Precise operation and maintenance quantification and intelligent recommendation method based on project dimension

Abstract

The invention relates to the field of computer internet and cloud computing, and discloses a project dimension-based precise operation and maintenance quantification and intelligent recommendation method, which comprises the following steps of: encoding distributed link tracking data into an ordered service gene sequence, and realizing automatic identification and resource quantification of a project by using a sequence alignment algorithm. By extracting a gene mutation rate and an epigenetic modification tag, system stability and external environment constraint are dynamically sensed. On the basis, a non-cooperative game model containing performance, cost and risk agents is constructed, game rules are dynamically reconstructed based on gene characteristics, risk sensitivity and constraint strategy space are adjusted in a non-linear mode, a biological adaptability evolution mechanism is simulated, a horizontal gene transfer mechanism is introduced, and a cold start decision of a new project is optimized. And finally, an optimal operation and maintenance configuration instruction is generated by solving a Nash equilibrium point. According to the method, accurate quantification, dynamic self-adaption and risk control of operation and maintenance decisions in a complex cloud environment are effectively realized.

Description

Technical Field

[0001] This invention relates to the fields of computer internet and cloud computing technology, specifically to a method for precise operation and maintenance quantification and intelligent recommendation based on project dimensions. Background Technology

[0002] With the popularization of cloud computing technology and the evolution of microservice architecture, the complexity of modern software systems is increasing exponentially. Cloud-native environments based on containerized orchestration have become the standard configuration of enterprise IT infrastructure. Amidst massive microservice instances and intricate call chains, achieving accurate, efficient, and automated operational decisions has become crucial for ensuring business continuity.

[0003] Existing operation and maintenance management technologies typically focus on physical resource metrics at the infrastructure level, mainly relying on preset static thresholds or simple rule engines to trigger actions. This coarse-grained resource perspective lacks the ability to perceive the upper-layer business logic, i.e., the "project" dimension. As a result, when multiple business scenarios share the underlying services, it is impossible to distinguish the business value and priority of traffic, often leading to non-core businesses preempting core resources or insufficient resource guarantees for critical paths.

[0004] Meanwhile, traditional static strategies appear too rigid in the face of highly dynamic traffic fluctuations and frequent code changes in cloud-native environments. Existing technical solutions often fail to effectively distinguish between normal resource demands due to business growth and abnormal resource consumption caused by code defects, lacking a comprehensive understanding of internal code changes and the external operational environment. This lack of contextual information makes it difficult for operational strategies to find a dynamic balance between aggressive resource provisioning and conservative risk control.

[0005] Furthermore, operational decisions inherently involve a multi-objective game between performance assurance, cost control, and system stability. Existing optimization algorithms typically focus on single-dimensional optimization, lacking a mechanism capable of quantifying the interests of all parties and seeking a globally stable solution in a non-cooperative environment. Especially for newly launched business projects or those lacking historical data, existing intelligent operation and maintenance models trained on massive amounts of historical samples often face severe "cold start" problems, unable to provide effective strategy recommendations in the initial stages of deployment, requiring a lengthy and high-risk trial-and-error phase. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a precise operation and maintenance quantification and intelligent recommendation method based on the project dimension. This solves the technical problems of existing cloud-native operation and maintenance technologies lacking the ability to perceive the business project dimension in a refined manner, making it difficult to distinguish between abnormal and normal traffic fluctuations caused by code changes, and making it difficult to achieve a dynamic balance and accurate decision-making among performance, cost and operational risks in a changing operation and maintenance environment.

[0007] To achieve the above objectives, the present invention provides the following technical solution: In its first aspect, the present invention provides a method for precise operation and maintenance quantification and intelligent recommendation based on project dimensions. This method addresses the problems of project identification and resource quantification by introducing the sequence alignment concept from bioinformatics, and combines multi-agent game theory to solve the dynamic optimization problem of operation and maintenance strategies. The method includes the following steps: collecting distributed link tracing data from cloud infrastructure and encoding it into ordered service gene sequences; using a sequence alignment algorithm, comparing the service gene sequences with benchmark sequences in a project benchmark gene library, determining the project affiliation of the service gene sequences based on the alignment results, and quantifying physical resource consumption based on the project affiliation.

[0008] While identifying project affiliation, the dynamic features of the service gene sequence are extracted. These dynamic features include at least the gene mutation rate reflecting the frequency of business logic changes and the epigenetic modification tag reflecting the external operation and maintenance environment. A multi-agent game model is constructed for the identified affiliation project. The multi-agent game model contains multiple agents representing different operation and maintenance objectives. Based on the dynamic features, the game rules of the multi-agent game model are dynamically reconstructed. The Nash equilibrium point of the reconstructed multi-agent game model is solved, and the strategy combination that reaches the Nash equilibrium state is generated as the operation and maintenance recommendation strategy.

[0009] In one alternative implementation, regarding the construction of the service gene sequence: In order to transform discrete operation and maintenance data into a comparable structured sequence, this invention extracts each service interaction in the distributed link tracing data as a base, and extracts the microservice node identifier. Interface method signature and operation type eigenvectors And combine them in chronological order to form service gene sequences. , where L is the sequence length.

[0010] In one alternative implementation, regarding the principle of item identification based on sequence alignment: This invention employs a dynamic programming algorithm to address the problem of ambiguous resource ownership in cloud-native environments. Specifically, a score matrix F is constructed, and the real-time observation sequence is calculated. With project benchmark sequence The similarity. Elements in the score matrix. Calculated recursively using the following formula: ; in, This is a vacancy penalty factor used to tolerate the absence or insertion of non-critical nodes; Matching scoring function. This involves calculating the maximum similarity score. Determine project ownership The physical resource consumption associated with this sequence will be accumulated to the project. In the resource account, thus achieving precise quantification based on logical dimensions.

[0011] In an optional implementation, regarding the extraction of gene dynamic characteristics: the gene mutation rate extracted by this invention. The formula used to quantify the instability of a project's architecture is as follows: ; in, To set the total number of call sequences within a time window, This represents the number of sequences that exhibit structural differences from the baseline sequence. Simultaneously, epigenetic modification tags are generated. This label is a binary feature vector that describes the external operation and maintenance environment context.

[0012] In one alternative implementation, regarding the construction of the multi-agent game model: To resolve multi-objective conflicts in operational decision-making, this invention constructs three non-cooperative agents: a performance agent, whose utility function... Negatively correlated with service response latency; cost agent, whose utility function Negatively correlated with resource overhead; risk agent, whose utility function With system crash probability Negative correlation.

[0013] In one alternative implementation, regarding the dynamic reconstruction of game rules (the core innovation of this invention): This invention utilizes genetic characteristics to alter the underlying mathematical structure of the game, rather than simply fine-tuning parameters. Specifically, it establishes a mapping relationship between gene mutation rate and the discourse power of the risk agent. When the mutation rate... As it increases, it passes through a monotonically increasing function. Modified risk utility function : ; This causes projects with more unstable architectures to have their game equilibrium points tilted towards conservative, stable strategies. Furthermore, establishing epigenetic modification tags... With strategy space The constraint relationship. When When given specific circumstances, force the locking of a specific subset of policies for a particular agent: ; This enables policy boundary control based on environmental perception.

[0014] In one alternative implementation, regarding the Nash equilibrium solution: this invention finds a strategy combination through an iterative optimal response algorithm. To satisfy the Nash equilibrium condition: ; in This represents the optimal strategy for all agents other than agent i. This equilibrium point represents the optimal operational recommendation under the current gene state and environmental constraints.

[0015] In one alternative implementation, regarding cold start optimization: To address the issue of insufficient historical data for newly added projects, this invention employs a horizontal gene transfer mechanism. This involves searching the benchmark library for data relevant to the new project. The mature project with the highest gene sequence similarity Its steady-state strategy parameters Mapped to the initial state of the new project This enables the transfer of experience across projects.

[0016] A second aspect of the present invention provides an electronic device including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method described in the first aspect.

[0017] A third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in the first aspect.

[0018] This invention provides a method for precise operation and maintenance quantification and intelligent recommendation based on the project dimension. It has the following beneficial effects: 1. This invention abandons the outdated methods of traditional operations and maintenance that rely on manual tagging or static configuration management databases to maintain project relationships, and innovatively introduces a bioinformatics sequence alignment algorithm. By abstracting the real-time microservice call chain as a "service gene," the system can dynamically identify which business logic "expression" each underlying resource call belongs to at a millisecond-level time granularity. This enables the system to accurately "segment" and attribute the loss of shared resources to specific projects, even in cloud-native environments with complex service meshes and frequent code iterations, achieving a leap from physical resource monitoring to business logic quantification.

[0019] 2. This invention establishes a non-cooperative game model composed of three intelligent agents representing performance, cost, and risk. This model not only simulates the trade-off process of human operations experts in multi-objective conflicts, but more importantly, it mathematically transforms the decision-making mechanism into a Nash equilibrium solution process. This means that the system no longer mechanically executes preset instructions, but automatically seeks the optimal solution under current resource constraints through the confrontation and compromise between intelligent agents, fundamentally avoiding the extreme situations of "sacrificing experience for saving money" or "infinite waste for stability" caused by single-objective optimization.

[0020] 3. A risk immunity mechanism based on architecture stability awareness has been implemented. This invention uniquely establishes a dynamic mapping mechanism between "gene mutation rate" and "risk discourse power." The system can keenly detect subtle changes in project code or call structure. Once an architecture instability is detected, the penalty weight of the risk agent is automatically increased in the game model. This mechanism gives the operation and maintenance system an early warning capability similar to a biological immune system—automatically tightening policy boundaries and prioritizing survival during high-risk periods of frequent business changes; while automatically releasing space and pursuing ultimate resource efficiency during periods of architectural stability, effectively preventing cascading failures caused by changes.

[0021] 4. This invention proposes an environment-aware policy locking mechanism. By identifying external environment tags such as major promotions, network shutdowns, and drills, the system can directly physically lock the policy space of the agent. This design cleverly solves the pain point that general algorithms struggle to cope with special business scenarios, ensuring that at critical moments, the system's decision-making logic can be strongly overridden by "business rules," preserving the intelligence of the algorithm while also taking into account the rigid constraints of the business scenario.

[0022] 5. Significantly reduces the cold start cost of launching new services. Addressing the challenges of cold starts due to a lack of historical data and difficulties in intelligent recommendation convergence for new projects, this invention utilizes a "horizontal gene transfer" mechanism to achieve cross-project experience reuse. By comparing the genetic similarity of business logic, the system can quickly identify mature and running "donor projects" and directly "transplant" their perfectly refined game strategy parameters to new projects. This allows newly launched services to receive near-expert-level operational configuration recommendations on day one, eliminating the need for lengthy adjustment and trial-and-error periods, greatly improving the reliability and efficiency of new service delivery. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the main flow of the method of the present invention; Figure 2 This is a system architecture and module interaction diagram for the operation of this invention; Figure 3 This is a schematic diagram illustrating the principle of service gene sequence encoding and alignment in this invention; Figure 4 This is a logic diagram of gene dynamic feature extraction and environmental perception in this invention; Figure 5 This is a schematic diagram of the dynamic game rule reconstruction mechanism based on gene features of the present invention; Figure 6 This is a schematic diagram of the device structure of the present invention and a cold start optimization flowchart based on horizontal gene transfer; Figure 7 This is a hardware structure block diagram of the electronic device of the present invention. Detailed Implementation

[0024] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] Please see the appendix Figures 1 to 7 This invention provides a method for precise operation and maintenance quantification and intelligent recommendation based on project dimensions, including the following steps: Step S1: Collect distributed link tracing data from the cloud infrastructure and encode it into an ordered service gene sequence.

[0026] In a cloud-native microservice architecture, every user request or internally triggered task generates a complete call trace in the distributed tracing system. This step collects this trace data in real time using probes (Agents) or the data plane of the service mesh pre-configured in each microservice node. The collected data fields include Trace ID, Span ID, parent Span ID, service name, interface name, call duration, and operation type.

[0027] To transform unstructured trace data into a mathematical form suitable for algorithm processing, the system first cleans the raw data, removing disconnected data caused by network jitter. Then, linearization processing logic is executed. For the directed acyclic graph (DAG) structure common in microservice calls, a topological sorting algorithm is used to convert the spans of parallel calls into linear sequences, ensuring the uniqueness and comparability of the sequence structures generated by the same business logic at different execution times.

[0028] Each service interaction after linearization is defined as a "base". For the t-th position in the sequence, its feature vector is extracted. As the value of a base. Eigenvector The mathematical expression is as follows: ; in, A unique identifier for a microservice node. Sign the API method to be called. For operation type enumeration values ​​(including HTTP, RPC, DB, MQ, etc.). A consistent hash function is used to map string identifiers to fixed-dimensional numeric vectors for easier subsequent computation. Combining the L feature vectors arranged in chronological order constitutes a complete service gene sequence S: ; The service gene sequence S uniquely represents the execution path and logical structure of a specific business operation on the underlying infrastructure.

[0029] Step S2: Using sequence alignment algorithms, the service gene sequence is compared with the benchmark sequence in the project's benchmark gene library to determine the project's affiliation and quantify resource consumption.

[0030] The system pre-maintains a project baseline gene library, which stores a set of typical gene sequences from each known project under stable operating conditions. For the service gene sequences generated in real time... The system employs an improved Needleman-Wunsch dynamic programming algorithm, comparing it with the benchmark sequence of the k-th item in the benchmark library. Perform a full comparison.

[0031] Construct a score matrix Where N and M are respectively and The sequence length.

[0032] any element in the matrix This represents the best alignment score between the first i bases of the observed sequence and the first j bases of the reference sequence. Its recursive calculation formula is defined as: ; In the formula, This is the gap penalty factor, which is negative and is used to penalize missing or inserted operations in the sequence, ensuring that the alignment algorithm is fault-tolerant. For the matching scoring function, if the Euclidean distance between two base feature vectors is less than a preset threshold, it is considered a match and a positive score is assigned; otherwise, it is considered a mismatch and a negative score is assigned.

[0033] Calculate the observation sequence Maximum similarity score for all items k The selection makes The largest project This will be used as the item to which the sequence belongs. If the maximum score is lower than the preset anomaly threshold, the sequence will be marked as "unclassified variation" and will not be included in the regular item statistics.

[0034] After determining the project ownership, the system performs precise resource quantification. It acquires resource consumption data (including CPU time, memory allocation, and I / O operations) from all span records in the trace. These physical resource consumption values ​​are then... Accumulate to the project according to the established attribution relationship. Real-time resource account middle: ; Through this step, the system achieves a shift from a physical resource perspective to a logical business perspective, enabling it to accurately calculate the actual resource consumption ratio of each project in the shared cluster.

[0035] Step S3: Extract the gene dynamics features of the service gene sequence, including gene mutation rate and epigenetic modification tags.

[0036] While performing sequence alignment, the system analyzes the temporal variation characteristics of the gene sequence in parallel. First, it calculates the gene mutation rate. This metric reflects the stability of the project's business logic. A sliding time window is defined, and statistics are compiled within that window for items belonging to the project. The number of sequences with structural differences from the baseline sequence among all observed sequences. With the total number of sequences The ratio: ; when A significant increase indicates that the project has recently undergone code releases, configuration changes, or unexpected call chain anomalies, and is in an unstable architectural state.

[0037] Simultaneously, the system connects to the context interface of the operation and maintenance platform to generate epigenetic modification tags. This label is a multi-dimensional binary vector used to describe the external macroscopic environment at the current moment. Each bit in the vector corresponds to a specific operation and maintenance scenario mode, for example: ; in This indicates that we are currently in peak business season. This indicates that the system is in a low-load period at night. This indicates that a fault simulation exercise is underway. These labels will serve as environmental constraints for subsequent game theory models.

[0038] Step S4: Construct a multi-agent game model that includes performance, cost, and risk agents.

[0039] For each identified item, the system instantiates an independent multi-agent game environment. This environment contains three agents with independent objective functions (utility functions), which engage in non-cooperative games within a shared decision space.

[0040] Define the system's state vector x as the set of operation and maintenance configuration parameters for this project.

[0041] Performance Agent It is dedicated to improving user experience. Its utility function... It is positively correlated with the service's average response time (latency) and throughput (throughput). To prevent unrestricted resource demand, its utility function includes a slight penalty for configuration changes.

[0042] Cost-based intelligent agents (FinOps Agents) It focuses on controlling infrastructure spending. Its utility function... The cost of renting cloud computing resources corresponding to the current configuration x is negatively correlated. The agent tends to choose the minimum resource specification that can meet the basic SLA.

[0043] Risk-based intelligent agents (SRE agents) It is dedicated to ensuring system reliability. Its utility function... With system crash probability The service rejection rate is negatively correlated with the data. The agent tends to conserve redundant resources (buffer) and set conservative circuit breaker strategies to resist traffic bursts.

[0044] There is an inherent conflict of interest among these three agents: the performance agent tends to expand, the cost agent tends to shrink, and the risk agent tends to maintain the status quo or increase redundancy.

[0045] Step S5: Dynamically reconstruct the game rules based on the dynamic characteristics of the genes.

[0046] To enable operational decisions to adapt to changes in the business lifecycle and external environment, the system uses the genetic features extracted in step S3 to make real-time corrections to the game model in step S4. This process is not a simple parameter adjustment, but rather changes the payoff structure and strategy boundaries of the game.

[0047] First, using gene mutation rate Reconstruct the weights of the risk agent. Establish a mapping function so that the risk agent's bargaining power (penalty coefficient) in the game increases exponentially with the mutation rate. The revised risk utility function. Expressed as: ; in and This is a constant for adjustment. This means that when the project architecture is undergoing frequent changes, any configuration that increases the risk of system crash will result in a significant utility penalty, thereby forcing the game outcome to converge towards a conservative strategy.

[0048] Secondly, using epigenetic modification tags Lock the agent's policy space. Define constraint operators. According to the current State, directly pruning the set of optional actions for a specific agent. For example, when detected When a "network shutdown period" is indicated, the system forcibly locks the shrinking action space of the cost agent, preventing it from issuing decision signals to reduce resources, no matter how economically advantageous the action may be.

[0049] Step S6: Solve for the Nash equilibrium point and generate the recommended operation and maintenance strategy.

[0050] Under the reconstructed game rules, the system solves for the Nash equilibrium of the multi-agent system using an iterative algorithm. In each iteration, each agent selects the optimal response policy that maximizes its own post-modification utility function based on the previous policies of other agents. When all agents' policies no longer change, i.e., when steady state is reached, the current configuration vector... This is the Nash equilibrium solution.

[0051] The equilibrium solution This represents the optimal balance between performance, cost, and risk under the current business stability characteristics and external environmental constraints. The system converts this configuration vector into executable operation and maintenance instructions, which are then sent to the underlying execution engine to complete the closed-loop implementation of the operation and maintenance strategy.

[0052] This embodiment details the encoding and construction process of the service gene sequence. In cloud-native distributed systems, the calling relationships between microservices constitute the physical carrier of business logic. In order to perform mathematical calculations and quantitative analysis on this complex topology, this invention abstracts and encodes it as a "service gene sequence".

[0053] In this invention, the construction of the service gene sequence begins with the preprocessing of the original distributed link tracing data. The system acquires the Trace data stream generated by the full-link tracing system through bypass monitoring or log collection. To address potential data out-of-order or packet loss during network transmission, a sliding time window-based reassembly strategy is implemented. For any given Trace, the system collects all Span objects it contains; each Span records an independent service call. Based on the reference relationship between the Span ID and the Parent Span ID, the system reconstructs the complete call tree for that request.

[0054] For the parallel call and asynchronous processing patterns commonly found in microservice architectures, the original call tree typically presents as a directed acyclic graph (DAG) structure. To facilitate subsequent sequence alignment algorithms, this embodiment employs a deterministic topological sorting algorithm to flatten the DAG structure into a linear sequence. The specific sorting rules are as follows: First, the calls are sorted according to their logical depth. For parallel calls within the same level, a secondary sort is performed based on the lexicographical order of the called service names. This strict sorting rule ensures that the sequence structures generated by the same business logic under different times and load conditions are unique and reproducible.

[0055] In this invention, each Span node after linearization is defined as a "base" in a gene sequence. To comprehensively characterize the business attributes of this node, three key features are extracted to construct a feature vector. These three features are: a unique identifier for the microservice node; and a unique identifier for the microservice node. (Service ID), used to distinguish different service units; the signature of the invoked interface method. (Method Signature) is used to distinguish different function entry points within the same service; and operation type enumeration values. (Operation Type) is used to distinguish behavioral patterns such as HTTP requests, RPC calls, database operations, or message queue production / consumption.

[0056] To convert the discrete features of the string type described above into a numerical form that can be efficiently processed by a computer, this embodiment uses a consistent hashing algorithm for encoding and mapping. For the base at position t in the gene sequence, its feature vector... Represented as: ; in, and These are hash mapping functions for service identifiers and interface signatures, respectively, which map strings of arbitrary length to a fixed range of integers. This is a one-hot encoding or enumeration index mapping function for operation types. Through this encoding method, complex text information is compressed into compact numerical vectors, significantly reducing storage overhead and laying a mathematical foundation for subsequent calculations of Euclidean distance or cosine similarity between vectors.

[0057] Ultimately, a complete business request is transformed into an ordered set of feature vectors, i.e., the service gene sequence S: ; Here, L represents the total step size of the call. The sequence S fully records the flow path and behavioral characteristics of business traffic within the underlying infrastructure. Just as biological genes determine trait expression, the service gene sequence determines the consumption pattern of physical resources by the business. This encoding process achieves a crucial transformation from unstructured log data to a structured mathematical model, forming the data foundation for accurate quantification and intelligent recommendation in this invention.

[0058] This embodiment elaborates on a sequence alignment and resource quantification algorithm based on dynamic programming. In cloud-native environments, the correspondence between resources and business projects often becomes blurred due to the reuse and dynamic orchestration of microservices. To achieve accurate resource attribution, this invention draws on the sequence alignment concept from bioinformatics and utilizes an improved Needleman-Wunsch dynamic programming algorithm to calculate the homology between real-time observed service gene sequences and preset project baseline sequences.

[0059] In this invention, a baseline gene library for the project is first constructed. This library contains several known projects. A representative set of benchmark sequences. For each real-time observation sequence to be identified. The system compares them with the baseline sequences of each project. Perform a full comparison. Assume the observation sequence... The length of the reference sequence is N. The length is M, and the core of the alignment process lies in constructing a matrix of size M. The score matrix F. Each element in the matrix... The similarity score represents the cumulative similarity score after the first i bases of the observed sequence are optimally aligned with the first j bases of the reference sequence.

[0060] The filling of this score matrix follows the recursive optimal substructure property. For any position Its score It depends on the maximum value from three sources: direct match / mismatch, gap introduced by the observed sequence, or gap introduced by the reference sequence. The specific recursive calculation formula is as follows: ; in, Let be the base matching scoring function. If the feature vector of the i-th base in the observed sequence... Eigenvector of the j-th base of the reference sequence If the similarity is higher than a preset threshold, it is considered a match, and the function returns a positive reward score. Conversely, if the match is not found, it is considered a mismatch and a negative penalty score is returned. . This is the vacancy penalty factor, which takes a negative value. The vacancy penalty mechanism is introduced to handle the possible absence or insertion of non-critical nodes in actual calls, thereby giving the algorithm the ability to tolerate minor structural variations.

[0061] After the matrix is ​​filled, the element in the bottom right corner of the matrix This is the globally optimal alignment score between the two sequences. This invention calculates the observed sequences... Maximum score for all project benchmark sequences Select the item that maximizes the score. As the final result of the attribution determination: ; If the maximum comparison score of all items is lower than the set anomaly threshold, the observed sequence is marked as a new unclassified variant and is not assigned a classification, thereby avoiding misclassification from interfering with resource statistics.

[0062] After determining the observation sequence Project ownership Subsequently, the system performs precise quantification of resource consumption. The system iterates through the Span data corresponding to each base in the sequence, extracts the recorded physical resource consumption index, and denotes it as... (Including CPU usage time, memory allocation, etc.). These resource consumptions are no longer considered as general infrastructure costs, but are explicitly added to the project. Dedicated resource account middle: ; Through the above steps, this invention achieves the transformation of resource billing from the "physical machine / container" dimension to the "business logic / project" dimension. Even if multiple projects share the same microservice container, this algorithm can accurately "cut" the resource consumption of each call and record it under the correct project name based on the genetic sequence characteristics of each call, completely solving the problem of unclear accounting in a shared resource environment.

[0063] This embodiment details the implementation methods for gene dynamic feature extraction and environmental perception. Based on accurate project attribution identification, to support subsequent intelligent decision-making, this invention further mines the evolutionary patterns of service gene sequences over time and the contextual information of the external environment, extracting gene mutation rates and epigenetic modification tags as key feature inputs.

[0064] In this invention, gene mutation rate Defined as a quantitative metric for measuring the stability of a specific project's business logic. In real-world operation and maintenance scenarios, frequent code releases, configuration changes, or instability in downstream dependencies can all lead to changes in the call chain structure, which manifests as a "mismatch" with the baseline sequence at the sequence level. To capture this dynamism, the system sets a sliding time window of length W. Within this window, the system continuously counts the total number of all observed sequences belonging to project k. And those sequences whose alignment scores with the benchmark sequence are below the perfect match threshold Number of differential sequences .

[0065] Gene mutation rate The calculation formula is as follows: ; in, This is an indicator function; it takes the value 1 when the condition is met, and 0 otherwise. When the value is close to 0, it indicates that the project's operating logic is highly stable, that is, in a "steady-state period"; when A significant increase indicates that the project is undergoing drastic structural adjustments or abnormal fluctuations, i.e., in a "transitional period." This indicator provides a direct mathematical basis for subsequent game theory models to assess the risk status of the system.

[0066] At the same time, this invention introduces epigenetic modification tags. This system is used to describe the external macro-level operational environment. Similar to epigenetic modifications in biology that affect gene expression without altering the gene sequence, the external operational environment does not change the code logic but significantly constrains the feasible space of operational strategies. The system interfaces with the enterprise's operational calendar system, monitoring and alarm platform, and release system through standard API interfaces to obtain the current environment status in real time.

[0067] Epigenetic modification tag It is constructed as a multidimensional binary feature vector: ; Each bit in the vector corresponds to a specific scene pattern. For example, This indicates peak business mode; when this bit is 1, it means that the system should prioritize throughput. The indicator shows a low-load nighttime mode, suggesting that the system can perform deep resource reclamation. The disaster recovery drill mode is indicated, suggesting that the expected fault injection may occur; This indicates a network freeze mode, implying that all unnecessary changes or operations are prohibited. (This is achieved by generating...) Vectors, in this invention, transform unstructured administrative instructions or environmental states into mathematical constraints that algorithms can understand, thereby achieving the organic integration of algorithmic decision-making and business rules.

[0068] This embodiment details the implementation mechanism of the multi-agent game model construction and dynamic rule reconstruction. To achieve adaptive operational decisions in complex cloud-native environments, this invention abandons the traditional static rule engine and instead constructs a dynamic game system comprising three non-cooperative agents. This system seeks the optimal balance between performance, cost, and risk by simulating competition and checks and balances among different stakeholders.

[0069] In this invention, for each identified item, three agents are instantiated: a performance agent and a performance agent. Cost-based intelligent agents and risk intelligence agents The system's decision variable x is defined as the resource allocation vector for the project.

[0070] Performance Intelligent Agent The goal is to maximize service quality. Its utility function... The design is such that it is negatively correlated with service response latency and positively correlated with service throughput. Specifically, it is expressed as follows: ; in, and These represent the predicted throughput and latency under the current configuration x. , These are the weighting coefficients. This is a minor penalty for configuration changes, used to prevent trivial and frequent expansions.

[0071] Cost-based intelligent agents The goal is to minimize resource overhead. Its utility function... The cost of renting cloud computing resources directly corresponding to configuration x Negative correlation: ; in, This is the penalty coefficient for violating the Service Level Agreement (SLA). This function causes the cost agent to always tend to choose the configuration with smaller resource specifications, provided that the basic SLA is met.

[0072] Risk Intelligent Agent The goal is to ensure high availability of the system. Its utility function... With system crash probability Negative correlation. In this embodiment, the average queue length of the system is estimated using the M / M / 1 model in queuing theory, and then the probability of a request being rejected or timed out is derived. : ; in, To improve system utilization, For the request arrival rate, To configure the service processing rate under x, K is the system's maximum queue capacity. The basic utility function of the risk agent is: ; In this invention, to achieve adaptive perception of the dynamic characteristics of a project, a rule reconstruction mechanism based on gene features is innovatively introduced. This mechanism is based on the aforementioned extracted gene mutation rate. and epigenetic modification tags The underlying rules of the game are adjusted dynamically in real time.

[0073] First, using gene mutation rate The utility function of the risk agent is reconstructed using a nonlinear weighted approach. A risk sensitivity adjustment function is defined for the system. The risk utility function is typically in exponential form. for: ; In the formula , The formula is a constant. The physical meaning of this formula is that when a project is in an unstable period with an extremely high mutation rate, the risk agent's sensitivity to the "collapse probability" will increase exponentially. This will lead to any strategy attempting to reduce redundant resources being severely penalized in subsequent game equilibrium solutions, forcing the entire system to tend towards a conservative strategy of "overprotection." Conversely, during a stable period, the risk weight returns to normal, allowing the cost agent to play a role in resource recovery.

[0074] Secondly, using epigenetic modification tags Policy space for each agent Apply hard constraints. Define the policy space mapping operator. For example, when In When bit is 1 (during the network closure period), the operator Forcefully slash the strategy space of the agent. Remove all action vectors that lead to resource reduction: ; In this way, the present invention realizes the biomimetic logic of "environment determines rules, rules determine strategies", ensuring that every decision of the algorithm conforms to mathematical optimality and strictly adheres to the rigid constraints of the business scenario, thus avoiding the inappropriate and aggressive actions of the pure algorithm model in special periods.

[0075] This embodiment details the implementation method of Nash equilibrium solution and policy generation process, as well as the cold start optimization mechanism based on horizontal gene transfer. After completing the construction of the game model and rule reconstruction, the core task of the system becomes finding a steady-state solution that can simultaneously satisfy the interests of all agents, i.e., the Nash equilibrium point, and converting this solution into actually executable operation and maintenance instructions.

[0076] In this invention, since the operation and maintenance configuration space is usually discrete and finite, an Iterated Best Response algorithm is used to solve for the Nash equilibrium. During algorithm initialization, the current configuration vector is set. This represents the current actual operating state of the system. In each iteration t, each agent observes the previous policy of other agents and selects the optimal response policy that maximizes its current utility function.

[0077] Specifically, for the i-th agent ( ), its first Wheel strategy The new rules are as follows: ; in, This represents the policy combination of agents other than agent i in round t. This represents the feasible policy space after being constrained by epigenetic modification tags. The system continuously monitors the magnitude of changes in the policy vectors. When the amplitude is less than the preset convergence threshold When the game reaches a steady state, output the current configuration vector. This is the final Nash equilibrium solution. This solution represents the optimal compromise between performance, cost, and risk under the current genetic stability characteristics and external environmental constraints.

[0078] To address the cold start problem—the lack of historical interaction data for newly added projects, which hinders the rapid convergence of game theory model parameters—this embodiment innovatively introduces the biological mechanism of "horizontal gene transfer." When the system identifies a completely new project... When it goes live, it does not start training its game model from scratch, but first starts the gene donor matching process.

[0079] System Extraction A small number of service gene sequences during the testing phase or early launch phase. This will be compared with all mature and operational projects in the project's benchmark gene bank. The sequence is compared with a baseline sequence. By calculating sequence similarity, a mature project with the closest business logic structure is selected as the "gene donor". ; Once a donor project is identified, the system performs a policy transfer operation. The donor project... The key parameters of the current steady-state game model are transferred to the new project through a linear mapping matrix H. Initial configuration status of the new project Set as: ; in, This is a correction vector based on the estimated traffic volume of the new project. Through this mechanism, the new project can directly inherit the optimal operation and maintenance strategy that has been validated over a long period of time in similar business scenarios. Just as bacteria acquire drug resistance through plasmid exchange, this greatly reduces the operation and maintenance risks and trial and error costs in the early stages of launching a new business.

[0080] The final generated recommendation strategy (or cold start strategy) The configuration file is formatted as a standard operation and maintenance configuration file and distributed to the cloud platform's control plane via API, completing a full closed loop from data perception and intelligent decision-making to automatic execution.

[0081] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for precise operation and maintenance quantification and intelligent recommendation based on project dimensions, characterized in that: Includes the following steps: S1: Collect distributed link tracing data in cloud infrastructure and encode it into an ordered service gene sequence; S2: Using a sequence alignment algorithm, the service gene sequence is compared with the benchmark sequence in the project benchmark gene library. Based on the alignment results, the project affiliation of the service gene sequence is determined, and the physical resource consumption is quantified based on the project affiliation. S3: While identifying the project affiliation, extract the dynamic features of the service gene sequence; The dynamic characteristics of the genes include at least the gene mutation rate, which reflects the frequency of business logic changes, and epigenetic modification tags, which reflect the state of the external operation and maintenance environment. S4: Construct a multi-agent game model for the determined ownership project, wherein the multi-agent game model contains multiple agents representing different operation and maintenance objectives; S5: Based on the dynamic characteristics of the genes, dynamically reconstruct the game rules of the multi-agent game model; S6: Solve for the Nash equilibrium point of the reconstructed multi-agent game model, and generate the strategy combination that reaches the Nash equilibrium state as the operation and maintenance recommendation strategy.

2. The method according to claim 1, characterized in that, In step S1, encoding the distributed link tracing data into an ordered service gene sequence specifically includes: Extract each service interaction from the distributed tracing data and define it as a base; extract the feature vector of the base, which includes the microservice node identifier, interface method signature, and operation type. The service gene sequence is formed by combining multiple bases in chronological order.

3. The method according to claim 1, characterized in that, In step S2, determining the project affiliation of the service gene sequence using a sequence alignment algorithm specifically includes: Construct a scoring matrix and use a dynamic programming algorithm to calculate the similarity score between the service gene sequence and the benchmark sequences of each project in the project benchmark gene library; A vacancy penalty factor is introduced during the calculation process to handle the missing or inserted non-critical nodes; The item with the highest similarity score is selected as the item to which the service gene sequence belongs; If the highest similarity score is lower than a preset threshold, the service gene sequence will be marked as an abnormal mutation or a new business and will not be assigned.

4. The method according to claim 1, characterized in that, In step S3, the gene mutation rate is extracted using the following method: Within a set time window, the frequency at which the service gene sequence belonging to a specific project exhibits structural differences or mismatches with the benchmark sequence of that project is defined as the gene mutation rate. The method for extracting the epigenetic modification tag is as follows: detect the current external operation and maintenance environment context, generate the corresponding binary feature vector as the epigenetic modification tag, and the external operation and maintenance environment context includes at least the business peak mode, the night maintenance mode and the disaster recovery drill mode.

5. The method according to claim 1, characterized in that, In step S4, the multiple agents in the multi-agent game model specifically include: Performance agent: Its utility function is negatively correlated with service response latency, and its policy space is resource expansion; Cost agent: Its utility function is negatively correlated with resource overhead, and its policy space is resource shrinkage; Risk agent: Its utility function is negatively correlated with the probability of system crash, and its policy space consists of configuring redundancy and circuit breaker threshold settings.

6. The method according to claim 5, characterized in that, In step S5, the dynamic reconstruction of the game rules of the multi-agent game model based on the dynamic characteristics of the genes specifically includes: Establish a mapping relationship between the gene mutation rate and the discourse power weight of the risk agent; When an increase in the gene mutation rate is detected, the penalty coefficient weight of the risk agent in the multi-agent game model is increased by a monotonically increasing function, so that the game balance is tilted towards stability.

7. The method according to claim 5, characterized in that, In step S5, the dynamic reconstruction of the game rules of the multi-agent game model based on the dynamic characteristics of the genes further includes: Establish a constraint mapping relationship between the epigenetic modification tags and the feasible policy space of each agent; Based on the current epigenetic modification tags, lock or release the policy space of a specific agent; When the epigenetic modification tag indicates a specific business model, the shrinkage operation is forcibly removed from the policy space of the cost agent.

8. The method according to claim 1, characterized in that, In step S6, solving for the Nash equilibrium point of the reconstructed multi-agent game model specifically includes: An iterative optimal response algorithm is used to simulate the multi-round strategy selection of each agent under the reconstructed game rules. When the policies of each agent no longer change with the policies of other agents and the Nash equilibrium condition is met, the current configuration vector is output as the operation and maintenance recommendation policy.

9. The method according to claim 1, characterized in that, The method also includes a cold-start optimization step based on horizontal gene transfer: When a new project is introduced and historical interaction data is lacking, a sequence alignment algorithm is used to search the historical project database for the mature project with the highest similarity to the initial service gene sequence of the new project as the gene donor. Through a horizontal gene transfer mechanism, the current Nash equilibrium strategy parameters of the gene donor are mapped to the initial state of the multi-agent game model of the new project.

10. The method according to claim 1, characterized in that, In step S2, the quantification of physical resource consumption specifically includes: After determining the project affiliation of the service gene sequence, obtain the actual CPU time slice and memory usage of each microservice node in the sequence; The CPU time slices and memory usage are accumulated into the resource accounts of the corresponding projects according to the determined project affiliation, forming a resource consumption vector at the project level.

Images