adding corresponding identification information, the generated new data segment is added to the original data segment set to complete the expansion of the data segment set. The formula is as follows:

[0065] ;

[0066] in, The length of the newly generated data segment, The number of original data segments selected from the data segment set. Let p be the length of the p-th selected original data segment;

[0067] ;

[0068] in, For the m-th feature value of the merged running data, Let p be the fusion weight of the p-th original data segment. This represents the m-th running data feature value of the p-th original data segment;

[0069] ;

[0070] in, The feature values ​​of the running data after linear interpolation at time t are: for Feature values ​​of runtime data after time-mapping for Feature values ​​of runtime data after time-mapping The target time point for interpolation completion;

[0071] S3. Set the data truncation length based on the data segment length of historical running tasks. Truncate the latest running data at the running node according to the truncation length, extract the data segment set corresponding to the running node, and perform similarity analysis between it and the latest running data. Select the most similar target data segment, and perform fault coefficient analysis on the target data segment in combination with the resource configuration of the running node to obtain the fault coefficient corresponding to the running node. Connecting the resource configuration of S1 and the data segment set of S2, through the latest data truncation, similarity matching, and fault coefficient calculation and correction, the real-time running status of the node is transformed into a quantifiable fault coefficient, realizing the digital expression of fault risk and replacing the fuzzy qualitative judgment.

[0072] In step S3, the average length of all data segments of the historical running tasks corresponding to each running node is calculated, and the average length is used as the data truncation length of the latest running data of that running node.

[0073] The latest running data, matching the data truncation length, is extracted from the real-time data acquisition terminals of each running node in chronological order from newest to oldest.

[0074] Extract the exclusive data segment set corresponding to each running node, perform similarity analysis between the latest running data and each data segment in the data segment set, obtain the similarity value between the latest running data and each data segment, and then select the data segment with the largest similarity value as the target data segment;

[0075] Extract the unique data segment set of each node, use the cosine similarity algorithm to calculate the similarity value between the latest extracted running data and each data segment in the set, and select the data segment with the largest similarity value as the target data segment; if there are multiple data segments with the same maximum similarity value, then all of these data segments are listed as target data segments.

[0076] In step S3, the running data of the target data segment is subjected to multi-dimensional fitting analysis with the resource configuration of the running node. Fault weight coefficients for each dimension are set, and the initial fault coefficient of the running node is obtained by weighted calculation. Then, the initial fault coefficient is corrected according to the length of the target data segment. The corrected value is the final fault coefficient of the running node.

[0077] The process involves performing minimum-maximum normalization on the running data of the target data segment and the resource configuration data of the corresponding running nodes to complete the fitting analysis of multi-dimensional features, eliminate dimensional differences, and obtain standardized fitting feature values ​​for each dimension. Then, a fault weight coefficient that meets the normalization condition is set for each fitting feature dimension. Based on the standardized fitting feature values ​​of each dimension, the initial fault coefficient of the running node is calculated by weighted summation. Finally, the normalization correction coefficient of the target data segment length is calculated, and the initial fault coefficient is linearly corrected using this coefficient to obtain the final fault coefficient of the running node. The value of the final fault coefficient is positively correlated with the node fault risk.

[0078] The higher the final failure coefficient value, the higher the risk of node failure;

[0079] The smaller the final failure coefficient, the lower the risk of node failure.

[0080] S4. Set fault optimization thresholds based on the remaining resource configuration of each running node, compare the fault coefficient of each running node with the corresponding optimization threshold for risk assessment, and determine that the running node is a node to be protected against fault tolerance when the fault coefficient is greater than the optimization threshold. Based on the remaining resource configuration of S1 and the fault coefficient of S3, the nodes that need fault tolerance protection are screened out through dynamic threshold setting and coefficient threshold comparison, avoiding the unreasonable judgment problem caused by "uniform threshold" and ensuring the adaptability of the fault tolerance judgment nodes.

[0081] In step S4, the remaining resource configuration is analyzed based on the resource configuration corresponding to the running node to obtain the remaining resource configuration of each running node;

[0082] Resource allocation consists of used resource allocation and remaining resource allocation;

[0083] From the complete resource configuration information of each running node, two categories of data are separated: used resource configuration and remaining resource configuration. The remaining resource configuration in terms of computing power, storage, network bandwidth, and hardware load is quantified and statistically analyzed to obtain the quantified value of the remaining resource configuration of each node. Then, the quantified value of the remaining resource configuration of each node is normalized, and the comprehensive value of the remaining resources is calculated. Based on the core rule that more remaining resource configuration results in a lower fault optimization threshold and fewer remaining resource configuration results in a higher fault optimization threshold, combined with the system's preset safety coefficient, a fault optimization threshold specific to each running node is set.

[0084] Summarize the remaining resource configuration of each running node and set the fault optimization threshold;

[0085] The more remaining resources are allocated, the lower the fault optimization threshold will be;

[0086] The less remaining resources are allocated, the higher the fault optimization threshold, as shown in the following formula:

[0087] ;

[0088] in, The comprehensive value for the remaining resource allocation of the i-th running node. Quantify the remaining resources in the d-th dimension of the i-th node. Quantify the total resource allocation value (used + remaining) for the i-th node in the d-th dimension.

[0089] ;

[0090] in, Optimize the fault threshold specifically for the i-th running node. The preset basic fault threshold (safety factor);

[0091] Compare the failure coefficient of each running node with the corresponding optimization threshold for risk assessment;

[0092] When the fault coefficient is greater than the optimization threshold, the running node is determined to be a node to be protected against fault tolerance.

[0093] If the fault coefficient is less than the optimization threshold, the running node is determined not to be a node to be protected against faults.

[0094] S5. For the node to be protected against faults, the remaining resource configuration of other running nodes is screened and selected based on the corresponding target data segment. The node to be protected against faults is then coordinated with the selected other running nodes to perform the same running tasks until the fault coefficient of the node to be protected against faults is less than the optimization threshold. Based on the target data segment in S3 and the node to be protected against faults in S4, the fault-tolerant protection is executed in a precise, dynamic, and efficient manner through coordinated node screening, coordinated scheduling of the same tasks, and real-time detection of closed-loop termination. This ensures optimal utilization of multi-node resources during the fault tolerance process and timely termination of fault tolerance.

[0095] In step S5, for the node to be protected against faults, the resource requirement parameters of the target data segment of the node to be protected against faults are calculated.

[0096] Based on the resource requirement parameter, other running nodes in the multi-node system whose remaining resource configuration meets the resource requirement parameter are selected. Priority is given to the normally operating node with the lowest remaining resource configuration value and the lowest node load rate. The other running nodes can be one or more. The steps are as follows:

[0097] For nodes identified as requiring fault tolerance protection, extract all matched target data segments and extract the resource usage time-series data of each target data segment within the complete execution time interval according to the four resource dimensions of computing power, storage, network bandwidth, and hardware load.

[0098] The resource usage time-series data of each dimension is statistically analyzed moment by moment. The peak resource usage of each dimension during the execution of the target data segment is selected and used as the basic resource requirement parameter of the corresponding dimension. Then, combined with the system's preset resource redundancy coefficient (to avoid insufficient demand caused by resource fluctuations), the basic resource requirement parameters of each dimension are fine-tuned to finally form a complete set of resource requirement parameters for the fault-tolerant protection node, ensuring that the parameters cover the peak and fluctuation requirements of resources in actual operation.

[0099] Based on the resource requirement parameter set of the node to be protected by fault tolerance, each normally operating node is compared dimension by dimension to check whether the remaining resource configuration value of each dimension is greater than or equal to the corresponding benchmark parameter. All candidate nodes that meet the resource requirements of all dimensions are selected. Then, the comprehensive load rate of each candidate node is calculated. The used resource configuration value and the total resource configuration value of each node in the four dimensions are statistically analyzed. The comprehensive load rate is calculated by weighted summation to eliminate the influence of load deviation in a single dimension. The candidate nodes are sorted in ascending order of comprehensive load rate. The node with the lowest load rate is selected as the collaborative fault tolerance node. According to the fault risk level of the node to be protected by fault tolerance, a single node (low risk) or multiple nodes (high risk) can be selected to form a collaborative fault tolerance cluster.

[0100] In step S5, all running tasks currently being executed by the fault-tolerant protection node are transmitted to other selected running nodes according to the task synchronization method, and the fault-tolerant protection node and other selected nodes are coordinated and scheduled to execute the same running tasks.

[0101] During the collaborative scheduling process, the fault coefficient changes of the nodes to be protected are monitored in real time. When the fault coefficient falls below the optimization threshold, the fault protection of other selected running nodes is stopped, and the independent task scheduling of each node is restored. The steps are as follows:

[0102] All running tasks currently being executed by the fault-tolerant protection node are sorted out and prioritized. The instruction set, execution context data, intermediate result data, and real-time resource usage information of each task are extracted in turn. At the same time, a dedicated data transmission channel is established between the fault-tolerant protection node and the selected collaborative fault-tolerant nodes, and unnecessary data interaction is closed to reduce transmission latency.

[0103] Through a dedicated transmission channel, the sorted task data is transmitted completely to the collaborative fault-tolerant node to ensure that the collaborative node obtains the same basic task execution information as the node to be fault-tolerant, thus avoiding task execution deviations. At the same time, the collaborative scheduling mechanism is activated to control the node to be fault-tolerant and the collaborative fault-tolerant node to execute the same running task synchronously, and to synchronize the task execution progress and resource usage data of both parties in real time to maintain the consistency of task execution status. By sharing tasks, the operating pressure of the node to be fault-tolerant is reduced.

[0104] When the fault coefficient remains below the optimization threshold, it is determined that the fault risk of the node to be protected has been reduced to a safe range. The fault protection termination command is triggered, and the node to be protected resumes independent task scheduling mode. At the same time, it cleans up temporary data in the collaboration process, closes the dedicated transmission channel, releases redundant resources of the collaboration node, and returns to normal operation.

[0105] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A task scheduling optimization and fault-tolerant processing method for multi-node collaborative control, characterized in that: Includes the following steps: S1. Obtain the list of running nodes, extract the resource configuration of the running nodes, and the historical running tasks and corresponding running data of the running nodes; S2. Divide the historical running task's running data into data segments based on the fluctuation range to obtain multiple data segments corresponding to the historical running task. Then, associate the data segments with similar data to form a data segment set. Generate new data segments based on the running data and data segment length in the data segment set. S3. Set the data truncation length according to the data segment length of historical running tasks. Truncate the latest running data at the running node according to the truncation length, extract the data segment set corresponding to the running node, and perform similarity analysis between it and the latest running data. Select the most similar target data segment, and perform fault coefficient analysis on the target data segment in combination with the resource configuration of the running node to obtain the fault coefficient corresponding to the running node. S4. Set the fault optimization threshold according to the remaining resource configuration of each running node, compare the fault coefficient of each running node with the corresponding optimization threshold for risk, and determine that the running node is a node to be protected against fault tolerance when the fault coefficient is greater than the optimization threshold. S5. For the node to be protected against faults, select the remaining resource configuration of other running nodes based on the corresponding target data segment, and coordinate the node to be protected against faults with the selected other running nodes to perform the same running tasks until the fault coefficient of the node to be protected against faults is less than the optimization threshold.

2. The task scheduling optimization and fault-tolerant processing method for multi-node collaborative control according to claim 1, characterized in that: In step S1, a multi-node operating system is connected, and the online status of all nodes in the domain is detected through the multi-node system. The operating nodes that are in the online operating state are selected and each operating node is given a unique identifier. Then, the unique identifiers of the operating nodes are summarized to obtain a list of operating nodes. The current configuration values ​​and remaining available values ​​of computing power, storage, network bandwidth, and hardware load of each running node are statistically analyzed and quantified to obtain the resource configuration of the running node; Retrieve task execution logs and data collection logs from each running node according to the time dimension, filter out completed and incomplete historical running tasks and classify them according to task number, and extract the running data of each historical running task during the execution process.

3. The task scheduling optimization and fault-tolerant processing method for multi-node collaborative control according to claim 1, characterized in that: In S2, a threshold for the fluctuation range of the running data is set, and the time series running data of the historical running tasks is segmented using the sliding window method. When the fluctuation range of the running data in the sliding window exceeds the threshold, segmentation is performed until the division of all historical running task running data is completed, resulting in multiple data segments corresponding to the historical running tasks. A data feature matching algorithm is used to calculate the similarity value of running data between any two data segments. An association threshold is set. When the similarity value of two data segments exceeds the association threshold, the two data segments are associated. Conversely, when the similarity value of two data segments does not exceed the association threshold, the two data segments are not associated. Among them, the same data segment can establish a relationship with multiple data segments that meet the similarity standard; The data segment set is a dedicated data segment set that is independently divided according to the running node. The historical running task data segments of each running node are only included in its own dedicated data segment set.

4. The task scheduling optimization and fault-tolerant processing method for multi-node collaborative control according to claim 1, characterized in that: In step S2, at least two data segments are selected from the data segment set. Feature fusion and interpolation are performed on the running data of the selected data segments. At the same time, the selected data segments are spliced ​​together according to their lengths to generate new data segments that conform to the characteristics of the running data. The length of the new data segment is the average length of the selected data segments. After adding corresponding identification information, the generated new data segment is added to the original data segment set to complete the expansion of the data segment set.

5. The task scheduling optimization and fault-tolerant processing method for multi-node collaborative control according to claim 1, characterized in that: In step S3, the average length of all data segments of the historical running tasks corresponding to each running node is calculated, and the average length is used as the data truncation length of the latest running data of that running node. The latest running data, matching the data truncation length, is extracted from the real-time data acquisition terminals of each running node in chronological order from newest to oldest. Extract the dedicated data segment set corresponding to each running node, perform similarity analysis between the latest running data and each data segment in the data segment set, obtain the similarity value between the latest running data and each data segment, and then select the data segment with the largest similarity value as the target data segment.

6. The task scheduling optimization and fault-tolerant processing method for multi-node collaborative control according to claim 1, characterized in that: In step S3, the running data of the target data segment is subjected to multi-dimensional fitting analysis with the resource configuration of the running node. Fault weight coefficients for each dimension are set, and the initial fault coefficient of the running node is obtained by weighted calculation. Then, the initial fault coefficient is corrected according to the length of the target data segment. The corrected value is the final fault coefficient of the running node. The higher the final failure coefficient value, the higher the risk of node failure; The smaller the final failure coefficient, the lower the risk of node failure.

7. The task scheduling optimization and fault-tolerant processing method for multi-node collaborative control according to claim 1, characterized in that: In step S4, the remaining resource configuration is analyzed based on the resource configuration corresponding to the running node to obtain the remaining resource configuration of each running node; Resource allocation consists of used resource allocation and remaining resource allocation; Summarize the remaining resource configuration of each running node and set the fault optimization threshold; The more remaining resources are allocated, the lower the fault optimization threshold will be; The less remaining resources are allocated, the higher the fault optimization threshold. Compare the failure coefficient of each running node with the corresponding optimization threshold for risk assessment; When the fault coefficient is greater than the optimization threshold, the running node is determined to be a node to be protected against fault tolerance. If the fault coefficient is less than the optimization threshold, the running node is determined not to be a node to be protected against faults.

8. The task scheduling optimization and fault-tolerant processing method for multi-node collaborative control according to claim 1, characterized in that: In step S5, for the node to be protected against faults, the resource requirement parameters of the target data segment of the node to be protected against faults are calculated. Based on the resource requirement parameter, other running nodes in the multi-node system whose remaining resource configuration meets the resource requirement parameter are selected. Priority is given to the normally operating nodes whose remaining resource configuration value is higher than the benchmark parameter and whose node load rate is the lowest. Among them, other running nodes can be one or more.

9. The task scheduling optimization and fault-tolerant processing method for multi-node collaborative control according to claim 1, characterized in that: In step S5, all running tasks currently being executed by the fault-tolerant protection node are transmitted to other selected running nodes according to the task synchronization method, and the fault-tolerant protection node and other selected nodes are coordinated and scheduled to execute the same running tasks. During the collaborative scheduling process, the fault coefficient of the node to be protected is monitored in real time. When the fault coefficient is lower than the optimization threshold, the fault protection of other selected running nodes is stopped, and the independent task scheduling of each node is restored.

Images