Container operation method based on artificial intelligence

Abstract

The invention discloses an artificial intelligence-based container operation method, which comprises the following steps of: scanning a basic mirror image to generate a unique gene code, constructing a behavior portrait library, and labeling an extended mirror image by using a historical data training model; calculating the similarity between a mirror image input by a user and a portrait library gene, screening a high-similarity mirror image, evaluating a combination success rate, and sorting and recommending after removing a conflicting mirror image; during distribution, the mirror images are fragmented, verification information is generated, node reputation scores are calculated, high-reputation nodes are selected for parallel transmission, and integrity is verified after target nodes are recombined; according to the scheme, accurate recommendation and conflict pre-judgment of the mirror image combination are realized, efficient and complete distribution is ensured, the intelligent degree of mirror image management is improved, the deployment failure rate is reduced, resource utilization is optimized, and a stable and extensible mirror image management scheme is provided for a containerized environment.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of computer software, and in particular to a container running method based on artificial intelligence. BACKGROUND

[0002] In the environment where the container technology is widely applied, the management and distribution of the images are faced with many challenges. The traditional container image management mode lacks effective extraction and utilization of deep features of the images, and it is difficult to accurately grasp the compatibility and functional association among the images. In the image deployment process, deployment failure is often caused by problems such as dependency conflicts and version mismatches, and it is difficult to predict and avoid in advance. At the same time, the image distribution lacks an intelligent scheduling mechanism, and the optimal transmission path cannot be dynamically selected according to the node state, so network congestion and transmission failure are likely to occur, which affects the efficiency and stability of the container running and is difficult to meet the needs of large-scale containerized applications.

[0003] With the continuous development of the container technology, the number of container images is increasing explosively, and higher requirements are put forward for accurate recommendation and efficient distribution of the images. In the prior art, the image recommendation is mostly based on simple rules, and it is difficult to comprehensively consider factors such as the functions, compatibility and conflicts of the images, so the success rate of the recommended image combination is low, and it is difficult to meet the diversified needs of users. In the image distribution link, the traditional method lacks an effective integrity checking mechanism, and it is difficult to ensure the accuracy of the image data in the distribution process. Moreover, the selection of the distribution node lacks scientific evaluation, and the distribution strategy cannot be dynamically adjusted according to the actual performance of the node, so the distribution efficiency is low, and the application of the container technology in complex scenarios is restricted. SUMMARY

[0004] The application provides a container running method based on artificial intelligence, realizes accurate recommendation and conflict prediction of the image combination, ensures efficient and complete distribution, improves the intelligent degree of image management, reduces the deployment failure rate and optimizes the resource utilization.

[0005] The application provides a container running method based on artificial intelligence, which comprises the following steps: S1, scanning the hierarchical hash and metadata of a basic image to generate a unique gene code, and storing the unique gene code into an extended image behavior portrait library; S2, extracting image combination features based on historical deployment extension records, labeling the extended images through a real-time compatibility prediction model and dynamically updating a label library; wherein the labels comprise function, compatibility and conflict labels; S3, calculating the gene similarity between a user input image and the portrait library, screening high-similarity images, evaluating the combination success rate by using a prediction model, eliminating images with known conflicts according to the conflict records in the label library and sorting according to the combination success rate; S4, according to the user demand quantity, intercept Top-N mirror from the sorted candidate list as the recommended result; S5, the mirror is fragmented, the content hash and the logical position coordinate are generated for each fragment, the root hash value, the total number of fragments and the target node list are summarized as the distribution check basis; the node reputation score is calculated according to the historical transmission success rate and response speed; S6, the high reputation node is preferentially selected as the initial source, the fragments are transmitted in parallel and the progress is monitored in real time, and the target node is reorganized according to the coordinates and the integrity is checked through the root hash.

[0006] Preferably, the dynamic updating of the label library comprises the following steps: specifically, extracting image combination information from all historical deployment records, including base image, extended image, deployment environment parameters and final deployment result, generating a unique gene code for each base image; annotating functional tags according to the image use through keyword matching, recording the mapping relationship between the base image gene code and the extended image for the successfully deployed combination, and forming a compatibility knowledge base; extracting structured data as auxiliary features; training an initial model using historical data, inputting image functions, gene codes and structured data combination features, and outputting compatibility scores; the results are fed back to the model after each new deployment, the compatibility association between the current gene code and the extended image is strengthened for successful cases, and the conflict feature library is updated for failed cases, and the prediction weight of the model for similar combinations is adjusted; when a new image is stored, the pre-defined function classification is automatically matched according to its description or metadata, and the base image gene code tag is added to the extended image when the model prediction success probability exceeds the threshold; when the model prediction failure probability is high, the conflict feature library is automatically associated to generate specific conflict tags and record the conflict solution.

[0007] Preferably, the unique gene code is generated by the following steps: starting from the bottom layer, the file system is scanned layer by layer, and the hash value of each layer is connected in turn to form a hierarchical hash chain that can uniquely identify the file system state of each layer of the base image; key structured data is extracted from the base image, including but not limited to operating system, architecture information, dependent library version and environment variables; all environment variables and their corresponding values set in the image are obtained; the generated hierarchical hash chain and the extracted structured metadata are spliced in a certain order; the spliced data string is processed using the SHA-256 algorithm; the SHA-256 algorithm converts an input data string of any length into a fixed length hash value, which is the gene code vector of the base image.

[0008] Preferably, S5 specifically includes: setting a fixed fragment size of 128 KB, if the last fragment is less than 128 KB, processing according to the actual size; if the mirror has been fragmented, checking whether the fragmentation strategy is consistent, otherwise re-fragmenting; starting from the head of the mirror file, cutting at intervals of 128 KB to generate a continuous fragment sequence, and assigning a unique logical position coordinate to each fragment; calculating the hash value of each fragment, splicing all fragment hashes in a fixed order, and calculating the top-level "root hash"; recording the core information root hash value, total number of fragments, and target node list, generating a mapping table of fragments and logical positions, and packaging as a verification metadata file.

[0009] Preferably, the generation of the mapping table of fragments and logical positions and the packaging as a verification metadata file includes: obtaining transmission task data of the target node in the past 24 hours from the monitoring system, including: success transmission times success_count, failure times fail_count, and average response time avg_latency; performing normalized calculation and weighted score on the success rate and response speed, and eliminating nodes with a reputation score below a threshold; if the node appears abnormal during transmission, temporarily reducing its reputation score and triggering re-calculation. The target node list is sorted in descending order of reputation score; selecting Top-K nodes for redundant distribution; if the high-reputation node load is too high, distributing part of the fragments to sub-optimal nodes in proportion.

[0010] Preferably, S5 further includes: S51, collecting the original data of each fragment, and calculating the entropy value by weighted summation; S52, comprehensively diagnosing and positioning the abnormal associated root cause from five dimensions, including file, resource, network, security, and dependency; S53, determining abnormal fragments and suspected abnormal fragments according to the associated root cause positioned by the five-dimensional diagnosis; S54, performing multi-target decision value evaluation on the suspected abnormal fragments, and introducing quantum lock enhancement technology to implement differentiated locking strategies in different scenarios; S55, constructing a dynamic gravitational model to simulate the gravitational effect of fragments in the distributed cluster; S56, dynamically adjusting the transmission path of the fragments based on real-time network topology information.

[0011] Preferably, the calculation of the entropy value includes: the core data is the original data of the fragment access frequency, time value, and associated dependency. The time value is calculated using an exponential decay model, and the formula is: wherein, is the decay coefficient, is the time since the last update; the correlation dependency is recorded by static code analysis and runtime tracking, and the dependency relationship between the shard and other shards, containers, and services is quantified as direct dependency and indirect dependency, and then the comprehensive score of the dependency degree = direct dependency times x 0.7 + indirect dependency depth x 0.3; the entropy value index of the shard is generated by comprehensively considering the three parameters, and the formula is: , wherein F is the access frequency normalized value, V is the time value normalized value, D is the correlation dependency normalized value, , , are weight coefficients, and the default values are 0.4, 0.3, and 0.3, respectively.

[0012] Preferably, the positioning abnormality is associated with a root cause, specifically: the five-dimensional abnormal events are sorted by timestamp, and the associated abnormalities occurring at the same time are identified; the abnormality association rule is preset, and the diagnostic result is automatically matched; the domain knowledge base is called for root cause reasoning in a complex scenario, the abnormal condition is reproduced in a test environment, and it is confirmed whether the same problem is triggered; repair measures are taken according to the root cause, and it is observed whether the abnormality is eliminated; the abnormality types, influence range and severity found in the five-dimensional diagnosis are summarized, and the abnormality association path is displayed in the form of a time axis and a dependency graph.

[0013] Preferably, in the S51, the entropy value is calculated, and further comprising: extracting shard multidimensional data features to generate entropy flow fingerprints that are unique and reflect the state changes of the shard; constructing a space-time association network to associate the states of the shard at different times and spaces, using the concept of quantum entanglement to simulate the dependency relationship between the shards; based on the entropy flow fingerprints and the space-time association network, the shard is restored to a state at a certain time in the past by using a time reversal algorithm.

[0014] Preferably, the shard is restored to a state at a certain time in the past by using a time reversal algorithm, specifically including: determining the target time point that needs to be traced back according to the demand of historical tracing; in the space-time association network, according to the target time point, finding the entropy flow fingerprints of the shard near the target time point; by matching with the pre-generated fingerprint library, finding the shard state fingerprint closest to the target time point; based on the matched shard state fingerprint and the space-time association network, using the time reversal algorithm, starting from the target time point, gradually deducing the historical state of the shard; the time reversal algorithm considers the association relationship and state transition rule between the shards, and restores the state of the shard before the target time point through iterative calculation; according to the reversed shard state, the corresponding business data and running indicators are obtained from the historical data storage, and the complete restoration of the historical state of the shard is realized.

[0015] One or more technical solutions provided in the application have at least the following technical effects or advantages: By generating a unique gene code of the base image and constructing a behavior image library, training a real-time compatibility prediction model combined with historical deployment data, precise recommendation and conflict prediction of image combination are realized; by adopting a sharding transmission and dynamic reputation node scheduling mechanism, the efficiency and integrity of image distribution are ensured, and the data reliability is guaranteed through root hash verification. The scheme significantly improves the intelligent level of image management, reduces the deployment failure rate, optimizes the resource utilization rate, and provides a stable and scalable image life cycle management solution for containerized environments.

[0016] By collecting multi-dimensional data of shards to calculate entropy values for accurate positioning of abnormalities, the root cause is quickly locked through five-dimensional diagnosis, and abnormal and suspected abnormal shards are processed in stages, and after multi-target evaluation of suspected abnormal shards, the quantum lock enhancement technology is introduced to realize differential locking, combined with a dynamic gravity model and path adjustment based on real-time topology, the distribution and communication of shards in a distributed cluster are effectively optimized, abnormalities can be quickly and accurately processed, the impact of failures is reduced, the stability and resource utilization of the cluster are improved, and business continuity is ensured, while the quantum lock technology enhances security and operational flexibility.

[0017] By intelligent feature selection, an entropy flow fingerprint that can accurately reflect the state change of the container shard is generated to ensure uniqueness and stability; by using artificial intelligence to simulate and quantify the space-time correlation of container shards, an intuitive network topology is constructed; with the help of intelligent matching and time reversal algorithm, the container shard can be quickly and accurately restored to a specific time state in the past, and the historical state can be completely restored, effectively improving the efficiency and accuracy of container running state tracing and fault troubleshooting. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 FIG. 1 is a flowchart of a container running method based on artificial intelligence according to an embodiment of the present application. DETAILED DESCRIPTION

[0019] In order to facilitate the understanding of the present application, the present application will be described more fully below with reference to the related drawings; the preferred embodiments of the present application are shown in the drawings, but the present application can be implemented in many different forms and is not limited to the embodiments described herein; on the contrary, the purpose of providing these embodiments is to make the disclosure of the present application more thorough and comprehensive.

[0020] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs; the terms used herein in the specification of the present application are only for the purpose of describing the specific embodiments and are not intended to limit the present application; the term "and / or" used herein includes any and all combinations of one or more related listed items.

[0021] Embodiment one: Figure 1 is a flowchart of a container running method based on artificial intelligence according to an embodiment of the present application.

[0022] As shown in Figure 1 , a container running method based on artificial intelligence comprises the following steps: S1, scanning the hierarchical hash and metadata of the base image to generate a unique genetic code and storing it in the extended image behavior portrait library.

[0023] Specifically, starting from the bottom layer, the file system is scanned layer by layer. For each layer of scanned files and directories, a recursive hash calculation is performed. First, the hash value of each file in the directory is calculated, and then the hash values of these files and the relevant information of the directory itself (such as permissions, timestamps, etc.) are combined to calculate the hash value of the directory. According to the order from top to bottom or from bottom to top, the hash value of each layer is connected in turn to form a complete hierarchical hash chain that can uniquely identify the file system state of each layer of the base image.

[0024] Extract key structured data from the base image, including but not limited to operating system (OS) type, clearly indicating whether the image is based on Linux, Windows, or other operating systems; architecture information, such as x86, ARM, etc.; dependency library version, recording the specific version number of various dependency libraries used in the image; environment variables, obtaining all environment variables and their corresponding values set in the image.

[0025] The hierarchical hash chain generated earlier and the extracted structured metadata are spliced in a certain order. For example, first the hierarchical hash chain and then the metadata, or according to actual needs to define a specific splicing rule to combine the two into a complete data string. The SHA-256 algorithm is used to process the spliced data string. The SHA-256 algorithm will convert any length of input data string into a fixed length (64 characters) hash value, which is the genetic code vector of the base image (this vector has uniqueness and can be used as a unique identifier for the base image).

[0026] The generated genetic code vector is stored in a specially designed extended image behavior portrait library. When storing, the corresponding index and association relationship need to be established, so that the genetic code information of the base image can be quickly and accurately queried and retrieved later.

[0027] Obtain relevant log files from the mirror repository (log files record various operation information in the mirror repository, including deployment records of extended images). Analyze the obtained mirror repository logs in detail, and filter out all deployment records of extended images from them. Each deployment record should include basic image information (such as basic image A), extended image information (such as extended image X), and deployment results (success or failure). Store the extracted extended images and their corresponding tag information in a time series database. The time series database can record data changes in chronological order, and is suitable for storing deployment record information with time attributes. When storing, ensure that the timestamp of the record is accurate and correct, so that the deployment situation and usage state of the extended image at different times can be clearly tracked.

[0028] S2, based on the historical deployment extension record, extract the image combination features, add function, compatibility and conflict labels to the extended image through the real-time compatibility prediction model, and dynamically update the label library.

[0029] Among them, the function, compatibility and conflict labels include: function labels, according to image function classification (such as web server, database). Compatibility labels, record the successfully combined basic image gene code (such as basic image A_Gene:a1b2c3...). Conflict labels, record the dependent conflict that leads to failure (such as glibc version incompatibility).

[0030] Specifically, extract image combination information from all historical deployment records, including basic images, extended images, deployment environment parameters and final deployment results (success / failure). Filter out invalid or duplicate records, and unify the image naming specification to generate a unique gene code for each basic image (refer to the process of generating a unique gene code in step S1).

[0031] According to the purpose of the image, the function label is marked by keyword matching. For successfully deployed combinations, record the mapping relationship between the basic image gene code and the extended image, form a compatibility knowledge base, and extract structured data as auxiliary features. For failed deployment combinations, analyze the logs to locate the conflict reasons, record the conflict types and the involved image and version information.

[0032] Use historical data to train the initial model, the input is the image combination features (function + gene code + structured data), and the output is the compatibility score (success probability). After each new deployment, the results are fed back to the model, the compatibility association between the current gene code and the extended image is strengthened for successful cases, and the conflict feature library is updated for failed cases, and the prediction weight of the model for similar combinations is adjusted.

[0033] When a new image is stored, it is automatically matched with a predefined function classification according to its description or metadata. When the success probability of the model prediction exceeds the threshold, add the base image gene encoding label to the extended image. When the failure probability of the model prediction is high, automatically associate the conflict feature library to generate specific conflict labels and record the conflict resolution.

[0034] S3, calculate the genetic similarity of the user input image and the image library, filter high similarity images, use the prediction model to evaluate the combination success rate, exclude images with known conflicts according to the conflict record in the label library, and sort by combination success rate.

[0035] Specifically, receive the image information submitted by the user (such as image name, version number, function description, dependency list, etc.), extract the key feature field, parse the keywords in the function description, and map them to the predefined function classification label, and extract the dependency library and its version range to generate a dependency feature vector.

[0036] Load all images with annotated gene encoding from the label library (each image contains gene encoding, function label, and compatibility / conflict record), and compare the dependency feature vector of the user image with the dependency feature vector of the image in the image library one by one. Calculate the overlap rate of the user image and the image library, and evaluate the version range compatibility (such as the user requires glibc>=2.28, and the image uses glibc-2.30, which scores high) If the function label matches, give an additional similarity bonus. Calculate the total similarity score according to the preset weight (such as dependency matching degree 60%, version compatibility 30%, function label 10%).

[0037] Keep the images with a similarity score higher than the preset threshold (defined in this application as 70 points), and generate a candidate list. If there are functionally redundant images in the candidate list, only keep the top 1-2 images with the highest similarity to avoid redundancy. For each candidate image, combine the features of the user input image and the candidate image to generate the input vector (gene encoding pair) required by the prediction model, and add the environmental context (such as the operating system version and hardware architecture of the target deployment environment). Call the real-time compatibility prediction model to output the combination success probability value, and record the prediction confidence if the model supports uncertainty estimation.

[0038] For each candidate image, check if there is a conflict record in the label library with the user input image (such as "glibc-2.28 in the user image and openssl-1.1.0 in the candidate image have version conflicts"). If the dependency library of the candidate image and the dependency library of the user image have known conflict chains (such as A depends on B, B conflicts with C, and the user image uses C), if such a situation exists, remove all candidate images with direct or indirect conflicts, and update the recommendation list.

[0039] Sort the candidate list by success rate from high to low. If the success rate is the same, recommend the image with higher similarity first. If there are sub-differences in the function tags (such as "time series database" vs "relational database"), recommend the one that fits better first. If there are fixable conflicts, arrange them in descending order of repair complexity.

[0040] Generate a detailed report for each recommended image, including: combination success rate and confidence. Similarity score and key matching features (such as "dependency library overlap rate 90%"). If there are conflicts, list the conflict details and suggest solutions (such as "upgrade the user's glibc to 2.30").

[0041] S4, according to the user's demand quantity, cut Top-N images from the sorted candidate list as the recommended results.

[0042] Where N is the user-specified quantity or the system default threshold. It should be noted that the user-specified quantity N is a positive integer.

[0043] Specifically, determine whether the user has explicitly specified the number of recommendations (such as requesting "recommend 5 images"). If not, use the system default threshold (default to recommend 3 images). If N exceeds the total length of the candidate list, automatically adjust it to the maximum available number of the candidate list (such as the user requires 10, but the candidate list only has 7, then take 7).

[0044] Get the candidate image list sorted by combination success rate from the output of the previous stage (S3). Confirm that each image in the list contains (image identification (name, version number), combination success rate and confidence, similarity score and function tag, conflict record (if any), list status check) information. If the candidate list is empty (such as all images are excluded by conflicts), trigger an exception handling process (such as returning "no available recommendations" or suggesting the user adjust the input conditions).

[0045] According to the user-specified N or default threshold, take the first N images from the sorted list. For example, if the sorted list is [image A (95%), image B (90%), image C (85%),...], and the user requires N=2, then take image A and image B.

[0046] If N=0, return an empty list and prompt the user to re-enter a valid number. If N=1, directly return the first ranked image, no further processing is needed. If the user's demand quantity changes (such as the user changes from "recommend 3" to "recommend 5"), re-execute the cutting operation.

[0047] If the user requires the recommended results to cover multiple functions (such as "1 database + 1 cache"), check the function tag distribution of the Top-N mirror. If there is function repetition (such as the first three are all database mirrors), keep the mirror with the highest success rate and replace the repeated items with mirrors of different functions. If the diversity requirement cannot be met, mark "function repetition warning" in the results. If multiple mirrors belong to different versions of the same base mirror (such as MySQL8.0 and MySQL5.7), only keep the version with the highest success rate unless the user explicitly requires multiple version comparison.

[0048] S5, the mirror is divided into 128KB slices, and the content hash and logical position coordinates are generated for each slice. The root hash value, total number of slices, and target node list are summarized as distribution verification basis; the node reputation score is calculated based on historical transmission success rate and response speed.

[0049] Specifically, set the fixed slice size to 128KB (if the last slice is less than 128KB, process it according to the actual size), if the mirror has been sliced (such as historical slice record exists), check if the slicing strategy is consistent, otherwise re-slice.

[0050] Starting from the mirror file header, cut it at intervals of 128KB to generate a continuous slice sequence (slice 1, slice 2… slice N), assign a unique logical position coordinate to each slice (such as slice sequence number / total number of slices, for example 3 / 10 represents the 3rd slice out of 10), and temporarily store the slices in the cache or temporary directory to avoid repeated reading of the original mirror. Calculate the hash value for each slice, for example: the hash of slice 1 is H1, the hash of slice 2 is H2, and so on. Concatenate all slice hashes in a fixed order (ascending slice sequence number) to calculate the top-level "root hash" (MerkleRoot hash), which is used for global verification.

[0051] Record the core information root hash value (as the final verification identifier for the integrity of the mirror), total number of slices, target node list (get the node IP or identifier (such as node A, node B) that needs to distribute the mirror from the scheduling strategy), generate a mapping table of slices and logical positions (such as slice 3→3 / 10) for positioning during transmission. Package the root hash, total number of slices, target node list, and slice index table into a verification metadata file.

[0052] Get the transmission task data of target nodes in the past 24 hours from the monitoring system, including: success count, fail count, average latency (unit: ms). Define success rate weight (70%), response speed weight (30%). Normalize success rate and response speed, calculate weighted score: reputation score = (0.7 × normalized_success) + (0.3 × normalized_speed). Example: node A success rate is 90%, average delay is 200 ms (threshold is 500 ms), then the score is 0.7 × 0.9 + 0.3 × 0.6 = 0.81.

[0053] Remove nodes with reputation score below threshold (0.5) to avoid selecting unreliable targets. If a node has an exception during transmission, temporarily reduce its reputation score and trigger recalculation. Sort the target node list from high to low according to reputation score. Select Top-K nodes (K = 2) for redundant distribution to improve fault tolerance. If the high-reputation node is overloaded, distribute some shards to sub-optimal nodes in proportion.

[0054] Split the shard index table by node (e.g. node A is responsible for shards 1-5, node B is responsible for shards 6-10), generate independent transmission tasks for each node, and attach verification information (hash value of target shard). Limit the number of concurrent transmissions per node to avoid network congestion.

[0055] After the node receives the shard, it immediately calculates the local hash and compares it with the shard hash in the metadata. If they do not match, request retransmission of the shard. After all shards are transmitted, the node reorganizes the image according to the shard index table, calculates the root hash and compares it with the value in the verification package to ensure integrity.

[0056] Trigger up to 3 retries for failed shards, each time selecting a sub-optimal reputation node. If all nodes fail, pause distribution and report an alarm, record the failure reason. Record the transmission status of each shard (success / failure), node selection, and hash verification result. Example log entry: shard 3 → node A → success → hash H1 matches. After each transmission is completed, dynamically adjust the node reputation score based on the result (e.g. success + 0.01 points, failure - 0.05 points).

[0057] S6, preferentially select high-reputation nodes as initial sources, transmit shards in parallel and monitor progress in real time, and target nodes reorganize according to coordinates and verify integrity through root hash.

[0058] Specifically, the latest calculated node reputation score is obtained from the S5 stage, and the reputation score weight is dynamically adjusted in combination with real-time monitoring data. The candidate node list is sorted from high to low according to the reputation score, and nodes with a reputation score lower than a threshold (0.6) are removed to avoid unreliable transmission. If there are multiple high-reputation nodes, the node with a close network topology distance to the target node (such as the same region or ISP) is preferentially selected. The Top-1 reputation node is selected as the initial transmission source (such as node A), and the Top-2 to Top-3 nodes are selected as backup sources for quick switching in the event of a main source failure. Example: if the main source node A fails to transmit, automatically switch to node B to continue transmitting the remaining shards.

[0059] The total shards are split according to node capabilities (such as node A is responsible for shards 1-5, and node B is responsible for shards 6-10), high-reputation nodes are preferentially allocated more shards, but the load of a single node is avoided to be too high (such as a single node can bear at most 60% of the shards). Transmission instructions are sent to each selected source node, including: target node list (nodes IP that need to receive the shard). Shards logical coordinates (such as shard 3 / 10) and content hash (used for verification).

[0060] The source node simultaneously pushes the shard to multiple target nodes (such as node A parallelly transmits shard 1 to nodes X, Y), and each transmission thread independently records the state (such as the number of bytes sent, transmission rate). The number of concurrent transmissions per node is dynamically adjusted, and the network bandwidth feedback (such as packet loss rate) is used to automatically limit the speed to avoid congestion.

[0061] The target node sends transmission progress to the control center every 5 seconds, and the progress data includes: current shard number and total shard number, transmitted byte number and total size, instantaneous transmission rate (KB / s). If the shard transmission is interrupted (such as 3 consecutive heartbeat timeouts), an alarm is immediately triggered and marked as "failed". Example alarm: node X fails to transmit shard 2, error code: NETWORK_TIMEOUT.

[0062] The transmission status of each shard is displayed in the form of a progress bar (such as "shard 4 / 10→ node Y→ 65%"), and abnormal shards are highlighted. If the transmission progress of a certain shard is seriously lagging (such as lower than 30% of the average progress), it is automatically switched from the slow node to the backup source node.

[0063] The target node temporarily stores the received fragments in a cache directory according to logical coordinates (such as fragment 1 / 10), for example, the cache path: / tmp / mirror_chunks / nodeX / chunk_3_of_10. The local hash of each fragment is calculated and compared with the hash value in the transmission instruction. If they do not match, the control center is requested to retransmit the fragment (and the reputation score of the corresponding source node is reduced). According to the fragment index table (such as fragment 1→ fragment 2→…→ fragment N), the fragments are sequentially spliced into a complete mirror file. For example, the merging command (logical description): catchunk_1chunk_2...chunk_10>final_image.iso. If a fragment fails to be transmitted and cannot be retransmitted, the entire mirror is marked as "incomplete" and the recombination is terminated.

[0064] The top-level hash of the recombined complete mirror file is calculated to generate a local root hash (local_root_hash). The local_root_hash is strictly compared with the root_hash in the verification metadata generated in the S5 stage. If local_root_hash=="A1B2C3..." (metadata value), the verification is passed. Otherwise, an image damage alarm is triggered and the difference hash value is recorded.

[0065] If the verification is passed, the control center is returned with the "image complete" state, and the temporary cache fragments are released. For example, the success log: node X mirror recombination is successful, the root hash matches, and the time consumption is 12 seconds. If the verification fails, the mirror is marked as "unavailable", the missing or damaged fragments are retransmitted from the backup source node, and if multiple retries still fail, a serious fault is reported and manual intervention is required.

[0066] The technical solutions in the embodiments of the present application have at least the following technical effects or advantages: By generating a unique genetic code of the base image and constructing a behavior portrait library, combining historical deployment data to train a real-time compatibility prediction model, precise recommendation and conflict prediction of image combinations are realized; the fragment transmission and dynamic reputation node scheduling mechanism are adopted to ensure the efficiency and integrity of image distribution, and the root hash verification is used to ensure data reliability. This scheme significantly improves the intelligent level of image management, reduces the deployment failure rate, optimizes the resource utilization rate, and provides a stable and scalable image lifecycle management solution for containerized environments.

[0067] Embodiment Two: In Embodiment One, a unified strategy is adopted for abnormal positioning and processing when a suspected abnormal condition occurs in a distributed cluster shard. However, different shards have significant differences in terms of their roles in the cluster, business criticality, traffic impact, and dependency relationships, and the abnormal performance and root cause are also diverse. In the face of these complex and diverse scenarios, the unified strategy is difficult to accurately adapt, and there are deficiencies in abnormal detection accuracy, processing timeliness, and resource utilization rationality, which cannot meet the needs of efficient and stable operation of the distributed cluster. To achieve more accurate and flexible processing of shard abnormalities and improve the overall performance of the cluster, Embodiment Two is proposed for optimization and improvement.

[0068] In some embodiments, the image is divided into 128KB shards, and step S5 further includes: S51, collect the original data of each shard, and calculate the entropy value by weighted summation.

[0069] Among them, the core data is the original data of shard access frequency, time value, and correlation dependency. The time value is calculated using an exponential decay model, and the formula is: , wherein, is the decay coefficient (such as 0.1 / hour), is the time since the last update of the shard. The correlation dependency is recorded through static code analysis and runtime tracking, recording the dependency relationship between the shard and other shards, containers, and services. The dependency complexity is quantified as direct dependency (the number of times the shard is directly called by other components) and indirect dependency (the depth of the association between the shard and other components through middleware), and then the dependency score is calculated as direct dependency times 0.7 + indirect dependency depth times 0.3. By combining the three parameters, the entropy value index of the shard is generated. The formula is: , wherein F is the normalized value of access frequency, V is the normalized value of time value, and D is the normalized value of correlation dependency, , , is the weight coefficient (default values are 0.4, 0.3, and 0.3, respectively). If the entropy value exceeds 1 or is lower than 0, the normalization parameters are re-verified or the weights are adjusted. For shards with sudden changes in entropy value (such as a change in entropy value >0.5 in a single calculation), secondary diagnosis (S52 five-dimensional diagnosis) is triggered.

[0070] S52, comprehensively diagnose and locate the abnormal correlation root cause of the distributed cluster from five dimensions.

[0071] Among them, the five dimensions include files, resources, networks, security, and dependencies.

[0072] Specifically, locate cluster problems caused by storage, access or integrity exceptions of positioning files; identify performance degradation problems of clusters caused by resource bottlenecks or waste; determine communication failure problems caused by network delays, packet loss or topology problems; record cluster risk problems caused by security attacks or configuration errors; and determine service unavailability problems caused by broken dependencies between components or long call chains. Sort the five-dimensional abnormal events by timestamp, identify simultaneous correlated abnormalities (such as file access failure coinciding with network packet loss time), and automatically match the diagnostic results according to preset abnormal correlation rules (such as "network packet loss rate > 5% -> file transfer timeout"). For complex scenarios, call the domain knowledge base for root cause reasoning. Reproduce the abnormal conditions in the test environment (such as limiting the network bandwidth of a certain node to 1 Mbps), and confirm whether the same problem is triggered. Take repair measures according to the root cause, and observe whether the abnormality is eliminated. Summarize the types of abnormalities found in five-dimensional diagnosis, the scope of influence and the severity, and display the abnormal correlation path in the form of a time axis and a dependency graph.

[0073] S53, according to the correlated root cause located by the five-dimensional diagnosis, determine abnormal shards and suspected abnormal shards.

[0074] Among them, for abnormal shards, immediately perform a cleaning operation; for suspected abnormal shards, set up an automatic migration mechanism, and migrate the suspected abnormal shards to a suitable node according to the real-time load and resource distribution of the system.

[0075] Specifically, map the root cause located by the five-dimensional diagnosis to specific shards (file dimension: if the root cause is a storage medium failure of a certain node, locate all storage shards on the node; resource dimension: if the root cause is a memory leak of a certain node, mark the computing shards with high memory usage on the node; network dimension: if the root cause is a high packet loss rate of a certain subnet, identify all communication shards in the subnet; security dimension: if the root cause is an attack on a certain node, isolate all service shards on the node; dependency dimension: if the root cause is that a certain dependent service is unavailable, mark the business shards that strongly depend on the service), and identify abnormal and suspected abnormal shards. Define abnormal level determination criteria: an abnormal shard meets any of the following conditions: directly causes business interruption; more than three dimensions are abnormal simultaneously in five-dimensional diagnosis (such as network packet loss + storage failure + resource exhaustion). A suspected abnormal shard meets any of the following conditions: single-dimensional abnormality but does not affect business; dependent service response delay exceeds threshold but does not completely fail.

[0076] In the cluster metadata management system, the abnormal shard state is marked as UNHEALTHY, new request allocation to the shard is stopped, and the load balancer is notified to remove the abnormal shard node from the service pool. If the abnormal shards are concentrated on a node (such as >3 abnormal shards), the node is marked as QUARANTINE, and new shard scheduling is prohibited. The five-dimensional indicators of suspected abnormal shards are continuously tracked, and if any indicator exceeds the specified range, migration evaluation is triggered. Nodes with resource utilization below the safety threshold are screened, and nodes in the same machine room or same availability zone as the original node are preferentially selected to reduce migration delay. If the shard depends on a specific service, a node that has deployed the service is selected, and it is verified whether the target node meets the shard running conditions (such as disk space, port occupation). A quick rollback path is prepared in case of migration failure. For stateless shards, instances are started directly on new nodes and configurations are synchronized; for stateful shards, metadata is migrated first, and then data increments are synchronized. A small amount of traffic is first routed to the new shard to verify functionality, and after confirming that there are no errors, the service discovery registry is updated and all traffic is switched to the new shard. After migration is complete, the original shard is kept running for 10 minutes to monitor whether the abnormality recurs. After confirming that there are no abnormalities, the original shard process is stopped and resources are cleaned up.

[0077] S54, multi-objective decision value evaluation is performed on the suspected abnormal shard, and quantum lock enhancement technology is introduced to implement differentiated locking strategies in different scenarios.

[0078] Among them, the multi-objective decision value evaluation considers three factors: cost (migration cost and locking cost), risk (fault escalation probability and dependency chain probability), and business value (criticality and traffic impact).

[0079] Specifically, each dimension indicator is normalized to the [0, 1] interval, the weight is dynamically adjusted according to the business scenario, the comprehensive value score is calculated: score = business value weight × business value + risk weight × (1-fault escalation probability) - cost weight × migration cost, and the suspected abnormal shard is classified and processed according to the score.

[0080] A unique quantum key is generated for each suspected abnormal shard, bound to the shard metadata, and the lock state is synchronized in real time to the cluster coordinator through quantum entanglement. Define scene-specific locking strategies (e.g., enable quantum obfuscation locks in high-security scenarios, and enable quantum annealing optimization lock combinations in high-concurrency scenarios). Deploy a quantum random number generator (QRNG) at the edge nodes of the cluster to generate an initial quantum key for each container, and securely transmit the key to the control plane through a quantum key distribution protocol (QKD). Modify the container runtime to support quantum lock API calls. The container runtime sends a locking request to the quantum lock manager, carrying the shard ID and the expected locking time. The manager generates a quantum entangled pair, returns half the key to the runtime, and stores the other half in a distributed quantum storage. The runtime uses the quantum key to hash the shard state, and the result is compared with the hash stored by the manager: if they match, the shard enters the "quantum locked" state; otherwise, trigger a fallback process (e.g., fall back to a traditional mutex lock). The runtime sends an unlock request, and the manager destroys the remote key copy through quantum teleportation. Record the lock log to the quantum blockchain to ensure audit tamper-proofing. If the quantum storage fails, automatically switch to the preloaded classical key (valid for 24 hours), and trigger quantum hardware redundancy switching. During network isolation, allow shards to continue operating locally using "quantum tokens", and synchronize the state through quantum error correction codes after partition recovery.

[0081] Track quantum lock operation time, shard migration success rate, business impact duration, and other indicators, and collect hardware-level data through quantum sensors. If high-value shard locking causes business timeout rate to rise, automatically reduce its quantum lock priority and increase classical lock timeout time. If the quantum lock failure rate is <0.1%, gradually expand the quantum lock application scenarios (e.g., from database shards to microservice instances). Temporarily disable quantum locks for low-value shards during sudden traffic scenarios to release resources for high-value shards. Use quantum annealing algorithm to quickly calculate the optimal lock combination. Enable "quantum obfuscation locks" for suspected attacked shards to prevent attackers from predicting lock release time (through random quantum bit flipping). Use quantum key pools for frequently locked shards (e.g., cache shards locked more than 10 times per minute) to reduce generation overhead. Preload quantum keys for frequently used shards into a trusted execution environment (TEE) to shorten the first lock delay to <1ms.

[0082] S55, build a dynamic gravity model to simulate the gravitational effect of shards in a distributed cluster.

[0083] S56, based on real-time network topology information, dynamically adjust the transmission path of the shard.

[0084] The transmission path of the dynamic adjustment of the shard is calculated by using the Dijkstra algorithm, and the triggering conditions are divided into periodic triggering (recalculating the paths of all active shards every 1 minute) and event triggering (detecting link failure and shard traffic surge).

[0085] The technical solutions in the embodiments of the present application have at least the following technical effects or advantages: By collecting multi-dimensional data of shards to calculate entropy values for accurately locating anomalies, the root cause is quickly locked by five-dimensional diagnosis, and abnormal and suspected abnormal shards are processed in stages. After multi-target evaluation of suspected abnormal shards, the quantum lock enhancement technology is introduced to realize differential locking. Combined with the dynamic gravity model and the path adjustment based on real-time topology, the distribution and communication of shards in the distributed cluster are effectively optimized, the abnormality can be quickly and accurately processed, the impact of failure is reduced, the stability and resource utilization of the cluster are improved, the business continuity is ensured, and the security and operation flexibility are enhanced by the quantum lock technology.

[0086] Embodiment Three: In Embodiment Two, after the cleaning operation on the abnormal shard, the problem of historical traceability failure may occur. Since the shard data is cleared, there is a lack of complete historical information to support the review of system operation history, the investigation of potential problems or the satisfaction of compliance audit requirements in the subsequent process, which brings challenges to the stable operation and safe management of the distributed cluster. To solve this problem, Embodiment Three improves Embodiment Two by building a complete historical traceability system to ensure that effective historical traceability can be achieved after the shard is cleaned.

[0087] In some embodiments, in step S51, the entropy value is calculated, and the method further includes: S511, extracting multi-dimensional data features of the shard to generate an entropy flow fingerprint that uniquely identifies and reflects the state change of the shard.

[0088] Specifically, with the help of artificial intelligence data mining and analysis capabilities, data is collected from multiple dimensions of container shards, including but not limited to basic information of container shards (such as shard ID, belonging container service, node location, etc.), running indicators (such as access frequency, response time, resource occupancy, etc.), dependency relationships (dependency with other container shards, containers, services), and business data (such as task type handled, data volume, etc.). Using artificial intelligence-based feature selection algorithms, such as deep learning-based feature importance evaluation algorithms, information gain based on machine learning, chi-square test, etc., the features that have a significant impact on the state of container shards are selected from the collected data. These features should accurately reflect the running state and trend of container shards, while reducing data redundancy and improving computing efficiency. The selected features are encoded and converted into a digital form that can be processed by a computer. The encoded features are combined and encrypted using artificial intelligence encryption technology, such as neural network-based encryption algorithms, to generate entropy stream fingerprints of container shards. The entropy stream fingerprints should be unique and stable, i.e., different container shard fingerprints are different, and the same container shard fingerprints at different times can reflect the changes in its state. For example, a hash algorithm is used to encrypt the feature combination to generate a fixed-length fingerprint string.

[0089] S512, construct a spatio-temporal association network to associate the states of the shards at different times and spaces, and use the concept of quantum entanglement to simulate the dependency relationship between the shards.

[0090] Specifically, according to the historical state records of container shards, a time series of container shard states is established in chronological order. Using artificial intelligence time series analysis technology, the change relationship between container shard states at different times is analyzed to determine the state transition rules and influencing factors. For example, by analyzing the access frequency changes of container shards at different time points, the time periods of access peaks and valleys are found out, as well as the causes of these changes.

[0091] Considering the physical location and logical relationship of container shards in a distributed cluster, the spatial association between container shards is established. Using artificial intelligence relationship analysis algorithms, the dependency relationship, communication relationship, and data interaction between container shards are analyzed to determine the degree of mutual influence between container shards. For example, if the failure of a container shard will cause the running abnormalities of other multiple container shards, then there is a strong spatial association between these container shards.

[0092] By utilizing the concept of quantum entanglement, combined with artificial intelligence simulation and quantification technology, the time and space correlation between container shards is simulated and quantified. An entanglement degree index between container shards is defined to measure the degree of correlation between container shards. The entanglement degree can be calculated according to the dependence strength, communication frequency, data transmission volume, and other factors between container shards. For example, a weighted average method is used to calculate the entanglement degree between container shards, and the weights are determined by the influence of different factors on the running of container shards, with the help of artificial intelligence weight optimization algorithm.

[0093] The time-space correlation network is constructed by taking container shards as nodes and the entanglement degree between container shards as the weight of the edge. This network can intuitively show the correlation between container shards at different times and spaces, providing network topology support for subsequent historical state restoration.

[0094] S513, based on the entropy flow fingerprint and the time-space correlation network, the shards are restored to a state at a certain time in the past by using the time reversal algorithm.

[0095] Specifically, according to the demand for historical tracing, the target time point that needs to be traced back is determined. This time point can be a specific time or a time period. For example, when troubleshooting a container business fault, it is necessary to trace back to a period of time before the fault occurs to understand the running state of the system.

[0096] In the time-space correlation network, according to the target time point, the entropy flow fingerprint of the container shard near the target time point is found. By matching with the pre-generated fingerprint library, using artificial intelligence similarity matching algorithm, such as similarity measurement model based on deep learning, the container shard state fingerprint closest to the target time point is found. For example, a similarity calculation algorithm (cosine similarity) is used to compare the similarity of the target fingerprint and the fingerprints in the fingerprint library, and the fingerprint with the highest similarity is selected as the matching result.

[0097] Based on the matched container shard state fingerprint and the time-space correlation network, the time reversal algorithm is used to gradually deduce the historical state of the container shard from the target time point. The time reversal algorithm considers the correlation between container shards and the state transition law, and restores the state of the container shard before the target time point by iterative calculation. For example, according to the entanglement degree between container shards and the state transition model, the state parameters of the container shard at the previous time are calculated.

[0098] According to the container shard state obtained by inversion, the corresponding business data and running indicators are obtained from the historical data storage, and the complete restoration of the historical state of the container shard is realized. For example, the container shard processing task type and data processing amount obtained by inversion are associated with the records in the historical database to obtain the specific business data content.

[0099] The technical solutions in the embodiments of the present application have at least the following technical effects or advantages: The entropy flow fingerprint accurately reflects the change of the container fragmentation state, ensuring uniqueness and stability. The artificial intelligence simulates and quantifies the space-time correlation of the container fragmentation, constructing an intuitive network topology. With the help of intelligent matching and time reversal algorithm, the container fragmentation can be quickly and accurately restored to a specific time state in the past, fully recovering the historical state, effectively improving the efficiency and accuracy of container operation state tracing and fault troubleshooting.

[0100] The above description is only the preferred embodiments of the present application and is not intended to limit the present application. For those skilled in the art, the present application can have various modifications and changes. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A container operation method based on artificial intelligence, characterized in that, include: S1, scan the hierarchical hash and metadata of the base image, generate a unique gene code, and store it in the extended image behavior profile library; S2 extracts image combination features based on historical deployment extension records, and dynamically updates the tag library by using a real-time compatibility prediction model to label the extension images; the tags include functional, compatibility and conflict tags. S3, calculate the gene similarity between the user input image and the image database, filter high similarity images, use the prediction model to evaluate the combination success rate, remove images with known conflicts according to the conflict records in the tag database, and sort them by combination success rate. S4. Based on the number of user requests, extract the Top-N mirror images from the sorted candidate list as the recommendation results; S5: Divide the image into shards, generate content hashes and logical location coordinates for each shard, and summarize the root hash value, total number of shards, and target node list as the basis for distribution verification; calculate node reputation scores based on historical transmission success rate and response speed. S6 prioritizes high-reputation nodes as the initial source, transmits fragments in parallel and monitors progress in real time, and the target nodes reassemble according to coordinates and verify integrity through root hash.

2. The container operation method based on artificial intelligence as described in claim 1, characterized in that, The dynamically updated tag library specifically includes: extracting image combination information from all historical deployment records, including base images, extended images, deployment environment parameters, and final deployment results; generating a unique gene code for each base image; labeling functional tags based on image usage using keyword matching; recording the mapping relationship between the base image gene code and the extended image for successfully deployed combinations to form a compatibility knowledge base; extracting structured data as auxiliary features; training an initial model using historical data, inputting image functions, gene codes, and structured data combination features, and outputting a compatibility score; feeding the results back to the model after each new deployment; strengthening the compatibility association between the current gene code and the extended image for successful cases; updating the conflict feature library for failed cases and adjusting the model's prediction weights for similar combinations; when a new image is added to the library, automatically matching a predefined functional category based on its description or metadata; adding a base image gene code tag to the extended image when the model predicts a success probability exceeding a threshold; automatically associating with the conflict feature library when the model predicts a high failure probability, generating specific conflict tags, and recording conflict solutions.

3. The container operation method based on artificial intelligence as described in claim 2, characterized in that, The process of generating a unique gene code includes: performing a comprehensive scan of the file system layer by layer from the bottom up; performing hash calculations recursively on the files and directories scanned at each layer; calculating the hash value for each file within a directory; combining the file hash values ​​with relevant information about the directory itself to calculate the hash value for that directory; concatenating the hash values ​​of each layer sequentially, either from top to bottom or bottom to top, to form a complete hierarchical hash chain that uniquely identifies the file system state of each layer of the base image; extracting key structured data from the base image, including but not limited to operating system, architecture information, dependency library versions, and environment variables, and obtaining all environment variables set in the image and their corresponding values; concatenating the generated hierarchical hash chain and the extracted structured metadata in a specific order; processing the concatenated data string using the SHA-256 algorithm; and converting the input data string of arbitrary length into a fixed-length hash value, which is the gene code vector of the base image.

4. The container operation method based on artificial intelligence as described in claim 1, characterized in that, S5 specifically includes: setting a fixed shard size of 128KB; if the last shard is less than 128KB, it is processed according to the actual size; if the image has been sharded, check whether the sharding strategy is consistent, otherwise re-shard; starting from the beginning of the image file, cut it at 128KB intervals to generate a continuous shard sequence, and assign a unique logical location coordinate to each shard; calculate the hash value for each shard, concatenate all shard hashes in a fixed order, and calculate the top-level "root hash"; record the core information root hash value, the total number of shards, and the target node list, generate a mapping table between shards and logical locations, and package it into a verification metadata file.

5. The container operation method based on artificial intelligence as described in claim 4, characterized in that, The process of generating a mapping table between shards and logical locations and packaging it into a verification metadata file includes: obtaining transmission task data of target nodes in the past 24 hours from the monitoring system, including: success_count, failure_count, and average response time avg_latency; normalizing and calculating a weighted score for success rate and response speed, and removing nodes with reputation scores below a threshold; if a node experiences an anomaly during transmission, temporarily lowering its reputation score and triggering a recalculation; sorting the target node list from high to low reputation score; selecting Top-K nodes for redundant distribution; and proportionally allocating some shards to suboptimal nodes if high-reputation nodes are overloaded.

6. The container operation method based on artificial intelligence as described in claim 1, characterized in that, The S5 also includes: S51 collects the raw data of each slice and calculates the entropy value by weighted summation; S52 performs a comprehensive diagnosis of distributed clusters from five dimensions to locate the root causes of anomalies; these five dimensions include files, resources, network, security, and dependencies. S53, Based on the associated root causes located by the five-dimensional diagnosis, determine the abnormal segments and suspected abnormal segments; S54 performs multi-objective decision-making value assessment on suspected abnormal segments and introduces quantum lock enhancement technology to implement differentiated locking strategies according to different scenarios; S55, construct a dynamic gravity model to simulate the gravitational effect of sharding in a distributed cluster; S56 dynamically adjusts the transmission path of fragments based on real-time network topology information.

7. The container operation method based on artificial intelligence as described in claim 6, characterized in that, The calculation of entropy includes: core data such as shard access frequency, timeliness value, and dependency; and the calculation of timeliness value using an exponential decay model, with the formula as follows: ,in, The attenuation coefficient is... The time since the last update of the shard; dependency ratio is recorded through static code analysis and runtime tracing, recording the dependencies between the shard and other shards, containers, and services, and the dependency complexity is quantified into direct and indirect dependencies. Therefore, the comprehensive dependency score = number of direct dependencies × 0.7 + depth of indirect dependencies × 0.3; combining these three parameters, the entropy index of the shard is generated, with the formula: Where F is the normalized value of access frequency, V is the normalized value of timeliness, and D is the normalized value of association dependency. , , The default values ​​for the weighting coefficients are 0.4, 0.3, and 0.3, respectively.

8. The container operation method based on artificial intelligence as described in claim 7, characterized in that, The specific steps for locating the root cause of anomalies are as follows: sorting the five-dimensional anomaly events by timestamp and identifying concurrently occurring related anomalies; pre-setting anomaly association rules and automatically matching diagnostic results; performing root cause reasoning by calling the domain knowledge base for complex scenarios, reproducing the anomaly conditions in the test environment, and confirming whether the same problem is triggered; taking remedial measures based on the root cause and observing whether the anomaly is eliminated; summarizing the anomaly types, impact range, and severity found in the five-dimensional diagnosis, and displaying the anomaly association path in the form of a timeline and dependency graph.

9. The container operation method based on artificial intelligence as described in claim 6, characterized in that, In step S51, calculating the entropy value further includes: Extract multi-dimensional data features from the shards to generate a unique entropy flow fingerprint that reflects changes in the shard state; A spatiotemporal correlation network is constructed to associate the states of the fragments in different times and spaces, and the dependency relationship between the fragments is simulated by using the concept of quantum entanglement. Based on entropy flow fingerprinting and spatiotemporal correlation networks, the fragments are restored to their state at a certain point in the past by using a time inversion algorithm.

10. The container operation method based on artificial intelligence as described in claim 9, characterized in that, The process of restoring a shard to its state at a specific point in the past using a time-reversal algorithm includes: determining the target time point to be traced back based on the need for historical tracing; searching for the entropy flow fingerprint of shards near the target time point in a spatiotemporal correlation network; finding the shard state fingerprint closest to the target time point by matching it with a pre-generated fingerprint database; using the matched shard state fingerprint and the spatiotemporal correlation network, and applying the time-reversal algorithm to progressively deduce the historical state of the shard from the target time point; the time-reversal algorithm considers the correlation between shards and the state transition rules, and restores the state of the shard before the target time point through iterative calculation; and obtaining the corresponding business data and operational indicators from historical data storage based on the inverted shard state to achieve a complete restoration of the shard's historical state.

Images