Storage system failure prediction method and device, electronic equipment and storage medium

Abstract

This application provides a storage system fault prediction method, apparatus, electronic device, and storage medium, relating to the field of storage system health prediction technology. It acquires storage system operating status data, business load data, and environmental data to construct a health model characterizing the device's health status and generates dynamic thresholds for anomaly detection based on business load data. The real-time collected data is analyzed based on this health model and dynamic thresholds to predict potential storage system faults. Furthermore, the root cause of the fault is determined through correlation analysis of multi-source data. Therefore, it can solve the problems in existing technologies caused by the use of fixed thresholds, a lack of in-depth analysis in single inspection methods, weak fault warning capabilities, and insufficient correlation of multi-source data, resulting in high false alarm rates, impact on storage service performance, high risk of data loss or service interruption, and difficulty in locating the root cause of faults.

Description

Technical Field

[0001] This application relates to the field of storage system health prediction technology, and in particular to a storage system fault prediction method and apparatus, electronic device and storage medium. Background Technology

[0002] Storage systems are core data management units in critical industries such as finance, healthcare, and telecommunications. With the exponential growth of data volume and the development of distributed and heterogeneous storage architectures, traditional health check technologies are no longer sufficient to meet high reliability requirements. Traditional solutions rely on hardware sensor data collection, basic performance monitoring, and human experience, with fixed threshold alarms as the core, covering detection aspects such as temperature, disk error rate, and RAID status. However, this is a passive maintenance model, failing to incorporate multi-dimensional feature modeling based on equipment model, age, and workload. Its core shortcomings include: high false alarm rates, leading to operational fatigue; rigid check scheduling, with full checks during peak periods severely impacting service performance; weak fault early warning capabilities, only able to respond after the fact, unable to predict early; and insufficient correlation of multi-source data, resulting in time-consuming root cause localization. These problems severely restrict the transformation of storage systems from "passive maintenance" to "proactive prevention." Summary of the Invention

[0003] This application provides a method and apparatus for predicting storage system failures, an electronic device, and a storage medium, to at least partially solve one of the technical problems in the related art.

[0004] This application provides a method for predicting storage system failures, including:

[0005] Acquire storage system operational status data, business load data, and environmental data; A health model representing the health status of equipment is constructed based on operational status data and environmental data, and dynamic thresholds for anomaly detection are generated by combining business load data. Based on the health model and dynamic thresholds, real-time collected data is analyzed to predict potential failures in the storage system and generate corresponding early warning information. Based on early warning information, the root cause of the failure is determined through correlation analysis of multi-source data.

[0006] This application also provides a storage system failure prediction device, comprising: The acquisition unit is used to acquire the storage system's operating status data, business load data, and environmental data. The first generation unit is used to build a health model representing the health status of the equipment based on the operating status data and environmental data, and to generate dynamic thresholds for anomaly detection in combination with business load data. The second generation unit is used to analyze the real-time collected data based on the health model and dynamic thresholds, predict potential failures of the storage system, and generate corresponding early warning information. The analysis unit is used to determine the root cause of a failure based on early warning information and through correlation analysis of multi-source data.

[0007] This application also provides an electronic device, including: a memory for storing a computer program; and a processor for executing the computer program to implement the steps of any of the above methods.

[0008] This application also provides a computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the steps of any of the above methods.

[0009] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of any of the above methods.

[0010] This application provides a storage system fault prediction method, apparatus, electronic device, and storage medium. By acquiring the storage system's operating status data, business load data, and environmental data, a health model characterizing the device's health status is constructed. Combined with business load data, a dynamic threshold for anomaly detection is generated. Based on this health model and dynamic threshold, real-time collected data is analyzed to predict potential storage system faults. Furthermore, the root cause of the fault is determined through multi-source data correlation analysis. Therefore, it can solve the problems in existing technologies caused by the use of fixed thresholds, single inspection methods lacking in-depth analysis, weak fault early warning capabilities, and insufficient multi-source data correlation, resulting in high false alarm rates, impact on storage service performance, high risk of data loss or service interruption, and difficulty in locating the root cause of faults. This achieves the technical effects of reducing false alarm rates, ensuring stable storage service performance, providing early warning of potential faults to reduce the risk of data loss and service interruption, and quickly and accurately locating the root cause of faults.

[0011] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0012] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein: Figure 1 A flowchart illustrating a storage system fault prediction method provided in an embodiment of this application; Figure 2 This is a schematic diagram of the structure of a storage system fault prediction device provided in an embodiment of this application. Detailed Implementation

[0013] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0014] To enable those skilled in the art to better understand the present application, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0015] The specific application environment architecture or specific hardware architecture on which the execution of the storage system fault prediction method depends is described here.

[0016] Embodiments of this application provide a method for predicting storage system failures. Figure 1 This is a flowchart illustrating a storage system fault prediction method provided in an embodiment of this application.

[0017] like Figure 1 As shown, the method includes the following steps: Step 101: Obtain the storage system's operating status data, business load data, and environmental data.

[0018] In the embodiments of this application, to achieve accurate assessment and efficient management of the health status of the storage system, it is first necessary to comprehensively collect basic data to support subsequent analysis. Specifically, this involves acquiring storage system operational status data (i.e., data reflecting the operational status of storage system hardware components and core functional parameters), business load data (i.e., data characterizing the operational pressure, data processing requirements, and business operation characteristics of the system), and environmental data (i.e., external environmental parameters affecting the stable operation of the storage system). This data collection process must ensure the comprehensiveness of data sources and the completeness of coverage dimensions, breaking through the limitations of traditional technologies that only collect single hardware or basic performance data. This provides multi-dimensional, high-quality basic data support for subsequent health model construction, dynamic threshold generation, and fault prediction and root cause analysis, ensuring the effectiveness of subsequent technical steps. As one implementation method, operational status data may include hardware status parameters such as disk error rate, RAID operational status, and device runtime, as well as core operational parameters such as data transmission rate and response latency. Business load data may include data read / write request volume, concurrent access volume, and peak business processing periods. Environmental data may include data related to data center temperature and humidity, power supply stability, and cooling system operational status.

[0019] By comprehensively acquiring multi-dimensional data on the operating status, business load, and environment of the storage system, it overcomes the shortcomings of traditional technologies in data collection, providing complete and comprehensive basic data support for subsequent health management-related analysis. This effectively avoids analytical biases caused by missing data or insufficient dimensions, laying a key foundation for improving the accuracy and reliability of storage system health management.

[0020] Step 102: Construct a health model representing the health status of the equipment based on the operating status data and environmental data, and generate dynamic thresholds for anomaly detection by combining business load data.

[0021] In the embodiments of this application, based on the acquisition of storage system operation status data, business load data, and environmental data, a health model (i.e., a quantitative assessment model comprehensively reflecting device hardware performance, operational stability, and environmental adaptability) is constructed using data modeling methods, leveraging the device's own operating characteristics reflected in the operation status data and the external influencing factors represented by the environmental data. Simultaneously, combining the system business carrying characteristics and operational pressure differences reflected in the business load data, dynamic thresholds for anomaly detection are generated through correlation analysis of multi-dimensional data. The construction of this health model focuses on the core influencing factors of device health status, achieving a systematic quantitative assessment of device health levels. The generation of dynamic thresholds overcomes the limitations of traditional fixed thresholds, enabling it to adapt to differences in different device models, service life, and business scenarios, ensuring the targeted and reasonable nature of anomaly detection. As one implementation method, the health model can be constructed using an algorithm that integrates hardware operating parameter trends and environmental parameter influence weights. The dynamic thresholds can be adaptively adjusted based on real-time changes in business load, historical data patterns, and device characteristic parameters.

[0022] By constructing a health model that fits the actual operation and environmental conditions of the equipment, and generating dynamic thresholds by combining business load data, the problem that traditional fixed thresholds cannot adapt to the differences in multiple scenarios is effectively solved, the false alarm rate of anomaly judgment is significantly reduced, and a scientific and reliable evaluation basis is provided for subsequent accurate fault prediction and anomaly analysis.

[0023] Step 103: Analyze the real-time collected data based on the health model and dynamic thresholds to predict potential failures of the storage system and generate corresponding early warning information.

[0024] In the embodiments of this application, based on the completed health status model and the generation of dynamic thresholds, multi-dimensional data of the storage system collected in real time is input into the health status model. Combined with the dynamic thresholds, a systematic analysis and judgment are performed. The core lies in using the model's ability to quantitatively assess the health status of the equipment and the scenario adaptability of the dynamic thresholds to identify abnormal characteristics hidden in data change trends. This allows for the proactive prediction of potential failures in the storage system and the generation of early warning information reflecting key information related to the failure. This process overcomes the limitations of traditional technologies that can only detect hardware anomalies that have already occurred, realizing a shift from "post-event response" to "proactive prevention," ensuring timely perception of potential risks. As one implementation method, the real-time collected data can correspond to the dimensions of the aforementioned operational status data, business load data, and environmental data. The early warning information can include key information such as the approximate type of potential failure, risk level, and expected impact range.

[0025] By combining health models and dynamic thresholds, real-time data is analyzed and potential faults are predicted. This effectively solves the problem of weak fault warning capabilities in traditional technologies. It can detect storage system risks in advance, significantly reduce the probability of data loss and service interruption, reserve sufficient processing time for operation and maintenance personnel, and ensure the high reliability of the storage system.

[0026] Step 104: Based on the early warning information, determine the root cause of the fault through correlation analysis of multi-source data.

[0027] In the embodiments of this application, after obtaining early warning information of potential faults, relying on the aforementioned collected multi-source data (i.e., storage system operating status data, business load data, and environmental data), multi-dimensional data correlation analysis is used to uncover the inherent correlations and abnormal causal logic between different data dimensions and the early warning information, thereby systematically tracing and accurately determining the root cause of the potential fault. This process overcomes the limitations of traditional technologies that lack multi-source data collaborative analysis. By integrating key factors such as the device's own operating characteristics, business load pressure, and external environmental influences, it avoids the positioning bias caused by single-data-dimensional analysis, ensuring the comprehensiveness and accuracy of root cause determination. As one implementation method, correlation analysis can match the abnormal parameter change trends in multi-source data based on the fault characteristics in the early warning information, and combine the dependencies and influence weights between data to pinpoint the core triggers of the fault.

[0028] By using correlation analysis of multi-source data to accurately determine the root cause of failure, the problem of difficulty and time-consuming root cause location in traditional technologies is effectively solved. This significantly shortens the operation and maintenance troubleshooting cycle, improves the pertinence and efficiency of fault handling, and provides key support for ensuring the rapid recovery and stable operation of the storage system.

[0029] This application provides a storage system fault prediction method. By acquiring the storage system's operating status data, business load data, and environmental data, a health model characterizing the device's health status is constructed. Combined with business load data, a dynamic threshold for anomaly detection is generated. Based on this health model and dynamic threshold, real-time collected data is analyzed to predict potential storage system faults. Furthermore, the root cause of the fault is determined through multi-source data correlation analysis. Therefore, this method can solve the problems in existing technologies, such as high false alarm rates, impact on storage service performance, high risk of data loss or service interruption, and difficulty in locating the root cause of faults due to the use of fixed thresholds, single inspection methods lacking in-depth analysis, weak fault early warning capabilities, and insufficient multi-source data correlation. This achieves the technical effects of reducing false alarm rates, ensuring stable storage service performance, providing early warning of potential faults to reduce the risk of data loss and service interruption, and quickly and accurately locating the root cause of faults.

[0030] In the embodiments involved in this application, there are various feasible specific implementation methods. To clearly and completely illustrate the technical solutions of this disclosure, the implementation methods listed below are merely exemplary and do not constitute a limitation on the scope of protection of this disclosure. That is, in addition to the implementation methods described below, other implementation methods that can be obtained by those skilled in the art based on the technical content disclosed in this disclosure through reasonable logical analysis, reasoning, or limited experimentation should also be covered within the scope of protection of this disclosure. The following specifically describes some exemplary implementation methods: As a specific implementation of this application, based on the basic solution, the acquisition of storage system operation status data, business load data and environmental data is further limited, including: synchronously collecting sensor data of storage devices through a distributed acquisition architecture, the sensor data including temperature, voltage, disk error rate and input / output latency; and extracting features from business access time-series data to obtain access pattern features.

[0031] Specifically, a distributed acquisition architecture is adopted to accurately acquire storage system operation status data, business load data, and environmental data. This architecture deploys distributed acquisition nodes on each node of the storage cluster. Each acquisition node establishes a real-time connection with the storage device sensors through a standardized communication protocol, synchronously collecting sensor data such as temperature, voltage, disk error rate, and input / output latency. During the acquisition process, a timestamp alignment mechanism ensures the temporal consistency of data across multiple nodes, while a data verification algorithm is used to filter out abnormal acquisition values ​​to ensure data accuracy. For business load data, business access time-series data is selected as the core analysis object. The time-series data is divided into segments by a sliding time window, and statistical analysis and pattern recognition algorithms are used to extract access pattern features, including key features such as access frequency per unit time, peak access period distribution, data read / write request ratio, and number of concurrent access users. During feature extraction, data smoothing technology is combined to reduce noise interference, ensuring that the access pattern features can truly reflect the characteristics of business operation load.

[0032] By synchronously collecting multi-dimensional sensor data through a distributed acquisition architecture, the real-time performance, completeness, and time-series consistency of data collection are ensured. Combined with targeted extraction of business access time-series data features, the core characteristics of business load are accurately captured, providing high-quality and highly relevant basic data for subsequent health model construction and dynamic threshold generation, effectively improving the accuracy and reliability of subsequent analysis.

[0033] As a specific implementation of this application, based on the basic scheme, it further defines a health status model that characterizes the health status of the equipment based on operational status data and environmental data, and generates a dynamic threshold for anomaly detection by combining business load data. This includes: performing time-series modeling on historical operational status data to obtain a specific health status scoring model for the equipment; performing periodic analysis on business load data, and generating a dynamic threshold function that changes over time by combining equipment aging factors.

[0034] Specifically, when constructing the health model, the historical operating status data of the storage system is first preprocessed to remove abnormal and noisy data and fill in missing values. Then, time-series modeling algorithms (such as Long Short-Term Memory Networks and Autoregressive Integral Moving Average Models) are used to perform deep training on the preprocessed historical data to explore the status change patterns of the equipment under different operating stages and environmental conditions, and output a specific health score model for the equipment. This model can intuitively represent the health level of the equipment through a quantitative score of 0-100. A score lower than the preset benchmark value indicates that the health status of the equipment is declining. When generating dynamic thresholds, the business load data is first statistically analyzed according to time periods such as daily, weekly, and monthly, and key features such as load peak, average, and fluctuation amplitude are extracted in each period. At the same time, parameters such as equipment runtime, cumulative read and write volume, and wear and tear of core components are quantified into equipment aging coefficients. Based on the above periodic load characteristics and aging coefficients, a dynamic threshold function that changes over time is constructed. This function can adaptively adjust the anomaly judgment criteria according to the periodic fluctuations of business load and the aging process of equipment. For example, during peak business periods or when the equipment is aging, the threshold range can be dynamically widened or tightened.

[0035] The device-specific health rating model obtained through time-series modeling can accurately match the individual operating characteristics of the device and avoid the adaptation bias of general models. By combining the dynamic threshold function of business load periodicity and device aging factors, the limitations of fixed thresholds are completely broken through, making the anomaly judgment criteria more in line with actual application scenarios, significantly reducing the false alarm rate, and providing more targeted judgment basis for subsequent fault prediction.

[0036] As a specific implementation of this application, based on the basic solution, it further defines the analysis of real-time collected data based on a health model and dynamic thresholds to predict potential failures of the storage system and generate corresponding early warning information, including: coupling the real-time health score with the dynamic threshold for analysis to calculate the probability of failure; when the probability of failure exceeds the preset early warning threshold, generating early warning information containing the failure type and the scope of impact.

[0037] Specifically, after preprocessing the multi-source data collected in real time from the storage system, it is input into a pre-constructed device-specific health scoring model to obtain a real-time health score, which directly reflects the current health status of the device. Subsequently, a coupling analysis is performed, extracting the threshold range corresponding to the current time period for the dynamic threshold function. Combining key parameters such as the deviation of the real-time health score from this threshold range and the short-term rate of change of the score (e.g., the rate of score decrease within 5 minutes), a Bayesian probability model or logistic regression algorithm is used to calculate the probability of failure. During the calculation, the correlation weights of data dimensions are simultaneously incorporated (e.g., the weight of disk error rate is higher than that of environmental temperature and humidity) to improve the accuracy of the probability calculation. Preset warning thresholds are based on the availability requirements of the storage system's industry (e.g., 99.999% for telecom-grade systems). The SLA is set to 85%. When the calculated probability of failure exceeds this preset warning threshold, the system automatically associates the abnormal features in the real-time data with the corresponding device operation dimensions (such as disk hardware dimension corresponding to excessive disk error rate, and performance dimension corresponding to abnormal input / output latency), determines the type of failure (such as disk decay failure, heat dissipation system failure, and performance bottleneck failure), and analyzes the scope of impact by combining the affected business modules and data storage partition information in the business load data. Finally, it generates warning information that includes the type of failure, risk level, scope of impact, and preliminary response direction.

[0038] By coupling real-time health scores with dynamic thresholds and accurately calculating fault probabilities, the accuracy of fault prediction is ensured, avoiding the one-sidedness of judgment based on a single indicator. The early warning information includes specific fault types and impact ranges, providing maintenance personnel with clear handling guidelines, significantly shortening fault response preparation time, and further reducing the risk of data loss and service interruption in the storage system.

[0039] As a specific implementation of this application, based on the basic solution, it further specifies that the root cause of the failure is determined by correlation analysis of multi-source data based on the early warning information, including: constructing a causal relationship network based on the dependency relationship between the device, component and business layer; performing probabilistic reasoning on the early warning information based on the causal relationship network, and outputting the confidence ranking of the root cause of the failure.

[0040] Specifically, the process begins by analyzing the data flow, control flow, and resource dependencies between storage system devices (such as disks, controllers, and interface modules), components (such as RAID components, heat dissipation components, and data transmission components), and the business layer (such as data read / write services, backup services, and concurrent access services). Each entity is treated as a network node, and dependencies are used as directed edges to construct a causal relationship network. Each directed edge is labeled with a dependency strength weight (range 0-1) calibrated based on historical fault data and operational experience. Upon receiving an early warning, key features such as the fault type and impact range are extracted to locate the corresponding cluster of related nodes in the causal relationship network. Abnormal parameters from real-time collected operational status data, business load data, and environmental data are used as input evidence for inference. A Bayesian network inference algorithm is employed to perform probabilistic inference on the cluster of related nodes, calculating the posterior probability of each potential root cause node triggering the current early warning. Finally, the confidence scores of the fault root causes are sorted from high to low based on their posterior probability values. The sorting results are synchronized with abnormal data evidence corresponding to each root cause (such as the correlation data between power supply voltage fluctuations and excessive disk error rates).

[0041] By constructing a causal relationship network that closely matches the actual dependencies of the system, a clear logical framework is provided for root cause analysis. Combined with probabilistic reasoning algorithms and confidence ranking, it not only accurately identifies the core causes of failures, but also allows maintenance personnel to prioritize high-confidence root causes, significantly shortening the fault investigation cycle and improving fault handling efficiency.

[0042] As a specific implementation of this application, based on the basic solution, the embodiment of this application further includes: dynamically adjusting the resource allocation strategy of health check tasks according to the predicted future business load; and adopting a resource-sensitive scheduling strategy to prioritize the use of idle system resources when executing check tasks.

[0043] Specifically, based on historical business load data and real-time business operation parameters of the storage system, a time-series prediction algorithm (such as Long Short-Term Memory networks) is used to accurately predict the business load intensity, peak distribution, and fluctuation trend for a preset period (such as 1 hour or 24 hours), generating future business load prediction results. Based on these prediction results, the resource allocation strategy for health check tasks is dynamically adjusted. Specifically, when a low point in the predicted business load is identified, the CPU utilization ratio, memory allocation, and IO bandwidth quota for check tasks are increased to support full-scale deep checks. When a peak in the predicted business load is identified, the resource utilization of check tasks is reduced, retaining only the minimum resources required for core basic checks. Simultaneously, the system resource monitoring module collects data such as CPU idle rate, remaining memory, and disk IO idle bandwidth in real time. Resource idle thresholds are set (such as CPU idle rate ≥ 65%, remaining memory ≥ 30%), and a resource-sensitive scheduling strategy is adopted. When system resources meet the thresholds, check tasks are prioritized for execution using idle resources. If resources are scarce, non-urgent check tasks are placed in a waiting queue and executed only after resources are released, avoiding competition for core resources with the business load.

[0044] By using dynamic resource allocation and resource-sensitive scheduling, the complete execution of health check tasks is ensured, while the impact of resource consumption during high-load periods on storage service performance is avoided, significantly improving system resource utilization and storage service stability.

[0045] As a specific implementation of this application, based on the basic solution, this application further includes, after predicting potential failures of the storage system and generating corresponding early warning information, automatically generating and executing predefined emergency response actions according to the failure type and impact range in the early warning information. The emergency response actions include switching data backup paths, adjusting load balancing strategies, or enabling redundant components.

[0046] Specifically, a mapping database of early warning information features and emergency response actions is pre-built. This database defines predefined emergency response actions and their execution priorities for different fault types (such as disk decay, data transmission path failure, and business load overload) and impact levels (such as single-node impact, multi-node impact, and core business impact). When the system generates an early warning, it automatically parses the fault type identifier and impact range parameters (including the affected business service identifier, data storage partition, and associated device nodes), and accurately matches the appropriate emergency response action based on the mapping database. If the fault type is a data transmission path anomaly and the impact range involves critical business data backup, a data backup path switching action is triggered, immediately terminating data transmission on the current fault path. The process activates the pre-configured redundant backup path. After verifying the link connectivity, subsequent data backup tasks are seamlessly migrated to the backup path. If the fault type is business load overload and the impact covers business processing across multiple nodes, the load balancing strategy is adjusted. The load allocation weight is recalculated based on the real-time load rate (CPU utilization, memory utilization, IO throughput) of each node, and excess business requests from the affected nodes are migrated to low-load nodes. If the fault type is hardware component failure and affects core storage functions, redundant components are activated. Activation commands are sent to preset redundant disks, redundant controllers, and other components. The operating status of the redundant components is verified synchronously. After confirming that they have successfully taken over the functions of the faulty component, the emergency response action is completed, and the action execution log and status feedback are recorded.

[0047] By automatically matching and executing early warning information with predefined emergency response actions, potential faults can be dealt with quickly without manual intervention, significantly shortening the response delay for fault handling. Targeted path switching, load adjustment, or redundancy activation actions can directly block the trend of fault escalation, effectively reducing the risk of data loss and service interruption time, and greatly improving the fault tolerance and continuous availability of the storage system.

[0048] It should be noted that the embodiments of this disclosure may include multiple steps. For ease of description, these steps are numbered, but these numbers are not a limitation on the execution time slots or execution order between the steps; these steps can be implemented in any order, and the embodiments of this disclosure do not limit this.

[0049] Corresponding to the aforementioned storage system failure prediction method, this disclosure also proposes a storage system failure prediction apparatus. Since the apparatus embodiments of this disclosure correspond to the aforementioned method embodiments, details not disclosed in the apparatus embodiments can be referred to the aforementioned method embodiments, and will not be repeated here.

[0050] Figure 2 This is a schematic diagram of the structure of a storage system fault prediction device provided in an embodiment of the present disclosure, as shown below. Figure 2 As shown, it includes: Acquisition unit 21 is used to acquire the storage system's operating status data, business load data, and environmental data; The first generation unit 22 is used to construct a health model representing the health status of the equipment based on the operating status data and environmental data, and to generate dynamic thresholds for anomaly detection in combination with business load data. The second generation unit 23 is used to analyze the real-time collected data based on the health model and dynamic thresholds, predict potential failures of the storage system, and generate corresponding early warning information. Analysis unit 24 is used to determine the root cause of the fault based on the early warning information and through correlation analysis of multi-source data.

[0051] This application provides a storage system fault prediction device. By acquiring the storage system's operating status data, business load data, and environmental data, a health model characterizing the device's health status is constructed. Combined with business load data, a dynamic threshold for anomaly detection is generated. Based on this health model and dynamic threshold, real-time collected data is analyzed to predict potential faults in the storage system. Furthermore, the root cause of the fault is determined through correlation analysis of multi-source data. Therefore, it can solve the problems in existing technologies caused by the use of fixed thresholds, single inspection methods lacking in-depth analysis, weak fault early warning capabilities, and insufficient correlation of multi-source data, which result in high false alarm rates, impact on storage service performance, high risk of data loss or service interruption, and difficulty in locating the root cause of faults. It achieves the technical effects of reducing false alarm rates, ensuring stable storage service performance, providing early warning of potential faults to reduce the risk of data loss and service interruption, and quickly and accurately locating the root cause of faults.

[0052] It should be noted that the foregoing explanation of the method embodiments also applies to the apparatus of this embodiment, and the principle is the same, so it is not limited in this embodiment.

[0053] For a description of the features in the embodiment corresponding to the storage system failure prediction device, please refer to the relevant description of the embodiment corresponding to the storage system failure prediction method, which will not be repeated here.

[0054] Embodiments of this application also provide an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to perform the steps in any of the above embodiments of the storage system failure prediction method.

[0055] Embodiments of this application also provide a computer-readable storage medium storing a computer program, wherein the computer program is configured to execute the steps in any of the above-described embodiments of the storage system fault prediction method when it is run.

[0056] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard disk, magnetic disk, or optical disk.

[0057] Embodiments of this application also provide a computer program product, which includes a computer program that, when executed by a processor, implements the steps in any of the above-described storage system fault prediction method embodiments.

[0058] Embodiments of this application also provide another computer program product, including a non-volatile computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps in any of the above-described storage system fault prediction method embodiments.

[0059] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0060] The foregoing has provided a detailed description of a storage system fault prediction method, apparatus, electronic device, and storage medium provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are merely for the purpose of helping to understand the method and its core ideas. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from its principles, and these improvements and modifications also fall within the protection scope of the claims of this application.

Claims

1. A method for predicting storage system failures, characterized in that, include: Acquire storage system operational status data, business load data, and environmental data; A health model characterizing the health status of the equipment is constructed based on the operational status data and environmental data, and a dynamic threshold for anomaly detection is generated by combining the business load data. Based on the health model and dynamic thresholds, the real-time collected data is analyzed to predict potential failures of the storage system and generate corresponding early warning information. Based on the aforementioned early warning information, the root cause of the fault is determined through correlation analysis of multi-source data.

2. The storage system fault prediction method according to claim 1, characterized in that, The acquisition of the storage system's operating status data, business load data, and environmental data includes: Sensor data from storage devices is collected synchronously using a distributed acquisition architecture. The sensor data includes temperature, voltage, disk error rate, and input / output latency. Feature extraction is performed on the time-series data of business access to obtain access pattern features.

3. The storage system fault prediction method according to claim 1, characterized in that, The process of constructing a health model characterizing the health status of equipment based on the operational status data and environmental data, and generating dynamic thresholds for anomaly detection by combining the business load data, includes: By performing time-series modeling on historical operating status data, a specific health score model for the equipment is obtained; Perform periodic analysis on business load data and generate a dynamic threshold function that changes over time by combining equipment aging factors.

4. The storage system fault prediction method according to claim 1, characterized in that, The process of analyzing real-time collected data based on the health model and dynamic thresholds to predict potential failures of the storage system and generate corresponding early warning information includes: The real-time health score is coupled with the dynamic threshold for analysis to calculate the probability of failure. When the probability of the fault occurring exceeds a preset warning threshold, a warning message containing the fault type and the scope of its impact is generated.

5. The storage system fault prediction method according to claim 1, characterized in that, The step of determining the root cause of the fault based on the early warning information through correlation analysis of multi-source data includes: Construct a causal relationship network based on the dependencies between devices, components, and the business layer; Based on the causal relationship network, probabilistic reasoning is performed on the early warning information, and the confidence ranking of the root causes of the fault is output.

6. The storage system fault prediction method according to claim 1, characterized in that, Also includes: The resource allocation strategy for health check tasks is dynamically adjusted based on the predicted future business load. A resource-sensitive scheduling strategy is adopted to prioritize the use of idle system resources when performing inspection tasks.

7. The storage system fault prediction method according to claim 4, characterized in that, After predicting potential failures of the storage system and generating corresponding early warning information, the method further includes: Based on the fault type and impact range in the warning information, predefined emergency response actions are automatically generated and executed. These emergency response actions include switching data backup paths, adjusting load balancing strategies, or enabling redundant components.

8. A storage system fault prediction device, characterized in that, include: The acquisition unit is used to acquire the storage system's operating status data, business load data, and environmental data. The first generation unit is used to construct a health model representing the health status of the equipment based on the operating status data and environmental data, and to generate a dynamic threshold for anomaly detection in combination with the business load data. The second generation unit is used to analyze the real-time collected data based on the health model and dynamic thresholds, predict potential failures of the storage system, and generate corresponding early warning information. The analysis unit is used to determine the root cause of the fault based on the warning information through correlation analysis of multi-source data.

9. An electronic device, characterized in that, include: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the storage system failure prediction method according to any one of claims 1-7.

10. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to execute the storage system fault prediction method according to any one of claims 1-7.

Images