Cluster state detection method, apparatus, device, and computer-readable storage medium

Abstract

A cluster state detection method, apparatus, device, and computer-readable storage medium are disclosed. The method includes: acquiring various multivariate time series data for each server node in a server cluster within each detection time period; and determining the matrix product T between adjacent matrices and the multivariate time series data corresponding to the i-th detection time period. i ; T i Input the trained GVAE and obtain the output T of the trained GVAE. i '; Determine T i and T i The difference characterization value is used; if the difference characterization value is greater than the threshold, it is determined that there is an abnormal server node in the server cluster. Through this embodiment, based on the multivariate time series of each server node in the server cluster, combined with the spatial correlation information between each server node and GVAE, the detection of whether there is an abnormal server node in the server cluster is realized.

Description

Technical Field

[0001] This application relates to the field of communication technology, specifically to a cluster status detection method, apparatus, device, and computer-readable storage medium. Background Technology

[0002] Gateway devices act as middleware between clients and servers, enabling network communication between clients and a specific server node in a server cluster. In existing technologies, load balancing strategies are typically used to determine which server node the gateway connects to. However, this approach can lead to the gateway connecting to an abnormal server node if an abnormal server node exists within the cluster.

[0003] Therefore, there is an urgent need for a technical solution to detect whether there are abnormal server nodes in a server cluster, so as to guide subsequent access actions based on the detection results. Summary of the Invention

[0004] To detect the presence of abnormal server nodes in a server cluster, this application provides a cluster status detection method, apparatus, device, and computer-readable storage medium.

[0005] In a first aspect, embodiments of this application provide a cluster state detection method, the cluster state detection method comprising:

[0006] Obtain multivariate time series data for each server node in the server cluster during each detection period. Each multivariate time series data includes performance measurement data of the node at multiple times.

[0007] Determine the matrix product T between the adjacent matrices and the multivariate time series group corresponding to the i-th detection time period. i The multivariate time series group corresponding to the i-th detection time period includes the multivariate time series of each server node in the i-th time period. The adjacency matrix is ​​used to characterize the spatial correlation information between each server node. i includes integers from 1 to N, and N is the number of detection time periods.

[0008] T i Input the trained Graph Variational Autoencoder (GVAE) and obtain the trained GVAE for T. i Output T i ', where the trained GVAE is processed by the encoder for T i After encoding to obtain feature information, the feature information is then decoded by a decoder to obtain the output T. i ';

[0009] Determine T i and T i 'Difference characterization value;

[0010] If the difference value is greater than the threshold, it is determined that there is an abnormal server node in the server cluster.

[0011] In conjunction with the first aspect, in one embodiment, prior to the step of obtaining the sets of multivariate time series of each server node in the server cluster within each detection time period, the method further includes:

[0012] Obtain standard multivariate time series data for each server node in the server cluster when they are in a normal state over a period of time.

[0013] Determine the characteristic distribution of each standard multivariate time series;

[0014] The JS divergence between each pair of server nodes is determined based on the characteristic distribution of each standard multivariate time series.

[0015] The adjacency matrix is ​​obtained based on the JS divergence of each pair of server nodes.

[0016] In conjunction with the first aspect, in one embodiment, prior to the step of obtaining the sets of multivariate time series of each server node in the server cluster within each detection time period, the method further includes:

[0017] Obtain standard multivariate time series data for each server node in the server cluster when they are in a normal state, within each time period.

[0018] Determine the matrix product t of adjacent matrices and the standard multivariate time series group corresponding to the i-th time period. i The standard multivariate time series group corresponding to the i-th time period includes each group of standard multivariate time series in the i-th time period when each server node is in a normal state. The adjacency matrix is ​​used to characterize the spatial correlation information between each server node. i includes integers from 1 to K, and K is the number of time periods.

[0019] t i Input a graph variational autoencoder (GVAE) and obtain the GVAE for t. i Output t i ', where GVAE uses an encoder to measure t i After encoding to obtain feature information, the feature information is then decoded by a decoder to obtain the output t. i ';

[0020] Determine t i and t i 'Difference characterization value;

[0021] Whether the stopping condition is met is determined based on the difference characterization value;

[0022] If the stopping condition is not met, the parameters of GVAE are adjusted, and the matrix product t of the standard multivariate time series group corresponding to the i-th time period is returned. i Steps;

[0023] If the stopping condition is met, the current GVAE is taken as the GVAE that has been trained.

[0024] In conjunction with the first aspect, in one implementation, the stopping condition is t. i and t i The difference representation value reaches a convergent state or t i and t i The difference representation value is less than the threshold.

[0025] In conjunction with the first aspect, in one implementation, the determination of T i and T i The steps for representing the difference values ​​include:

[0026] T i and T i Input the reconstruction error calculation formula, and use the calculated reconstruction error as T. i and T i The difference characterization value, where the reconstruction error is calculated using the following formula:

[0027]

[0028] MSE represents the reconstruction error.

[0029] In conjunction with the first aspect, in one implementation, the determination of T i and T i The steps for representing the difference values ​​include:

[0030] T i and T i 'Input the formula for calculating the restoration error, and use the calculated restoration error as T' i and T i The difference characterization value, where the reduction error is calculated using the following formula:

[0031]

[0032] MAE represents the restoration error.

[0033] In conjunction with the first aspect, in one implementation, after the step of determining that an abnormal server node exists in the server cluster if the difference characterization value is greater than a threshold, the method further includes:

[0034] Based on the inverse matrix of the adjacent matrix and T i'Obtain the output multivariate time series corresponding to each set of multivariate time series for each server node in the i-th detection time period;'

[0035] For each server node, based on its N sets of multivariate time series within N detection time periods and the N sets of output multivariate time series corresponding to the N sets of multivariate time series, the difference characterization value corresponding to the server node is obtained.

[0036] If the difference representation value corresponding to a server node is greater than the threshold, the server node is determined to be an abnormal server node.

[0037] Secondly, embodiments of this application provide a cluster status detection device, the cluster status detection device comprising:

[0038] The acquisition module is used to acquire various multivariate time series of each server node in the server cluster within each detection time period. Each multivariate time series includes performance measurement data of the node at multiple times.

[0039] The first determining module is used to determine the matrix product T between the adjacent matrices and the multivariate time series group corresponding to the i-th detection time period. i The multivariate time series group corresponding to the i-th detection time period includes the multivariate time series of each server node in the i-th time period. The adjacency matrix is ​​used to characterize the spatial correlation information between each server node. i includes integers from 1 to N, and N is the number of detection time periods.

[0040] Processing module, used to process T i Input the trained Graph Variational Autoencoder (GVAE) and obtain the trained GVAE for T. i Output T i ', where the trained GVAE is processed by the encoder for T i After encoding to obtain feature information, the feature information is then decoded by a decoder to obtain the output T. i ';

[0041] The second determining module is used to determine T. i and T i 'Difference characterization value;

[0042] The judgment module is used to determine that there is an abnormal server node in the server cluster if the difference characterization value is greater than a threshold.

[0043] Thirdly, embodiments of this application provide a cluster state detection device, which includes a processor, a memory, and a cluster state detection program stored in the memory and executable by the processor, wherein when the cluster state detection program is executed by the processor, it implements the steps of the cluster state detection method as described in the first aspect.

[0044] Fourthly, embodiments of this application provide a computer-readable storage medium storing a cluster state detection program, wherein when the cluster state detection program is executed by a processor, it implements the steps of the cluster state detection method as described in the first aspect.

[0045] The beneficial effects of the technical solutions provided in this application include:

[0046] In this embodiment, the multivariate time series of each server node in the server cluster within each detection time period are obtained. Each multivariate time series includes performance measurement data of the node at multiple times. The matrix product T between the adjacent matrix and the multivariate time series group corresponding to the i-th detection time period is determined. i The multivariate time series group corresponding to the i-th detection time period includes the multivariate time series groups of each server node within the i-th time period. The adjacency matrix is ​​used to represent the spatial correlation information between each server node, where i includes integers from 1 to N, and N is the number of detection time periods. i Input the trained Graph Variational Autoencoder (GVAE) and obtain the trained GVAE for T. i Output T i ', where the trained GVAE is processed by the encoder for T i After encoding to obtain feature information, the feature information is then decoded by a decoder to obtain the output T. i '; Determine T i and T i The difference characterization value is used; if the difference characterization value is greater than the threshold, it is determined that there is an abnormal server node in the server cluster. Through this embodiment, based on the multivariate time series of each server node in the server cluster, combined with the spatial correlation information between each server node and GVAE, the detection of whether there is an abnormal server node in the server cluster is realized. Attached Figure Description

[0047] Figure 1 This is a flowchart illustrating an embodiment of the cluster state detection method of this application;

[0048] Figure 2 A diagram illustrating data processing for GVAE;

[0049] Figure 3 A schematic diagram for determining the characteristic distribution of a standard multivariate time series;

[0050] Figure 4 This is a functional module diagram of an embodiment of the cluster status detection device of this application;

[0051] Figure 5 This is a schematic diagram of the hardware structure of the cluster status detection device involved in the embodiment of this application. Detailed Implementation

[0052] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application.

[0053] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0054] Firstly, embodiments of this application provide a cluster status detection method.

[0055] In one embodiment, reference is made to Figure 1 , Figure 1 This is a flowchart illustrating an embodiment of the cluster state detection method of this application. Figure 1 As shown, the cluster status detection methods include:

[0056] Step S10: Obtain each set of multivariate time series for each server node in the server cluster during each detection time period. Each set of multivariate time series includes performance measurement data of the node at multiple times.

[0057] In this embodiment, the length of the detection time period is set according to actual needs, for example, one day or two days. The number of detection time periods is set according to actual needs, for example, 5 or 10, denoted as N here.

[0058] Performance measurement data for each server node is collected at multiple points within a given time period. The number of points is set according to actual needs, such as 50 or 100, and is denoted as P here.

[0059] Let Xai be the multivariate time series of server node a during the i-th detection time period, where Xai = (Xai1, Xai2, Xai3, ..., Xai...).p ). Among them, Xai p This represents the performance measurement data of node a at time p within the i-th detection time period, and so on.

[0060] For ease of explanation, assume the server cluster contains server node 1, server node 2, and server node 3, with N equal to 2 and p equal to 5. The retrieved data includes:

[0061] X11=(X111,X112,X113,X114,X115),

[0062] X214, X215), X31=(X311, X312, X313,

[0063] X123, X124, X125), X22 = (X221, X222, X223, X224, X225) and X32 = (X321,

[0064] X322, X323, X324, X325).

[0065] The performance metrics include one or more of the following: CPU utilization, memory utilization, I / O throughput, bandwidth, maximum latency, minimum latency, packet loss rate, data transfer rate, protocol type, service type, and topology data of the server nodes. It's easy to understand that the performance metrics are preprocessed data from the raw values; this preprocessing can be normalization.

[0066] Step S20: Determine the matrix product T between the adjacent matrices and the multivariate time series group corresponding to the i-th detection time period. i The multivariate time series group corresponding to the i-th detection time period includes the multivariate time series of each server node in the i-th time period. The adjacency matrix is ​​used to characterize the spatial correlation information between each server node. i includes integers from 1 to N, and N is the number of detection time periods.

[0067] In this embodiment, referring to the example above, the multivariate time series group corresponding to the first detection time period is [X11, X21, X31], and the multivariate time series group corresponding to the second detection time period is [X12, X22, X32]. Then, the matrix product T1 between adjacent matrix A and [X11, X21, X31], and the matrix product T2 between adjacent matrix A and [X12, X22, X32] can be determined.

[0068] Step S30, T iInput the trained Graph Variational Autoencoder (GVAE) and obtain the trained GVAE for T. i Output T i ', where the trained GVAE is processed by the encoder for T i After encoding to obtain feature information, the feature information is then decoded by a decoder to obtain the output T. i ';

[0069] In this embodiment, refer to Figure 2 , Figure 2 A diagram illustrating data processing for GVAE. (Example) Figure 2 As shown, the encoder (convolutional neural network) of GVAE encodes the input, and the encoded feature information is sent to the decoder (another convolutional neural network) of GVAE so that the decoder can decode the feature information to obtain the output.

[0070] Among them, the graph variational autoencoder (GVAE), also known as the variational graph autoencoder (VGAE), is a product of combining graph autoencoder (GAE) and variational autoencoder (VAE).

[0071] Graph variational autoencoders use graph neural networks to learn spatial information about data through graph matrices, which is highly correlated with the relationships between cluster nodes, compared to traditional neural networks.

[0072] Based on the above examples, when the input is T1, the output is T1'; when the input is T2, the output is T2'.

[0073] Step S40, determine T i and T i 'Difference characterization value;

[0074] In this embodiment, based on the above example, the difference characterization value needs to be calculated based on T1, T1', T2, and T2'.

[0075] Further, in one embodiment, step S40 includes:

[0076] T i and T i Input the reconstruction error calculation formula, and use the calculated reconstruction error as T. i and T i The difference characterization value, where the reconstruction error is calculated using the following formula:

[0077]

[0078] MSE represents the reconstruction error.

[0079] Further, in one embodiment, step S40 includes:

[0080] T i and T i 'Input the formula for calculating the restoration error, and use the calculated restoration error as T' i and T i The difference characterization value, where the reduction error is calculated using the following formula:

[0081]

[0082] MAE represents the restoration error.

[0083] In this embodiment, either the reconstruction error or the restoration error can be used as T. i and T i The difference representation value can also be a value determined by averaging or weighting the reconstruction error and the restoration error, etc., as T. i and T i 'The difference characterization value'.

[0084] Step S50: If the difference characterization value is greater than the threshold, it is determined that there is an abnormal server node in the server cluster.

[0085] In this embodiment, the threshold is set according to actual needs. If the difference indicator value is greater than the threshold, it indicates that the input and output differ significantly, thus determining that there is an abnormal server node in the server cluster.

[0086] In this embodiment, the multivariate time series of each server node in the server cluster within each detection time period are obtained. Each multivariate time series includes performance measurement data of the node at multiple times. The matrix product T between the adjacent matrix and the multivariate time series group corresponding to the i-th detection time period is determined. i The multivariate time series group corresponding to the i-th detection time period includes the multivariate time series groups of each server node within the i-th time period. The adjacency matrix is ​​used to represent the spatial correlation information between each server node, where i includes integers from 1 to N, and N is the number of detection time periods. i Input the trained Graph Variational Autoencoder (GVAE) and obtain the trained GVAE for T. i Output T i ', where the trained GVAE is processed by the encoder for T i After encoding to obtain feature information, the feature information is then decoded by a decoder to obtain the output T. i '; Determine T i and T iThe difference characterization value is used; if the difference characterization value is greater than the threshold, it is determined that there is an abnormal server node in the server cluster. Through this embodiment, based on the multivariate time series of each server node in the server cluster, combined with the spatial correlation information between each server node and GVAE, the detection of whether there is an abnormal server node in the server cluster is realized.

[0087] Furthermore, in one embodiment, before step S10, the method further includes:

[0088] Obtain the standard multivariate time series of each server node in the server cluster when they are in a normal state over a period of time; determine the feature distribution of each standard multivariate time series; determine the JS divergence between each pair of server nodes based on the feature distribution of each standard multivariate time series; and obtain the adjacency matrix based on the JS divergence between each pair of server nodes.

[0089] In this embodiment, it is assumed that the server cluster contains server node 1, server node 2, and server node 3, and the value of P is 5. The obtained standard multivariate time series include:

[0090] Y1 = (Y11, Y12, Y13, Y14, Y15), Y2 = (Y21, Y22, Y23, Y24, Y25), and Y3 = (Y31, Y32, Y33, Y34, Y35). Here, Y1 is a standard multivariate time series data for server node 1 in a normal state over a given time period, Y11 is the performance measurement data of server node 1 in a normal state at the first moment within a given time period, and so on.

[0091] The following explanation uses the determination of the characteristic distribution of the standard multivariate time series Y1 as an example:

[0092] Reference Figure 3 , Figure 3 A schematic diagram for determining the characteristic distribution of a standard multivariate time series. (e.g.) Figure 3 As shown, a standard multivariate time series (Y11, Y12, ..., Y1) is used. t As the input to the Adjacent Matrix Variational Autoencoder (AM-VAE), the latent variables z are learned through a set of GRUs (Gated Recurrent Units) in the AM-VAE. This allows the acquisition of the same mean μ and variance σ in the time domain, ultimately obtaining the latent variable z in the time dimension. Throughout the training process, the input is the original standard multivariate time series, which is encoded using GRUs to obtain latent variables Z1 to Zt. Then, another set of GRUs decodes the data, outputting a new standard multivariate time series (Y11', Y12', ..., Y1...) after encoding and decoding. tThe loss calculation involves determining the similarity between the input and output standard multivariate time series, adjusting the parameters of the AM-VAE based on the loss, and then returning to training until the loss is less than a preset value or the loss converges. The mean μ and variance σ at this point are then used as the feature distribution of the standard multivariate time series. The goal of training is to reduce the difference between the output and input, thereby ensuring that the latent variables can truly represent the features in the standard multivariate time series.

[0093] It should be noted that different standard multivariate time series use different AM-VAEs to determine their characteristic distributions. For example, the characteristic distribution of a standard multivariate time series Y1 is determined by one AM-VAE, the characteristic distribution of a standard multivariate time series Y2 is determined by another AM-VAE, and so on.

[0094] Thus, we can obtain the characteristic distribution of Y1, denoted as P1~(μ1, σ1), the characteristic distribution of Y2, denoted as P2~(μ2, σ2), and the characteristic distribution of Y3, denoted as P3~(μ3, σ3).

[0095] Based on this, the JS divergence JS(P1, P2) corresponding to server node 1 and server node 2 can be obtained:

[0096]

[0097] Similarly, we can obtain JS(P1, P3) and JS(P2, P3).

[0098] Furthermore, the elements at each position in the adjacent matrix A are pre-defined based on the JS divergence of the corresponding two server nodes, and it is stipulated that the element value is 1 when the JS divergence is greater than a preset value, and 0 otherwise. Based on this, the adjacent matrix can be obtained after obtaining the JS divergence of each pair of server nodes.

[0099] In this embodiment, AM-VAE is used to learn the adjacency matrix for graph variational autoencoders, and then combined with graph variational autoencoders (GVAE) to fully explore the temporal and spatial correlation information of the data.

[0100] Furthermore, in one embodiment, before step S10, the method further includes:

[0101] Step S01: Obtain the standard multivariate time series data for each server node in the server cluster during different time periods when each server node is in a normal state.

[0102] This embodiment is basically the same as step S10, except that the standard multivariate time series obtained in this embodiment is limited to server nodes in a normal state.

[0103] Step S02: Determine the matrix product t of the adjacent matrices and the standard multivariate time series group corresponding to the i-th time period. i The standard multivariate time series group corresponding to the i-th time period includes each group of standard multivariate time series in the i-th time period when each server node is in a normal state. The adjacency matrix is ​​used to characterize the spatial correlation information between each server node. i includes integers from 1 to K, and K is the number of time periods.

[0104] Step S03, t i Input a graph variational autoencoder (GVAE) and obtain the GVAE for t. i Output t i ', where GVAE uses an encoder to measure t i After encoding to obtain feature information, the feature information is then decoded by a decoder to obtain the output t. i ';

[0105] Step S04, determine t i and t i 'Difference characterization value;

[0106] The specific embodiments of steps S02 to S04 can be referred to the embodiments of steps S20 to S40 above, and will not be repeated here.

[0107] Step S05: Detect whether the stopping condition is met based on the difference characterization value;

[0108] In this embodiment, the stopping condition is t. i and t i The difference representation value reaches a convergent state or t i and t i The difference representation value is less than the threshold.

[0109] Step S06: If the stopping condition is not met, adjust the parameters of GVAE and return the matrix product t of the standard multivariate time series group corresponding to the i-th time period, which determines the adjacent matrix. i Steps;

[0110] Step S07: If the stopping condition is met, then the current GVAE is taken as the GVAE that has been trained.

[0111] In this embodiment, the trained GVAE can produce an output that is approximately identical to the input, and the training is based on standard multivariate time series data from each server node in the server cluster during various time periods when all server nodes are in normal operation. In this case, if subsequent multivariate time series data from each server node in the server cluster during various detection time periods are obtained and used as input to the trained GVAE, if the difference between the output and input of the trained GVAE is greater than a threshold, it indicates the presence of abnormal server nodes in the server cluster; if the difference is not greater than the threshold, it indicates that all server nodes in the server cluster are in normal operation.

[0112] Furthermore, in one embodiment, after step S50, the method further includes:

[0113] Step S60, based on the inverse matrix of the adjacent matrix and T i 'Obtain the output multivariate time series corresponding to each set of multivariate time series for each server node in the i-th detection time period;'

[0114] In this embodiment, it is assumed that there are server node 1, server node 2, and server node 3 in the server cluster, and N is 2. The multivariate time series of server node a in the i-th detection time period is denoted as Xai, where Xai = (Xai1, Xai2, Xai3, ..., Xai...). p ). Among them, Xai p This represents the performance measurement data of node a at time p within the i-th detection time period, and so on.

[0115] The multivariate time series group corresponding to the first detection time period is [X11, X21, X31], and the multivariate time series group corresponding to the second detection time period is [X12, X22, X32].

[0116] The matrix product T1 of adjacent matrices A and [X11, X21, X31] corresponds to the output T1′. Then, the output multivariate time series for each server node in the first time period can be obtained based on the following formula:

[0117] [X11', X21', X31'] = T1'A -

[0118] Among them, A - It is the inverse of the adjacent matrix A.

[0119] Wherein, X11' is the output multivariate time series corresponding to the multivariate time series X11 of server node 1 in the first time period, X21' is the output multivariate time series corresponding to the multivariate time series X21 of server node 2 in the first time period, and X31' is the output multivariate time series corresponding to the multivariate time series X31 of server node 3 in the first time period.

[0120] Similarly, we get [X12', X22', X32'].

[0121] Step S70: For each server node, based on its N sets of multivariate time series in N detection time periods and the N sets of output multivariate time series corresponding to the N sets of multivariate time series, obtain the difference characterization value corresponding to the server node.

[0122] Step S80: If the difference representation value corresponding to the server node is greater than the threshold, then the server node is determined to be an abnormal server node.

[0123] In this embodiment, taking server node 1 as an example, the difference representation value corresponding to server node 1 is obtained based on X11, X11', X12, and X12', and the calculation method is as follows:

[0124]

[0125] or,

[0126] The difference representation value can be determined using either of the two calculation methods mentioned above. The threshold is set according to the selected calculation method.

[0127] If the difference representation value corresponding to server node 1 is greater than the threshold, it means that the N multivariate time series of server node 1 within the N detection time period are significantly different from the N output multivariate time series obtained by the trained GVAE, indicating that server node 1 is an abnormal server node.

[0128] Similarly, the difference representation values ​​for other server nodes are determined, which will not be elaborated here.

[0129] Secondly, embodiments of this application also provide a cluster status detection device.

[0130] In one embodiment, reference is made to Figure 4 , Figure 4 This is a functional module diagram of an embodiment of the cluster status detection device of this application. Figure 4 As shown, the cluster status detection device includes:

[0131] The acquisition module 10 is used to acquire each set of multivariate time series of each server node in the server cluster during each detection time period. Each set of multivariate time series includes the performance measurement data of the node at multiple times.

[0132] The first determining module 20 is used to determine the matrix product T between the adjacent matrices and the multivariate time series group corresponding to the i-th detection time period. i The multivariate time series group corresponding to the i-th detection time period includes the multivariate time series of each server node in the i-th time period. The adjacency matrix is ​​used to characterize the spatial correlation information between each server node. i includes integers from 1 to N, and N is the number of detection time periods.

[0133] Processing module 30 is used to process T i Input the trained Graph Variational Autoencoder (GVAE) and obtain the trained GVAE for T. i Output T i ', where the trained GVAE is processed by the encoder for T i After encoding to obtain feature information, the feature information is then decoded by a decoder to obtain the output T. i ';

[0134] The second determining module 40 is used to determine T. i and T i 'Difference characterization value;

[0135] The judgment module 50 is used to determine that there is an abnormal server node in the server cluster if the difference characterization value is greater than the threshold.

[0136] Furthermore, in one embodiment, the cluster state detection device further includes a learning module, used for:

[0137] Obtain standard multivariate time series data for each server node in the server cluster when they are in a normal state over a period of time.

[0138] Determine the characteristic distribution of each standard multivariate time series;

[0139] The JS divergence between each pair of server nodes is determined based on the characteristic distribution of each standard multivariate time series.

[0140] The adjacency matrix is ​​obtained based on the JS divergence of each pair of server nodes.

[0141] Furthermore, in one embodiment, the cluster state detection device further includes a training module, used for:

[0142] Obtain standard multivariate time series data for each server node in the server cluster when they are in a normal state, within each time period.

[0143] Determine the matrix product t of adjacent matrices and the standard multivariate time series group corresponding to the i-th time period. i The standard multivariate time series group corresponding to the i-th time period includes each group of standard multivariate time series in the i-th time period when each server node is in a normal state. The adjacency matrix is ​​used to characterize the spatial correlation information between each server node. i includes integers from 1 to K, and K is the number of time periods.

[0144] t i Input a graph variational autoencoder (GVAE) and obtain the GVAE for t. i Output t i ', where GVAE uses an encoder to measure t i After encoding to obtain feature information, the feature information is then decoded by a decoder to obtain the output t. i ';

[0145] Determine t i and t i 'Difference characterization value;

[0146] Whether the stopping condition is met is determined based on the difference characterization value;

[0147] If the stopping condition is not met, the parameters of GVAE are adjusted, and the matrix product t of the standard multivariate time series group corresponding to the i-th time period is returned. i Steps;

[0148] If the stopping condition is met, the current GVAE is taken as the GVAE that has been trained.

[0149] Furthermore, in one embodiment, the stopping condition is t. i and t i The difference representation value reaches a convergent state or t i and t i The difference representation value is less than the threshold.

[0150] Furthermore, in one embodiment, the second determining module 40 is configured to:

[0151] T i and T i 'Input the reconstruction error calculation formula, and use the calculated reconstruction error as T' i and T i The difference characterization value, where the reconstruction error is calculated using the following formula:

[0152]

[0153] MSE represents the reconstruction error.

[0154] Furthermore, in one embodiment, the second determining module 40 is configured to:

[0155] T i and T i 'Input the formula for calculating the restoration error, and use the calculated restoration error as T' i and T i The difference characterization value, where the reduction error is calculated using the following formula:

[0156]

[0157] MAE represents the restoration error.

[0158] Furthermore, in one embodiment, the determination module 50 is also used for:

[0159] Based on the inverse matrix of the adjacent matrix and T i 'Obtain the output multivariate time series corresponding to each set of multivariate time series for each server node in the i-th detection time period;'

[0160] For each server node, based on its N sets of multivariate time series within N detection time periods and the N sets of output multivariate time series corresponding to the N sets of multivariate time series, the difference characterization value corresponding to the server node is obtained.

[0161] If the difference representation value corresponding to a server node is greater than the threshold, the server node is determined to be an abnormal server node.

[0162] The functions of each module in the above-mentioned cluster state detection device correspond to the steps in the above-mentioned cluster state detection method embodiment, and their functions and implementation processes will not be described in detail here.

[0163] Thirdly, embodiments of this application provide a cluster status detection device, which can be a personal computer (PC), laptop computer, server, or other device with data processing capabilities.

[0164] Reference Figure 5 , Figure 5 This is a schematic diagram of the hardware structure of the cluster status detection device involved in the embodiments of this application. In the embodiments of this application, the cluster status detection device may include a processor, a memory, a communication interface, and a communication bus.

[0165] The communication bus can be of any type and is used to interconnect the processor, memory, and communication interface.

[0166] The communication interface includes input / output (I / O) interfaces, physical interfaces, and logical interfaces used for interconnecting devices within the cluster status monitoring device, as well as interfaces used for interconnecting the cluster status monitoring device with other devices (such as other computing devices or user equipment). Physical interfaces can be Ethernet interfaces, fiber optic interfaces, ATM interfaces, etc.; user equipment can be displays, keyboards, etc.

[0167] Memory can be various types of storage media, such as random access memory (RAM), read-only memory (ROM), non-volatile RAM (NVRAM), flash memory, optical storage, hard disk, programmable ROM (PROM), erasable PROM (EPROM), electrically erasable PROM (EEPROM), etc.

[0168] The processor can be a general-purpose processor, which can call the cluster state detection program stored in memory and execute the cluster state detection method provided in the embodiments of this application. For example, the general-purpose processor can be a central processing unit (CPU). The method executed when the cluster state detection program is called can be referred to in the various embodiments of the cluster state detection method of this application, and will not be repeated here.

[0169] Those skilled in the art will understand that Figure 5 The hardware structure shown does not constitute a limitation of this application and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0170] Fourthly, embodiments of this application also provide a computer-readable storage medium.

[0171] The present application has a cluster state detection program stored on a computer-readable storage medium, wherein when the cluster state detection program is executed by a processor, it implements the steps of the cluster state detection method described above.

[0172] The method implemented when the cluster state detection program is executed can be referred to in various embodiments of the cluster state detection method of this application, and will not be repeated here.

[0173] It should be noted that the sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0174] The terms "comprising" and "having," and any variations thereof, in the specification, claims, and accompanying drawings of this application are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to such process, method, product, or apparatus. The terms "first," "second," and "third," etc., are used to distinguish different objects, etc., and do not indicate a sequence, nor do they limit "first," "second," and "third" to different types.

[0175] In the description of the embodiments of this application, terms such as "exemplary," "for example," or "for instance" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as "exemplary," "for example," or "for instance" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplary," "for example," or "for instance" is intended to present the relevant concepts in a concrete manner.

[0176] In the description of the embodiments of this application, unless otherwise stated, " / " means "or". For example, A / B can mean A or B. The "and / or" in the text is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist simultaneously, and B exists alone. In addition, in the description of the embodiments of this application, "multiple" means two or more.

[0177] In some processes described in the embodiments of this application, multiple operations or steps are included in a specific order. However, it should be understood that these operations or steps may not be executed in the order they appear in the embodiments of this application, or they may be executed in parallel. The sequence number of the operation is only used to distinguish different operations, and the sequence number itself does not represent any execution order. In addition, these processes may include more or fewer operations, and these operations or steps may be executed sequentially or in parallel, and these operations or steps may be combined.

[0178] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device to execute the methods described in the various embodiments of this application.

[0179] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A cluster state detection method, characterized in that, The cluster status detection method includes: Obtain standard multivariate time series data for each server node in the server cluster when they are in a normal state over a period of time. Determine the characteristic distribution of each standard multivariate time series; The JS divergence between each pair of server nodes is determined based on the characteristic distribution of each standard multivariate time series. The adjacency matrix is ​​obtained based on the JS divergence of each pair of server nodes; Obtain multivariate time series data for each server node in the server cluster during each detection period. Each multivariate time series data includes performance measurement data of the node at multiple times. Determine the matrix product of adjacent matrices and the multivariate time series group corresponding to the i-th detection time period. The multivariate time series group corresponding to the i-th detection time period includes the multivariate time series of each server node in the i-th time period. The adjacency matrix is ​​used to characterize the spatial correlation information between each server node. i includes integers from 1 to N, and N is the number of detection time periods. Will Input the trained Graph Variational Autoencoder (GVAE) and obtain the trained GVAE targeting... Output ', where the trained GVAE is processed by the encoder. After encoding to obtain feature information, the feature information is then decoded by a decoder to obtain the output. '; Sure and 'Difference characterization value; If the difference value is greater than the threshold, it is determined that there is an abnormal server node in the server cluster.

2. The cluster state detection method as described in claim 1, characterized in that, Before the step of obtaining the multivariate time series of each server node in the server cluster within each detection time period, the method further includes: Obtain standard multivariate time series data for each server node in the server cluster when they are in a normal state, within each time period. Determine the matrix product of adjacent matrices and the standard multivariate time series group corresponding to the i-th time period. The standard multivariate time series group corresponding to the i-th time period includes each group of standard multivariate time series in the i-th time period when each server node is in a normal state. The adjacency matrix is ​​used to characterize the spatial correlation information between each server node. i includes integers from 1 to K, and K is the number of time periods. Will Input a graph variational autoencoder (GVAE) and obtain the GVAE's target... Output ', where GVAE uses an encoder to... After encoding to obtain feature information, the feature information is then decoded by a decoder to obtain the output. '; Sure and 'Difference characterization value; Whether the stopping condition is met is determined based on the difference characterization value; If the stopping condition is not met, the parameters of GVAE are adjusted, and the matrix product of the adjacent matrices and the standard multivariate time series group corresponding to the i-th time period is returned. Steps; If the stopping condition is met, the current GVAE is taken as the GVAE that has been trained.

3. The cluster state detection method as described in claim 2, characterized in that, The stopping condition is and The difference representation value has reached a convergence state or and The difference representation value is less than the threshold.

4. The cluster state detection method as described in claim 1, characterized in that, The determination and The steps for representing the difference values ​​include: Will and 'Input the reconstruction error calculation formula, and use the calculated reconstruction error as...' and The difference characterization value, where the reconstruction error is calculated using the following formula: MSE represents the reconstruction error.

5. The cluster state detection method as described in claim 1, characterized in that, The determination and The steps for representing the difference values ​​include: Will and 'Input the formula for calculating the restoration error, and use the calculated restoration error as...' and The difference characterization value, where the reduction error is calculated using the following formula: MAE represents the restoration error.

6. The cluster state detection method according to any one of claims 1 to 5, characterized in that, Following the step of determining that an abnormal server node exists in the server cluster if the difference characterization value is greater than a threshold, the method further includes: The inverse matrix based on the adjacent matrix and 'Obtain the output multivariate time series corresponding to each set of multivariate time series for each server node in the i-th detection time period;' For each server node, based on its N sets of multivariate time series within N detection time periods and the N sets of output multivariate time series corresponding to the N sets of multivariate time series, the difference characterization value corresponding to the server node is obtained. If the difference representation value corresponding to a server node is greater than the threshold, the server node is determined to be an abnormal server node.

7. A cluster status detection device, characterized in that, The cluster status detection device includes: The learning module is used to obtain various standard multivariate time series of each server node in the server cluster under normal conditions over a period of time; determine the feature distribution of each standard multivariate time series; determine the JS divergence between each pair of server nodes based on the feature distribution of each standard multivariate time series; and obtain the adjacency matrix based on the JS divergence between each pair of server nodes. The acquisition module is used to acquire various multivariate time series of each server node in the server cluster within each detection time period. Each multivariate time series includes performance measurement data of the node at multiple times. The first determining module is used to determine the matrix product of adjacent matrices and the multivariate time series group corresponding to the i-th detection time period. The multivariate time series group corresponding to the i-th detection time period includes the multivariate time series of each server node in the i-th time period. The adjacency matrix is ​​used to characterize the spatial correlation information between each server node. i includes integers from 1 to N, and N is the number of detection time periods. Processing module, used to process Input the trained Graph Variational Autoencoder (GVAE) and obtain the trained GVAE targeting... Output ', where the trained GVAE is processed by the encoder. After encoding to obtain feature information, the feature information is then decoded by a decoder to obtain the output. '; The second determining module is used to determine... and 'Difference characterization value; The judgment module is used to determine that there is an abnormal server node in the server cluster if the difference characterization value is greater than a threshold.

8. A cluster status detection device, characterized in that, The cluster state detection device includes a processor, a memory, and a cluster state detection program stored in the memory and executable by the processor, wherein when the cluster state detection program is executed by the processor, it implements the steps of the cluster state detection method as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a cluster state detection program, wherein when the cluster state detection program is executed by a processor, it implements the steps of the cluster state detection method as described in any one of claims 1 to 6.

Images