Software performance bottleneck monitoring system for database interaction

Abstract

The invention relates to the technical field of database monitoring, in particular to a database interaction-oriented software performance bottleneck monitoring system, which comprises a three-dimensional tensor construction module for acquiring a transaction operation chain of each node and a latch contention waveform and an IO bandwidth pulse sequence associated with the transaction operation chain in real time through a distributed transaction probe to generate a three-dimensional tensor; the tensor decomposition module is used for inputting the three-dimensional tensor into a decomposition model to generate a competitive intensity spectrum; the dynamic back pressure regulation and control module is used for identifying competition hotspots according to the competition intensity spectrum, and injecting back pressure pulses with phase difference into a transaction scheduler when detecting that a latch contention waveform and IO bandwidth pulsation form resonance; according to the method, active feedback intervention of a database kernel layer on competitive behaviors is realized, the problem of centralized preemption of lock resources caused by frequency resonance can be effectively broken through especially through scheduling offset and a progressive exit mechanism, and the instantaneous congestion probability of the system and the transaction-level delay fluctuation amplitude are reduced on the premise of not influencing the transaction consistency.

Description

Technical Field

[0001] The present invention relates to the technical field of database monitoring, and in particular to a software performance bottleneck monitoring system oriented to database interaction. Background Art

[0002] As enterprise-level business systems rapidly evolve towards distributed, heterogeneous architectures, the operational performance of databases, as core data management components, has become a key factor affecting system response speed and overall throughput. In a high-concurrency transaction environment, resource contention between database nodes is common, especially in dimensions such as latch holding and I / O read / write conflicts, which are more likely to cause short-term, large-scale performance jitter, leading to a surge in transaction execution delays. In severe cases, it can even induce risks such as transaction deadlock and resource starvation.

[0003] Existing performance monitoring tools are mostly based on operating system-level metric sampling or transaction-level log analysis. They lack the ability to directly observe database kernel-level resource behavior and are unable to capture subtle changes in contention states. Furthermore, traditional performance modeling methods often employ single-dimensional metric analysis or fixed-structure decomposition models, making it difficult to accurately characterize the contention transmission paths and timing resonance patterns between multiple nodes. This results in poor generalization of performance bottleneck identification results and high hysteresis in regulatory feedback.

[0004] In addition, most current database systems lack active intervention mechanisms. When high-intensity resource contention breaks out, scheduling strategies often rely on static priority settings or simple timeout retry logic, making it difficult to effectively suppress resonant congestion and unable to dynamically adapt to the transaction execution order adjustment under sudden IO fluctuations. Summary of the Invention

[0005] The present invention provides a software performance bottleneck monitoring system for database interaction. The performance bottleneck monitoring system is oriented to the database interaction environment and has microscopic perception capabilities and spatiotemporal collaborative analysis capabilities. It can achieve a detailed characterization of resource conflict patterns in a multi-dimensional competitive feature space, and optimize and regulate the system execution path in real time based on a dynamic back pressure mechanism, thereby fundamentally improving transaction processing stability and system response performance.

[0006] A software performance bottleneck monitoring system for database interaction, including: 3D tensor building module: This module uses distributed transaction probes to collect transaction operation chains of each node and their associated latch contention waveforms and I / O bandwidth pulsation sequences in real time to generate a 3D tensor. Tensor decomposition module: performs tensor decomposition of the competition pattern, inputs the three-dimensional tensor into the decomposition model, separates the core feature matrix representing global resource competition and the subsidiary feature matrix representing local transaction interference, and generates a competition intensity spectrum; Dynamic back pressure control module: identifies contention hotspots based on the contention intensity spectrum. When it is detected that the latch contention waveform resonates with the IO bandwidth pulsation, a back pressure pulse with a phase difference is injected into the transaction scheduler. The back pressure pulse includes a frequency suppression parameter and a transaction priority remapping table.

[0007] Optionally, the three-dimensional tensor construction module specifically includes: Deploy a distributed transaction probe cluster. Each probe is embedded in the database kernel layer and synchronously captures the transaction operation chain of each node and its associated latch contention waveform and I / O bandwidth pulsation sequence within a preset time window. Construct a 3D tensor: The first dimension is the time axis, where each node collects data aligned with the global clock; The second dimension is the node topology axis, which encodes the mapping relationship between physical node locations and logical shards; The third dimension is the competition index axis, which quantifies the latch request frequency into a frequency domain energy spectrum and converts the IO rate gradient into a time domain envelope signal.

[0008] Optionally, the transaction operation chain includes a cross-node operation hash mark and a timestamp sequence; The latch contention waveform includes request frequency, hold duration, and wait queue depth; The capture of the IO bandwidth pulsation sequence includes extracting the read and write operation rate variation gradient in real time through a throughput differential calculator.

[0009] Optionally, the three-dimensional tensor construction module also includes spatiotemporal regularization of the three-dimensional tensor through a competition coupling analyzer. When a shared latch is detected in a cross-node transaction operation chain, a competition association mark is injected into the corresponding time slice to generate an enhanced tensor with a competition conduction mark.

[0010] Optionally, the tensor decomposition module comprises decomposing the three-dimensional tensor input using an improved TuckER decomposition model, where the improved TuckER decomposition model comprises: A sliding time window mechanism is introduced to dynamically adjust the rank parameter of the tensor core, so that the core feature matrix dimension can be adaptively expanded with the competition intensity. Add a competitive coupling constraint term to the decomposition loss function.

[0011] Optionally, the decomposition process includes an iterative step: Initialize the core tensor G, whose dimension is determined by principal component analysis of the frequency domain energy spectrum of the competitive indicator axis; Alternately optimize the attachment moments A and B, where A represents the competition propagation pattern in the node topology dimension and B represents the competition evolution trend in the time dimension; The decomposition stopping condition is triggered when the cosine similarity of the core matrices of adjacent iterations is greater than 0.95.

[0012] Optionally, the generation of the competition intensity spectrum includes modal expansion of the core feature matrix G, calculating the weight value of each dimension along the competition index axis, weighted fusion of the frequency domain energy spectrum amplitude, time domain envelope signal strength, and node topology coupling degree, and outputting a spectrum graph with three-dimensional characteristics of time-space-intensity to form a competition intensity spectrum.

[0013] Optionally, the dynamic back pressure control module specifically includes: Resonance peak detection: Extract key data sets from the competition intensity spectrum, perform cross-dimensional coupling analysis on the key data sets, traverse each pair of frequency and time combination items, and determine whether they meet the synchronization conditions in the time dimension. If they do, calculate the corresponding resonance contribution value based on the energy intensity at the latch frequency and the change amplitude of the IO envelope signal at the corresponding time point. Accumulate all resonance contribution values that meet the synchronization conditions to obtain a resonance index. The resonance index is compared with the resonance trigger threshold. If it exceeds, it is determined to be in a resource resonance state, triggering the back-pressure pulse control process. Back-pressure pulse generation mechanism: Identify the dominant resonant frequency from the frequency domain energy spectrum. Based on the topological distance between the dominant resonant frequency and each node, calculate the phase difference used to break the resonant synchronization. Build a delay suppression mechanism for lock request scheduling based on this phase difference. Based on the current changes in the I / O interference intensity of each node, establish priority remapping rules to automatically assign low scheduling priority to transactions with drastic changes in I / O load.

[0014] Control instruction execution: The calculated phase difference value is converted into a scheduling delay instruction and injected into the database kernel layer lock scheduler. The priority remapping table is loaded through the transaction scheduler to reorder the execution of currently active transactions.

[0015] Optionally, the key data set includes a main response frequency set representing latch resource competition in the frequency domain and a mutation moment set representing IO communication interference on the time axis.

[0016] Optionally, the execution of the control instruction also includes delaying the processing of potential interference transactions, continuously monitoring the changes in the resonance index, and starting the control exit procedure after it drops to a safety release threshold.

[0017] Beneficial effects of the present invention: The present invention deploys a high-frequency transaction probe cluster at the database kernel layer to synchronously collect cross-node transaction chains, latch contention waveforms, and IO bandwidth pulsation sequences, and construct a three-dimensional tensor, thereby improving sampling granularity and timing accuracy. It also realizes key competition parameters such as lock holding status, request frequency, and IO gradient fluctuation, and completely restores the coupling behavior trajectory during the instantaneous resource contention process, providing a data basis for subsequent bottleneck identification and control.

[0018] Based on the TuckER tensor decomposition algorithm, the present invention introduces a dynamic kernel rank adjustment mechanism driven by a sliding time window, and embeds a coupling constraint term based on the competition label tensor in the loss function, so that the dimension of the core feature matrix can be adaptively adjusted with the complexity of the competition pattern, which not only retains the cross-node propagation trend, frequency domain resonance component and local transaction disturbance characteristics, but also avoids the overfitting interference of global features.

[0019] The resonance detection mechanism proposed in the present invention comprehensively considers the time-frequency coupling relationship between the locking frequency energy spectrum and the IO envelope mutation, and judges whether the system enters a high-risk synchronization state through the cross-dimensional resonance index. On this basis, a phase disturbance injection strategy, a delay control distribution model based on load characteristics, and real-time transaction priority remapping rules are constructed, which realizes the active feedback intervention of the database kernel layer on the competitive behavior. In particular, through the scheduling offset and progressive exit mechanism, it can effectively break the concentrated preemption of lock resources caused by frequency resonance, reduce the system instantaneous congestion probability and transaction-level delay fluctuation amplitude without affecting transaction consistency, and enhance the system operation stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only for the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0021] Figure 1 A schematic diagram of system operation logic according to an embodiment of the present invention; Figure 2 Schematic diagram of system function modules according to an embodiment of the present invention. DETAILED DESCRIPTION

[0022] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments. It is also noted that, to provide a more detailed description, the following embodiments are best and preferred embodiments, and those skilled in the art may employ alternative methods for implementing certain known technologies. Furthermore, the accompanying drawings are intended only to provide a more detailed description of the embodiments and are not intended to limit the present invention.

[0023] It should be noted that references in the specification to "one embodiment," "an embodiment," "exemplary embodiments," "some embodiments," etc. indicate that the described embodiments may include specific features, structures, or characteristics, but not necessarily every embodiment will include such specific features, structures, or characteristics. Furthermore, when specific features, structures, or characteristics are described in conjunction with an embodiment, it is within the knowledge of persons skilled in the relevant art to implement such features, structures, or characteristics in conjunction with other embodiments (whether or not explicitly described).

[0024] In general, terms can be understood, at least in part, from their use in context. For example, depending at least in part on the context, the term "one or more" as used herein can be used to describe any feature, structure, or characteristic in the singular sense, or can be used to describe a combination of features, structures, or characteristics in the plural sense. Additionally, the term "based on" can be understood as not necessarily intended to convey an exclusive set of factors, but can instead, depending at least in part on the context, allow for the presence of other factors that are not necessarily explicitly described.

[0025] like Figure 1-Figure 2 As shown in the figure, the software performance bottleneck monitoring system for database interaction includes: 3D tensor building module: This module uses distributed transaction probes to collect transaction operation chains of each node and their associated latch contention waveforms and I / O bandwidth pulsation sequences in real time to generate a 3D tensor. Tensor decomposition module: This module performs tensor decomposition of competition patterns, inputs a three-dimensional tensor into the decomposition model, separates the core feature matrix representing global resource competition and the subsidiary feature matrix representing local transaction interference, and generates a competition intensity spectrum. Dynamic back-pressure control module: Identifies contention hotspots based on the contention intensity spectrum. When it detects that the latch contention waveform resonates with the I / O bandwidth pulsation, it injects a back-pressure pulse with a phase difference into the transaction scheduler. The back-pressure pulse includes frequency suppression parameters and a transaction priority remapping table.

[0026] Specific example description: When distributed transactions cause latch contention among multiple nodes (such as in Oracle RAC scenarios): The first step is to capture the latch operation frequency (e.g., 500 times per second) and I / O throughput fluctuation (e.g., ±30%) of each node. The second step is to use tensor decomposition to find that the two form a resonance peak at a frequency of 125Hz (which cannot be correlated with traditional monitoring tools); The third step injects back-pressure pulses: adding a 50ms phase delay to high-frequency latch requests and boosting the priority of I / O-intensive transactions. In the fourth step, the system eliminated the resonance phenomenon within 20ms, and the transaction throughput returned to stability (the measured response speed was relatively fast).

[0027] The three-dimensional tensor building blocks specifically include: a) Deploy a distributed transaction probe cluster, each probe is embedded in the database kernel layer, ms is the time window, and the following data are captured synchronously: 1. Transaction operation chain data: ,in, Indicates the On the node The hash mark of the operation chain, The corresponding timestamp, Representation node The number of transaction operation chains on the .

[0028] 2. Latch contention waveform: ,in, Representation node The latch contention waveform vector on Indicates the request frequency (the number of requests for the same lock per unit time). Indicates the lock holding duration, Indicates the waiting queue depth.

[0029] 3.IO bandwidth pulse sequence: Among them, B i (t) represents the IO bandwidth pulse sequence vector on node i, R i (t) represents the cumulative throughput sequence of IO read operations, represents the read rate gradient (time differential of IO read throughput), W i (t) represents the cumulative throughput sequence of IO write operations, Represents the write rate gradient (time differential of IO write throughput).

[0030] b) Construct a three-dimensional tensor , its three dimensions are defined as follows: 1. Time Dimension :According to the global clock synchronization sampling, each frame corresponds to ms collection fragment.

[0031] 2. Node topology dimension n: represents the physical node number and logical shard mapping identifier to which the sampled data belongs, in the form of n=max(i), that is, mapping the physical node i to the logical shard topology position.

[0032] 3. Competitive indicator dimension :include: Frequency domain energy spectrum of latch request frequency : ,in Represents Fourier transform, which is used to obtain the frequency domain characteristics of the signal; Time domain envelope signal of IO rate gradient : , Represents the signal envelope function.

[0033] c) Enhanced tensor generation: For the above three-dimensional tensor Enter the competitive coupling analyzer: Determine whether there is cross-node shared latch access (e.g., two transaction operation chains share the same lock ID); If there is shared behavior, locate the corresponding time slice and node groups , then inject the competitive conduction flag into the tensor: ,in Inject value for competition conduction flag, is the corresponding indicator dimension index, usually the lock frequency or IO fluctuation position, is the node pair where resource competition occurs.

[0034] Specific example data flow: When a transaction on node A updates a data page, triggering a read operation on node B: The probe captures a sudden increase in the latch holding time of node A to 120ms (baseline value 80ms); Node B detected that the IO read latency climbed from 15ms to 90ms; The competitive coupling analyzer finds the association between the two through the hash mark of the operation chain; Generate a red warning flag tensor within the time slice [TS-45, TS+5]; Based on this, the back pressure control module prioritizes limiting the batch write transactions of node A.

[0035] The tensor decomposition module specifically includes: a) Improved TuckER decomposition model design: for three-dimensional tensors Perform a modified TuckER decomposition, including the following: Dynamic rank adjustment: introducing a sliding time window mechanism ,The rank of the core tensor is dynamically set according to the ,change of competition intensity within each window, so that the ,dimension of the core feature tensor changes adaptively.

[0036] The loss function with competitive coupling constraints is expressed as: ,in, is the competition association tag tensor (generated from the S1 enhanced tensor), represents the intermediate feature matrix in the decomposition process (representing potential competitive factors), is the core tensor (global resource competition core), is the node dimension attachment matrix (propagation mode), is the time dimension attachment matrix (evolution trend), represents element-wise product, Represents the L1 norm, which is used to guide sparse expression. represents the Frobenius norm, which is used to measure the tensor decomposition error. represents the mode-3 product (tensor modular multiplication along the third dimension), is the regularization weight, dynamically dependent on the tensor structure adjustment, see below.

[0037] b) Decompose the iterative process: 1. Initialize the core tensor : Set the initial rank dimension based on the frequency domain principal component analysis (PCA) results of the competitive indicator axis in the tensor; 2. Alternating optimization of the attachment matrix: optimization ,encoding the competitive conduction structure among the nodes; optimization ,capturing the dynamic evolution of competition intensity in the time dimension.

[0038] 3. Iteration termination condition: the core tensor of the current iteration Compared with the previous time The iteration is terminated when the cosine similarity of satisfies the following conditions: .

[0039] c) Competition intensity spectrum generation: 1. Core Tensor Along the index dimension Direction expansion into a matrix ; 2. Extract the three main weight factors on the indicator dimension: Frequency domain energy spectrum amplitude (lock contention frequency characteristics); Time domain envelope strength (IO fluctuation amplitude characteristics); Node coupling degree (cross-node interference density characteristics); 3. Perform weighted fusion and output a competition intensity spectrum in the form of three-dimensional coordinates : ;in, represents the intensity dimension, is the normalized weight, represents the frequency domain energy spectrum feature, Indicates the IO envelope signal strength characteristics, Represents the node competition coupling coefficient.

[0040] Definition of dynamic regularization term weight coefficient: ; in, represents the number of non-zero elements in the competition mark tensor (indicating the number of real competition instances), is the total number of elements in the tensor, represents the current core tensor rank, Represents an S-shaped nonlinear function used to smoothly control the constraint strength.

[0041] It should be further explained that the extraction of the three main weight factors in the competition index dimension is as follows: 1. Frequency domain energy spectrum amplitude (lock contention frequency characteristics): Extract the lock request frequency signal of each node in each time window; Performing Fourier transform on the signal to obtain a frequency domain amplitude spectrum; Extract the maximum response peak within the preset frequency range (50–250 Hz) and record its frequency position and intensity; The spectrum peak is used as the main characteristic of frequency competition to identify potential lock contention resonance phenomenon.

[0042] 2. Time Domain Envelope Strength (IO Fluctuation Amplitude Characteristics) Get the IO read and write rate gradient time series; Use Hilbert transform to construct analytical signal and extract its envelope curve; Calculate the maximum value, average change rate and instantaneous peak span of the envelope curve within the window to reflect the intensity of burst IO interference; Its peak value and change gradient are used as the amplitude characteristic factors reflecting the IO disturbance.

[0043] 3. Node coupling (cross-node interference density characteristics): Based on the contention tag information in the enhanced tensor, node pairs with shared latch access behavior are identified; Count the number of competitive associations between each pair of nodes and their temporal concentration (using the competition density per unit time); According to the competition graph between nodes, the sum of the connection edge weights of each node is calculated as the competition coupling degree of the node; The node coupling degree is normalized and mapped to the competition index dimension as the main factor reflecting the intensity of cross-node conflicts.

[0044] The specific example decomposition process of the tensor decomposition module is as follows: When distributed transactions are detected causing latch oscillations on multiple nodes: 1. The three-dimensional tensor size is [time axis: 1200 frames, node axis: 8 nodes, index axis: 15 dimensions]; 2. The decomposition initialization core matrix is 5×5×5, and the cosine similarity reaches 0.97 after 3 iterations; 3. A sudden increase in the weight of the third dimension was detected in the core matrix G (the strength value increased from 0.15 to 0.82); 4. The competitive intensity spectrum shows that this dimension corresponds to: Frequency domain energy peak @128Hz (latch operation resonant frequency); Temporal envelope intensity 0.75 (IO gradient change exceeds threshold); Node coupling degree is 0.91 (nodes 3 and 5 form a competitive channel); 5. The phase difference of the reverse pressure pulse that triggers the S3 stage of this spectrum feature is set to 1 / 128 Hz ≈ 7.8 ms.

[0045] The dynamic back pressure control module specifically includes: a) Resonant peak detection: 1. Extract from the competition intensity spectrum: Peak frequency set of the latch frequency domain energy spectrum : , is the frequency The energy spectrum amplitude of The set of mutation time points of the IO envelope signal : , The threshold for determining the sudden change of the IO envelope slope is used to determine whether the rate of change of the IO bandwidth envelope curve G(t) is drastic enough to constitute a "sudden change event". The gradient data is sampled according to time. , calculate the 90% quantile of its distribution, setting: , suitable for dynamic flow systems, automatically maintaining a relatively stable discrimination ratio.

[0046] 2. Calculate the resonance index: ; in, is the synchronization tolerance function, indicating the frequency Cycle and time point The degree of closeness (within the ±50μs tolerance range is recorded as 1, otherwise it is recorded as 0).

[0047] 3. Technical points 3. Resonance determination: If , then the back pressure control mechanism is triggered, The resonance trigger threshold is set at the initial stage of system startup, and a 5-minute training cycle is set to calculate the resonance index without control intervention during this period. distribution, calculate its expected value and standard deviation , set the resonance trigger threshold to: , Set the value to 3 (control trigger sensitivity) to reduce false triggering.

[0048] b) Back pressure pulse generation: 1. Phase difference calculation: Determine the main resonant frequency : ; Calculate the back pressure pulse phase difference : ;in, represents the damping coefficient estimated by the topological distance between nodes.

[0049] 2. Frequency suppression parameter injection: Set the latch request delay to: ,in, is the current load factor of the system. Normal distribution is used to avoid lock contention behavior occurring at the phase synchronization point; 3. Transaction priority remapping: For IO-intensive transactions, reset the priority to: ;in, Indicates the IO urgency coefficient, which indicates the sensitivity to bandwidth degradation. Represents the real-time gradient of the IO bandwidth envelope signal, is the adjusted transaction priority, is the original transaction priority, It is the real-time slope of the IO signal (reflecting the sudden trend).

[0050] c) Control instruction execution process: 1. Microsecond-level control injection: The phase difference parameter Encoded as a clock offset instruction with microsecond precision; injected into the database kernel layer scheduler to delay latch scheduling events.

[0051] 2. Priority table loading and scheduling rescheduling: The transaction scheduler loads the updated priority remapping table; according to Adjust the transaction execution order online to avoid peak resource usage.

[0052] 3. Regulation and Exit Mechanism: Monitoring resonance index Drop to the safety release threshold After the following, the back pressure gradual evacuation process is started, and the safety release threshold is set to 60% of the trigger threshold according to the trigger threshold ratio: , The recommended value is 0.6; Use exponential decay strategy to release phase delay: ,in, Expressed as the time constant for regulation exit.

[0053] The present invention encompasses any alternatives, modifications, equivalents, and solutions that fall within the spirit and scope of the present invention. To provide a thorough understanding of the present invention, specific details are described in detail below in connection with the preferred embodiments of the present invention, but those skilled in the art will be able to fully understand the present invention without these detailed descriptions. Furthermore, to avoid unnecessary confusion regarding the essence of the present invention, well-known methods, processes, procedures, components, and circuits have not been described in detail.

[0054] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A software performance bottleneck monitoring system for database interaction, characterized by: include: 3D tensor building module: This module uses distributed transaction probes to collect transaction operation chains of each node and their associated latch contention waveforms and I / O bandwidth pulsation sequences in real time to generate a 3D tensor. Tensor decomposition module: performs tensor decomposition of the competition pattern, inputs the three-dimensional tensor into the decomposition model, separates the core feature matrix representing global resource competition and the subsidiary feature matrix representing local transaction interference, and generates a competition intensity spectrum; Dynamic back pressure control module: identifies contention hotspots based on the contention intensity spectrum. When it is detected that the latch contention waveform resonates with the IO bandwidth pulsation, a back pressure pulse with a phase difference is injected into the transaction scheduler. The back pressure pulse includes a frequency suppression parameter and a transaction priority remapping table.

2. The database interaction-oriented software performance bottleneck monitoring system according to claim 1, characterized in that: The three-dimensional tensor construction module specifically includes: Deploy a distributed transaction probe cluster. Each probe is embedded in the database kernel layer and synchronously captures the transaction operation chain of each node and its associated latch contention waveform and I / O bandwidth pulsation sequence within a preset time window. Construct a 3D tensor: The first dimension is the time axis, where each node collects data aligned with the global clock; The second dimension is the node topology axis, which encodes the mapping relationship between physical node locations and logical shards; The third dimension is the competition index axis, which quantifies the latch request frequency into a frequency domain energy spectrum and converts the IO rate gradient into a time domain envelope signal.

3. The database interaction-oriented software performance bottleneck monitoring system according to claim 2, characterized in that: The transaction operation chain includes a cross-node operation hash mark and a timestamp sequence; The latch contention waveform includes request frequency, hold duration, and wait queue depth; The capture of the IO bandwidth pulsation sequence includes extracting the read and write operation rate variation gradient in real time through a throughput differential calculator.

4. The database interaction-oriented software performance bottleneck monitoring system according to claim 2, characterized in that: The three-dimensional tensor construction module also includes temporal and spatial regularization of the three-dimensional tensor through a competition coupling analyzer. When a shared latch is detected in a cross-node transaction operation chain, a competition association mark is injected into the corresponding time slice to generate an enhanced tensor with a competition conduction mark.

5. The database interaction-oriented software performance bottleneck monitoring system according to claim 1, characterized in that: The tensor decomposition module comprises decomposing the three-dimensional tensor input using an improved TuckER decomposition model, wherein the improved TuckER decomposition model comprises: A sliding time window mechanism is introduced to dynamically adjust the rank parameter of the tensor core, so that the core feature matrix dimension can be adaptively expanded with the competition intensity. Add a competitive coupling constraint term to the decomposition loss function.

6. The database interaction-oriented software performance bottleneck monitoring system according to claim 5, characterized in that: The decomposition process includes the iterative steps: Initialize the core tensor G, whose dimension is determined by principal component analysis of the frequency domain energy spectrum of the competitive indicator axis; Alternately optimize the attachment moments A and B, where A represents the competition propagation pattern in the node topology dimension and B represents the competition evolution trend in the time dimension; The decomposition stopping condition is triggered when the cosine similarity of the core matrices of adjacent iterations is greater than 0.

95.

7. The database interaction-oriented software performance bottleneck monitoring system according to claim 6, characterized in that: The generation of the competition intensity spectrum includes modal expansion of the core feature matrix G, calculation of the weight value of each dimension along the competition index axis, weighted fusion of the frequency domain energy spectrum amplitude, time domain envelope signal strength, and node topology coupling degree, and output of a spectrum graph with three-dimensional characteristics of time-space-intensity to form a competition intensity spectrum.

8. The database interaction-oriented software performance bottleneck monitoring system according to claim 1, characterized in that: The dynamic back pressure control module specifically includes: Resonance peak detection: Extract key data sets from the competition intensity spectrum, perform cross-dimensional coupling analysis on the key data sets, traverse each pair of frequency and time combination items, and determine whether they meet the synchronization conditions in the time dimension. If they do, calculate the corresponding resonance contribution value based on the energy intensity at the latch frequency and the change amplitude of the IO envelope signal at the corresponding time point. Accumulate all resonance contribution values that meet the synchronization conditions to obtain a resonance index. The resonance index is compared with the resonance trigger threshold. If it exceeds, it is determined to be in a resource resonance state, triggering the back-pressure pulse control process. Back-pressure pulse generation mechanism: This mechanism identifies the dominant resonant frequency from the frequency domain energy spectrum. Based on the topological distance between the dominant resonant frequency and each node, it calculates the phase difference needed to break resonant synchronization. A delay suppression mechanism for lock request scheduling is built around this phase difference. Priority remapping rules are established based on the current I / O interference intensity changes of each node, automatically assigning low scheduling priority to transactions with drastically changing I / O loads. Control instruction execution: The calculated phase difference value is converted into a scheduling delay instruction and injected into the database kernel layer lock scheduler. The priority remapping table is loaded through the transaction scheduler to reorder the execution of currently active transactions.

9. The database interaction-oriented software performance bottleneck monitoring system according to claim 8, characterized in that: The key data set includes a main response frequency set representing latch resource competition in the frequency domain and a mutation time set representing IO communication interference in the time axis.

10. The database interaction-oriented software performance bottleneck monitoring system according to claim 9, characterized in that: The execution of the control instruction also includes delaying the processing of potential interference transactions, continuously monitoring the changes in the resonance index, and starting the control exit procedure after it drops to a safety release threshold.

Images