A database migration engineering experience reuse method, device and medium

Abstract

The embodiment of the application discloses a database migration engineering experience reuse method, equipment and medium, belongs to the technical field of database, solves the problem that the same kind of abnormality repeatedly occurs in different tasks in the process of database migration, and zero start is responded, so that the diagnosis efficiency is low and the diagnosis cost is high. Including, when executing the database migration task, the operation event is collected, and the original operation event sequence is generated; after detecting the end of the task, based on the original operation event sequence, the diagnosis path reverse reconstruction processing is executed, the disposal path is generated; the disposal path is associated with the multi-dimensional scene fingerprint and the abnormal description information corresponding to the task, a structured experience knowledge unit is generated, and is stored to the experience knowledge base; when detecting the abnormality in the new database migration task execution process, based on the scene characteristics corresponding to the abnormality, multi-dimensional weighted matching is carried out in the experience knowledge base, to determine the reference disposal path.

Description

Technical Field

[0001] This application relates to the field of database technology, and in particular to a method, device and medium for reusing engineering experience in database migration. Background Technology

[0002] In the context of digital transformation, enterprise-level database migration and synchronization has become a core engineering practice characterized by high frequency and complexity. Whether it's heterogeneous migration across different database types, version upgrades, real-time synchronization, or post-event data consistency verification, all face complex technical challenges.

[0003] Currently, the industry primarily relies on automated tools to handle standardized data flow and object transformation processes. However, in real production environments, especially when encountering abnormal scenarios where tools fail to pre-program their processing logic, manual diagnosis and handling by engineers' on-site experience remains crucial. Experienced engineers typically attempt to recover tasks by manually executing SQL, adjusting task parameters, and selectively skipping conflicting objects, but these processes are often ad-hoc and exploratory.

[0004] Existing intelligent operation and maintenance systems and the expert systems built into database tools mostly focus on monitoring and alarming of continuously running systems and root cause localization of common faults, or on building static handling rules for known anomaly patterns. For dynamic handling processes in complex migration tasks that heavily rely on personal experience, there is a lack of effective automated data collection and structured data preservation mechanisms. Valuable handling experience is often scattered in engineers' personal notes. This leads to similar anomalies being repeatedly addressed from scratch in different tasks, resulting in low handling efficiency and high diagnostic costs. Summary of the Invention

[0005] This application provides a method, device, and medium for reusing engineering experience in database migration to solve the following technical problem: during database migration, similar anomalies are repeatedly dealt with from scratch in different tasks, resulting in low handling efficiency and high diagnostic costs.

[0006] The embodiments of this application adopt the following technical solutions: This application provides a method for reusing engineering experience in database migration. It includes: collecting operational events throughout the task lifecycle during database migration to generate an original operational event sequence; after the task is detected as completed, performing reverse reconstruction of diagnostic paths based on the original operational event sequence to generate a handling path; associating the handling path with the multi-dimensional scenario fingerprint and anomaly description information corresponding to the current task to generate structured experience knowledge units, which are then stored in an experience knowledge base; wherein the scenario fingerprint is a multi-dimensional feature vector defined based on the database migration scenario; when an anomaly is detected during the execution of a new database migration task, performing multi-dimensional weighted matching in the experience knowledge base based on the scenario features corresponding to the anomaly; and based on the multi-dimensional weighted matching result, executing a main path recommendation or bypass recommendation strategy to determine a reference handling path in the experience knowledge base, and outputting the reference handling path as a handling suggestion for the current anomaly.

[0007] In one implementation of this application, during the execution of a database migration task, operation events within the task lifecycle are collected to generate an original operation event sequence. Specifically, this includes: during the execution of the database migration task, operation events are collected at stage switching nodes, exception triggering nodes, manual intervention nodes, and task final state nodes to generate an original operation event sequence; a stage switching node is a node where the task undergoes a state transition at least between the pre-inspection stage, the full transfer stage, the incremental synchronization stage, the data comparison stage, the repair stage, and the completion stage; an exception triggering node is a node where the tool detects an error, alarm, or verification inconsistency; a manual intervention node is a node where the engineer performs operations including pausing or resuming the task, modifying configuration parameters, skipping specified objects, manually executing SQL commands, and calling the tool's repair function; and a task final state node is a node where the task transitions to a success, failure, or manual termination state.

[0008] In one implementation of this application, after the completion of the task is detected, a diagnostic path reverse reconstruction process is performed based on the original operation event sequence to generate a disposal path. Specifically, this includes: taking the abnormal triggering event as the starting anchor point, extracting all operation events from the anchor point to the end of the current task in the original operation event sequence to generate a candidate disposal event sequence; scanning the candidate disposal event sequence and marking operation events belonging to a preset intervention type as decision nodes; wherein, the preset intervention type includes at least one of parameter modification, object skipping, manual SQL execution, and tool repair function call; merging and recording multiple status query operations issued consecutively for the same target object within a preset time window into a single status confirmation operation, and filtering out intermediate decision operations whose operation results are covered by the operations of subsequent decision nodes to obtain a disposal action chain; and annotating the disposal action chain with disposal information based on the final state of the database migration task to obtain a disposal path; wherein, the disposal information includes at least one of the following: task success or failure status, total path execution time, and residual side effects information.

[0009] In one implementation of this application, the handling path is associated with the multi-dimensional scene fingerprint and anomaly description information corresponding to the current task to generate a structured experience knowledge unit. Specifically, this includes: extracting multi-dimensional scene feature vectors based on the overall context information of the current task to construct a multi-dimensional scene fingerprint; wherein the multi-dimensional scene fingerprint includes at least one of the following: database type combination, version range, task type, anomaly category, migration scale range, and anomaly occurrence stage; extracting anomaly description information from the anomaly triggering event that triggers the handling path; wherein the anomaly description information includes at least one of anomaly identifier and error information; associating and combining the handling path, multi-dimensional scene fingerprint, anomaly description information, and handling information to generate a structured experience knowledge unit; and setting initial confidence weights for the structured experience knowledge unit based on the handling information; wherein paths indicated as successful in the handling information are assigned a first initial confidence value, and paths indicated as failed in the handling information are assigned a second initial confidence value, with the first initial confidence value being higher than the second initial confidence value.

[0010] In one implementation of this application, multi-dimensional weighted matching is performed in an experience knowledge base based on the scene features corresponding to the anomaly. Specifically, this includes: extracting feature values ​​from multiple dimensions within the task environment corresponding to the current anomaly to form a current scene fingerprint; determining a hot track knowledge base and a cold track knowledge base in the experience database; automatically transferring experience knowledge units between the hot track and cold track knowledge bases based on confidence weights and call records; calculating the weighted similarity of the current scene fingerprint and the target scene fingerprint of a pre-stored experience knowledge unit in the hot track knowledge base across multiple dimensions to obtain a first similarity set; calculating the weighted similarity of the current scene fingerprint and the target scene fingerprint of a pre-stored experience knowledge unit in the cold track knowledge base across multiple dimensions to obtain a second similarity set; and selecting one or more experience knowledge units with similarities higher than a preset matching threshold from the first and second similarity sets in descending order of weighted similarity, as a candidate recommendation set for matching.

[0011] In one implementation of this application, based on the multi-dimensional weighted matching results, a main path recommendation or a side path recommendation strategy is executed to determine a reference handling path in the experience knowledge base. Specifically, this includes: if there are experience knowledge units with similarity greater than a first preset threshold, then main path recommendation is initiated, historical failure case information corresponding to experience knowledge units that meet the confidence weight conditions is obtained, an applicable condition description is generated, and a recommendation panel is generated based on the experience knowledge units that meet the confidence weight conditions and the applicable condition description; if there are no experience knowledge units with similarity greater than the first preset threshold, then side path inference processing is initiated, one or more dimensions in the scene fingerprint are sequentially filtered out, and similarity calculation and matching are re-executed with the downgraded dimension combination, and a corresponding decay coefficient is applied to the confidence of the matching result to obtain the reference handling path of the downgraded recommendation.

[0012] In one implementation of this application, one or more dimensions in the scene fingerprint are sequentially filtered out, and similarity calculation and matching are re-executed with the downgraded dimension combination. A corresponding attenuation coefficient is applied to the confidence of the matching result. Specifically, this includes: filtering out the version interval dimension and migration scale interval dimension in the scene fingerprint, matching with the remaining four dimensions, and multiplying the confidence of the matching result by a first attenuation coefficient to complete the first-level downgrade; if the first-level downgrade does not match any experience knowledge units that meet the similarity conditions, the task type dimension is further filtered out, matching with the remaining three dimensions, and multiplying the confidence of the matching result by a second attenuation coefficient to complete the second-level downgrade; if the second-level downgrade does not match, cross-database type combination matching is performed only based on the anomaly category dimension and occurrence stage dimension, and the confidence of the matching result is multiplied by a third attenuation coefficient to complete the third-level downgrade; wherein, the first attenuation coefficient is greater than the second attenuation coefficient, and the second attenuation coefficient is greater than the third attenuation coefficient.

[0013] In one implementation of this application, after outputting the reference handling path as a handling suggestion for the current exception, the method further includes: if the handling suggestion is adopted and the task is successful, increasing the confidence weight of the experience knowledge unit corresponding to the handling suggestion; if the handling suggestion is adopted but the task fails, decreasing the confidence weight of the experience knowledge unit corresponding to the handling suggestion, and updating the applicable condition boundary description corresponding to the handling suggestion based on the failure scenario; if the handling suggestion is not adopted but the task is successful, storing the currently executed handling path as a new experience knowledge unit in the experience knowledge base.

[0014] This application provides an engineering experience reuse device for database migration, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, enable the at least one processor to: collect operation events during the task lifecycle and generate an original operation event sequence during the execution of a database migration task; wherein the operation events include at least an anomaly triggering event and a corresponding manual intervention event; after the task is detected to be completed, perform a diagnostic path reverse reconstruction process based on the original operation event sequence to generate a disposal path; associate the disposal path with the multi-dimensional scenario fingerprint and anomaly description information corresponding to the current task to generate a structured experience knowledge unit and store it in an experience knowledge base; wherein the scenario fingerprint is a multi-dimensional feature vector defined based on the database migration scenario; when an anomaly is detected during the execution of a new database migration task, perform multi-dimensional weighted matching in the experience knowledge base based on the scenario features corresponding to the anomaly; and, based on the multi-dimensional weighted matching result, execute a main path recommendation or bypass recommendation strategy to determine a reference disposal path in the experience knowledge base and output the reference disposal path as a disposal suggestion for the current anomaly.

[0015] This application provides a non-volatile computer storage medium storing computer-executable instructions. These instructions are configured to: collect operational events within the task's lifecycle during database migration task execution, generating an original operational event sequence; wherein the operational events include at least an anomaly triggering event and a corresponding manual intervention event; upon detection of task completion, perform reverse reconstruction of diagnostic paths based on the original operational event sequence to generate a handling path; associate the handling path with the multi-dimensional scenario fingerprint and anomaly description information corresponding to the current task to generate structured experience knowledge units, which are then stored in an experience knowledge base; wherein the scenario fingerprint is a multi-dimensional feature vector defined based on the database migration scenario; when an anomaly is detected during the execution of a new database migration task, perform multi-dimensional weighted matching in the experience knowledge base based on the scenario features corresponding to the anomaly; based on the multi-dimensional weighted matching result, execute a main path recommendation or bypass recommendation strategy to determine a reference handling path in the experience knowledge base, and output the reference handling path as a handling suggestion for the current anomaly.

[0016] The above-mentioned technical solutions adopted in this application embodiment can achieve the following beneficial effects: This application embodiment solves the problem of missing experience data by recording key operation events imperceptibly during the tool's task execution, providing a data foundation for subsequent experience structuring and intelligent reuse. Secondly, it automatically transforms the original operation event sequence into a structured processing path, eliminating the dependence on manual processing and transforming experience accumulation from manual-driven to system-automated. By strongly associating the processing path with multi-dimensional scene fingerprints and specific anomaly descriptions, subsequent retrieval matching is based on similarity calculations of multi-dimensional feature vectors, improving the accuracy and scene relevance of recommendations. This application embodiment achieves full coverage of experience and zero-blank recommendation by executing main path recommendation strategies and side path recommendation strategies, maximizing the utilization rate of historical knowledge. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings: Figure 1 A flowchart illustrating a method for reusing engineering experience in database migration, as provided in this application embodiment; Figure 2 This is a schematic diagram of a database migration engineering experience reuse device provided in an embodiment of this application.

[0018] Figure label: 200: Reuse of engineering experience for database migration; 201: Processor; 202: Memory. Detailed Implementation

[0019] This application provides a method, device, and medium for reusing engineering experience in database migration.

[0020] To enable those skilled in the art to better understand the technical solutions in this application, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this application.

[0021] Figure 1 A flowchart of a database migration engineering experience reuse method provided in this application embodiment is shown below. Figure 1 As shown, the method for reusing engineering experience in database migration includes the following steps: Step 101: When executing the database migration task, collect the operation events within the task lifecycle and generate the original operation event sequence.

[0022] In one implementation of this application, during the execution of a database migration task, operation events are collected at stage switching nodes, exception triggering nodes, manual intervention nodes, and task final state nodes to generate an original operation event sequence. Stage switching nodes are nodes where the task undergoes a state transition at least between the pre-inspection stage, full transfer stage, incremental synchronization stage, data comparison stage, repair stage, and completion stage. Exception triggering nodes are nodes where the tool detects errors, alarms, or verification inconsistencies. Manual intervention nodes are nodes where engineers perform operations including pausing or resuming the task, modifying configuration parameters, skipping specified objects, manually executing SQL commands, and invoking tool repair functions. Task final state nodes are nodes where the task transitions to a success, failure, or manual termination state.

[0023] Specifically, in this embodiment, the task event acquisition module is embedded in the tool's task execution engine. It automatically detects and records key operational events during task execution via embedded data points, eliminating the need for manual triggering by engineers and serving as the data entry point for the entire system. This embodiment targets four types of acquisition nodes throughout the task lifecycle, using the process data of database migration / synchronization tasks as the structured acquisition scope: Stage Switching Node: Automatically triggered when a task undergoes a state transition between stages such as pre-inspection, full transfer, incremental synchronization, data comparison, repair, and completion. Recorded information includes: source library type / version, target library type / version, number / scale range of migrated objects, current network latency, elapsed time, stage completion rate, and cumulative alarm count for this stage.

[0024] Anomaly Trigger Node: Automatically triggered when the tool detects an error, alarm, or inconsistency. Recorded information includes: structured anomaly type identifier, anomaly context (stage of occurrence, names and quantities of involved objects, error code and description), and a complete snapshot of the environment at the time of the anomaly.

[0025] Manual Intervention Node: Triggered when an engineer performs any of the following proactive actions: pausing / resumpting a task, modifying configuration parameters, skipping a specified object, manually executing an SQL command, or invoking a tool's repair function. Recorded information includes: intervention type identifier, intervention parameter values, and task status changes before and after the intervention. This node is a core data source for capturing engineers' tacit experience.

[0026] Task Final State Node: Triggered when the task workflow reaches success, failure, or manual termination. Recorded information includes: final status label, total time elapsed, cumulative number of exceptions, cumulative number of manual interventions, and number of unprocessed objects. This node also serves as an automatic trigger signal for the experience accumulation process.

[0027] Step 102: After the task is detected to have ended, perform reverse reconstruction of the diagnostic path based on the original operation event sequence to generate a disposal path.

[0028] In one implementation of this application, an abnormal triggering event is used as the starting anchor point. All operation events from the anchor point to the end of the current task are extracted from the original operation event sequence to generate a candidate disposal event sequence. The candidate disposal event sequence is scanned, and operation events belonging to a preset intervention type are marked as decision nodes. The preset intervention type includes at least one of parameter modification, object skipping, manual SQL execution, and tool repair function call. Multiple status query operations issued consecutively for the same target object within a preset time window are merged and recorded as a single status confirmation operation. Intermediate decision operations whose results are overridden by subsequent decision nodes are filtered out to obtain a disposal action chain. Based on the final state of the database migration task, disposal information is annotated on the disposal action chain to obtain a disposal path. The disposal information includes at least one of the following: task success / failure status, total path execution time, and residual side effects.

[0029] Specifically, engineers' actions when handling anomalies are often out of order and exploratory, with the original event stream containing a large number of invalid queries and repetitive operations that cannot be directly reused. Based on this, embodiments of this application propose a diagnostic path reverse reconstruction algorithm to construct structured, reusable units of experiential knowledge.

[0030] First, anomaly anchor point alignment is performed. Using the anomaly triggering event as the anchor point, all operational events from the moment the anomaly was triggered until the problem was resolved or the task terminated are extracted and arranged in ascending order of timestamps to form a candidate handling event sequence. That is, this process first uses the anomaly triggering event recorded within the task lifecycle as the starting anchor point and locates it in the original operational event sequence based on its timestamp. Then, all operational events within the entire time interval from that anchor point until the task reaches its final state are extracted. These events include, but are not limited to, other possible subsequent anomalies, manual intervention operations performed by engineers, automatic repair attempts performed by the system, and task state change events. Finally, the extracted events are arranged in chronological order to form a candidate handling event sequence.

[0031] Secondly, decision nodes are identified. A rule engine scans the candidate sequence, marking parameter modification, object skipping, manual SQL execution, and repair function call operations as decision nodes, forming the core framework of the handling path. Non-decision nodes (such as duplicate log queries and status polling) are filtered out in the next step.

[0032] Next, redundant operations are compressed, consecutive invalid operations of the same type are merged, transitional steps that have no substantial impact on the final result are removed, and the set of operations necessary to form a minimal effective action chain is retained, reducing noise in knowledge units. Specifically, embodiments of this application identify multiple status query operations initiated consecutively within a preset short time window for the same target database object, such as the same data table. These operations are usually repetitive actions performed by engineers to confirm a certain status, and are merged and recorded as a single logical operation representing the intention to confirm the status. The second category is the elimination of invalid intermediate operations. By analyzing the logical relationships between decision nodes, if the operation result of a certain decision node is completely covered or replaced by the operation of another subsequent decision node, then the intermediate decision node is regarded as an invalid transitional step and is eliminated from the chain. After the above merging and elimination operations, the remaining decision nodes are connected in their original chronological order to form a coherent action chain. Furthermore: First, a dynamic data object state transition network is constructed. After capturing the original sequence of operation events starting from an anomaly triggering event, it is parsed to identify the specific database object affected by each operation event and the state change it causes. Using object identifiers, state values ​​as nodes, and operation events as directed edges, a directed graph structure is dynamically built in memory to form a network depicting the object state transition process. Second, network backtracking weaving is performed starting from the anomaly object. Using the core anomaly data object reported in the anomaly event and its state node at the time of the anomaly as the initial anchor point, the operation sequence is traversed backward, with each subsequent event incrementally woven into the network as a new edge, thereby generating a view of the state evolution path. Next, redundant operations are merged based on the network topology. The system scans the network, identifies multiple edges within a short time window that target the same state transition for the same object and have the same operation type, and merges and aggregates these edges for diagnostic purposes, achieving intelligent compression of repetitive probing operations. Then, by using a predefined database operation semantic state overlay rule base, the edges in the network are analyzed. If it is found that the state node guided by a certain edge is bound to be overwritten or eliminated by the operation performed by another edge targeting the same object, the former is determined to be an intermediate transitional operation that can be overwritten, and it is removed from the core processing path backbone. At the same time, it is folded into a note of the transitional steps previously attempted by subsequent edges and retained. Finally, a refined processing action chain is generated.

[0033] Finally, the handling results are labeled. Based on the status label (success / failure / partial success) of the task's final state node, the compressed handling path is labeled with the results, and the total execution time and side effects of the additional path are recorded, providing a basis for assigning initial values ​​to the confidence weights. Specifically, the labeling information first includes the final result state of the task corresponding to the handling action chain, such as success, failure, or partial success. Second, the actual total execution time of all operations in the handling action chain from the initial exception trigger to the task's final state is calculated. In addition, information on any residual side effects that may occur after executing the handling action chain is analyzed and recorded, such as whether temporary tables were left in the target database or whether certain constraints were disabled but not re-enabled. After labeling, the refined handling action chain carrying the handling result information is encapsulated into a structured handling path, which represents a verified solution pattern for this exception.

[0034] Step 103: Associate the handling path with the multi-dimensional scene fingerprint and anomaly description information corresponding to this task to generate structured experience knowledge units and store them in the experience knowledge base.

[0035] In one implementation of this application, a multi-dimensional scene feature vector is extracted based on the overall context information of the current task to form a multi-dimensional scene fingerprint. The multi-dimensional scene fingerprint includes at least one of the following: database type combination, version range, task type, exception category, migration scale range, and exception occurrence stage. In the exception triggering event that triggers the handling path, exception description information is extracted. The exception description information includes at least one of: exception identifier and error information. The handling path, multi-dimensional scene fingerprint, exception description information, and handling information are associated and combined to generate a structured experience knowledge unit. Based on the handling information, an initial confidence weight is set for the structured experience knowledge unit. Paths indicated as successful by the handling information are assigned a first initial confidence value, and paths indicated as failed by the handling information are assigned a second initial confidence value. The first initial confidence value is higher than the second initial confidence value.

[0036] Specifically, this application embodiment is based on the following: The product types of the source and target databases are read from the task configuration file and combined into a database type combination dimension; the major version number is parsed from the database connection information and categorized into a preset version range dimension; the task type dimension is determined based on the task startup parameters to identify whether it belongs to full migration, incremental synchronization, or data comparison and repair; the predefined classification code is parsed from the abnormal triggering event used as an anchor point and used as an abnormality category dimension; the number of tasks or actual transmitted objects is analyzed and mapped to discrete migration scale range dimensions such as large-scale, medium-scale, or small-scale; simultaneously, the specific stage of the task when the abnormality occurs, such as the full transmission stage or the incremental synchronization stage, is recorded and used as the abnormality occurrence stage dimension. Finally, these dimension values ​​together constitute a structured feature vector, i.e., the multi-dimensional scene fingerprint of this task scenario.

[0037] Based on the initial exception triggering event of this handling process, standardized exception description information is extracted. The extracted content includes: an exception type identifier natively defined by the migration tool or database, which is typically explicitly defined in the tool's error code system or the database's SQLSTATE; the exception occurrence timestamp; and the complete original error message text, which usually contains a specific error description, violated constraints, or failed SQL fragments. Furthermore, information about affected key data objects is extracted from the context load of the exception event, such as the conflicting database table name, constraint name, or specific data record characteristics. This information collectively constitutes the structured exception description information.

[0038] After obtaining the handling path, multi-dimensional scenario fingerprint, anomaly description information, and handling information including success / failure status and execution time, an associative combination operation is performed to generate structured experience knowledge units. Specifically, a data structure object with a fixed pattern is created, with the handling path as the core operational logic of the unit. Subsequently, the extracted multi-dimensional scenario fingerprint and anomaly description information are associated and bound as the scenario index attribute and problem description attribute of the unit, respectively. Simultaneously, the final success / failure status of the task, the total execution time of the path, and any recorded residual side effects information from the handling information are bound as the result effect attribute of the unit. Through this combination, specific anomaly handling instances are transformed into experience knowledge units comprising a problem scenario, problem description, solution, and solution effect. After the experience knowledge unit is generated, an initial confidence weight is assigned to it based on the final task status recorded in its handling information to quantify the initial reliability of the experience carried by the knowledge unit. The allocation rule is as follows: if the handling information explicitly indicates that the task ultimately achieves a successful state after applying the handling path, the system assigns a high initial first confidence value to the knowledge unit, which is set to a preset positive base value. If the task of processing information ultimately fails, then the knowledge unit is assigned a second confidence initial value that is significantly lower than the first initial value.

[0039] Step 104: When an anomaly is detected during the execution of the new database migration task, multi-dimensional weighted matching is performed in the experience knowledge base based on the scenario features corresponding to the anomaly.

[0040] In one implementation of this application, feature values ​​of multiple dimensions are extracted from the task environment corresponding to the current exception to form the current scene fingerprint. A hot track knowledge base and a cold track knowledge base are determined in the experience database; the experience knowledge units in the hot track knowledge base and the cold track knowledge base are automatically transferred according to the confidence weight and call records. The current scene fingerprint and the target scene fingerprint of the pre-stored experience knowledge units in the hot track knowledge base are weighted and similarly calculated across multiple dimensions to obtain a first set of similarities. The current scene fingerprint and the target scene fingerprint of the pre-stored experience knowledge units in the cold track knowledge base are then weighted and similarly calculated across multiple dimensions to obtain a second set of similarities. The first set of similarities and the second set of similarities are then selected in descending order of weighted similarity, with one or more experience knowledge units having a similarity higher than a preset matching threshold, to serve as the matched candidate recommendation set.

[0041] Specifically, this application embodiment includes an experience knowledge base. The experience knowledge base operates as an independent storage service, consisting of two physically separate storage units: a hot track knowledge base and a cold track knowledge base, and is equipped with an automatic transfer mechanism.

[0042] The "hot track" knowledge base stores experiential knowledge units that have been accessed within the last 90 days or have a confidence weight higher than 0.4. The recommendation engine prioritizes searching from hot tracks to ensure recommendation response speed and quality. The "cold track" knowledge base stores archived knowledge units with a confidence weight consistently below 0.4 or no access records within the last 90 days. Cold tracks serve as a backup, participating in the search when hot tracks fail to find a match.

[0043] Dual-track circulation rules: Hot track to cold track migration (degradation): Knowledge units will be automatically migrated from hot tracks to cold tracks when either of the following conditions is met: the confidence weight is below 0.4 for 30 consecutive days and / or there are no call records for 90 days: Cold track to hot track conversion activation (upgrade): When a knowledge unit in the cold track is hit by the recommendation engine and the task is successful after the engineer adopts it, the unit is automatically moved back to the hot track, the confidence weight is reset to 0.6, and it enters the normal closed-loop update process.

[0044] This application's embodiments employ a dual-track flow mechanism to effectively activate dormant long-tail experiences through archiving, preventing high-quality, low-frequency knowledge from being permanently marginalized. Using a six-dimensional scene fingerprint as the index key, it supports two query modes: single-dimensional exact matching and multi-dimensional weighted fuzzy search. Hot tracks are prioritized over cold tracks in search ranking. Each knowledge unit retains a complete historical version record, supporting the tracing of confidence weight changes and the reasons for each update.

[0045] Furthermore, from the real-time context of the current task execution, specific feature values ​​for each dimension required to construct a multi-dimensional scene fingerprint are extracted. First, the scope of the hot track knowledge base used for priority retrieval is determined, which stores recently invoked, knowledge units with confidence weights above the activity threshold, or newly generated knowledge units. Simultaneously, the system also prepares to access the cold track knowledge base as a fallback retrieval source, which stores archived knowledge units with consistently low confidence weights or those that have not been accessed for a long time. The system loads or calculates the latest flow status of knowledge units between the two tracks. This status is automatically maintained by the background process according to preset rules, such as migrating hot track units with weights below the threshold for several consecutive days to the cold track, or activating units in the cold track that have been successfully invoked again and migrating them back to the hot track, ensuring that the knowledge set used during retrieval is up-to-date and hierarchically organized by utility.

[0046] Using the constructed current scene fingerprint as the query vector, a multi-dimensional weighted similarity calculation is performed on the target scene fingerprint associated with each experiential knowledge unit stored in the hot track knowledge base. During the calculation, a preset weight coefficient is assigned to each dimension of the scene fingerprint. This coefficient reflects the importance of that dimension in matching; for example, the database type combination dimension has the highest weight, followed by the version range, and the migration scale range has a relatively lower weight. For each target scene fingerprint, the system calculates its feature value similarity with the current scene fingerprint in each dimension, multiplies it by the corresponding dimension weight, and finally sums the weighted similarities of all dimensions to obtain a comprehensive similarity score between zero and one. After traversing all units in the hot track knowledge base, a first set of similarities is obtained. The same multi-dimensional weighted similarity calculation process is then performed on the cold track knowledge base. That is, using the same current scene fingerprint and the same weight coefficients for each dimension, the calculation is performed with the target scene fingerprint of each experiential knowledge unit in the cold track knowledge base to obtain a second set of similarities. The first and second sets of similarities are merged to form a global list containing all searchable knowledge units and their similarity scores. Subsequently, the global list is sorted from high to low according to weighted similarity scores. This embodiment of the application includes a direct recommendation threshold, which defines the minimum similarity standard for determining a valid match. From the sorted list, all empirical knowledge units with similarity scores higher than the preset matching threshold are selected from top to bottom. These selected units represent all candidate solutions in historical experience that are highly similar to or somewhat related to the current abnormal scenario, collectively forming a candidate recommendation set.

[0047] Step 105: Based on the multi-dimensional weighted matching results, execute the main path recommendation or bypass recommendation strategy to determine the reference handling path in the experience knowledge base, and output the reference handling path as a handling suggestion for the current anomaly.

[0048] In one implementation of this application, if there are empirical knowledge units with a similarity greater than a first preset threshold, then the main path recommendation is initiated. Historical failure case information corresponding to each empirical knowledge unit that meets the confidence weight condition is obtained, and an applicable condition description is generated. Based on the empirical knowledge units that meet the confidence weight condition and the applicable condition description, a recommendation panel is generated. If there are no empirical knowledge units with a similarity greater than the first preset threshold, then bypass inference processing is initiated. One or more dimensions in the scene fingerprint are sequentially filtered out, and the similarity calculation and matching are re-executed with the downgraded dimension combination. A corresponding decay coefficient is applied to the confidence of the matching result to obtain a reference processing path for downgraded recommendations.

[0049] Specifically, if an experiential knowledge unit has a similarity greater than a first preset threshold, candidates are sorted in descending order of confidence weight, and the top three are selected. For each recommendation, common features from historical failure cases are automatically extracted to generate an explanation of applicable condition boundaries. A recommendation panel appears embedded in the tool's anomaly handling interface, displaying for each recommendation: a summary of the handling path, historical usage count, success rate, and applicable condition boundaries. Engineers can adopt, ignore, or mark feedback. The system records the set of differences between the engineer's actual execution path and the recommended path, and passes this data to the closed-loop update module. Furthermore, once a handling path is determined as a candidate recommendation, all historical experiential knowledge units that used that path but resulted in failure are retrieved from the experiential knowledge base, forming a failure case set. Next, the system extracts the multi-dimensional scenario fingerprint from each unit in the failure case set and aligns it with the scenario fingerprint of the current anomaly, constructing a "failure-current difference feature matrix." This matrix not only records the feature values ​​of each dimension of the failure case but also encodes its differences from the scenario fingerprint of the current anomaly. Subsequently, an improved association rule mining algorithm was used to analyze the matrix. First, single-dimensional high-frequency failure features were screened, identifying feature values ​​that frequently appeared in the failure case set and deviated from the current abnormal scenario fingerprint. Then, multi-dimensional feature association analysis was performed to calculate the co-occurrence relationships between these single-dimensional risk features, identifying feature combinations with high support in failure cases and significantly leading to path failure confidence when they appear together, thus discovering complex risk conditions. Next, the mined high-frequency failure features and strongly correlated combinations were summarized into structured boundary rules, and a risk intensity index combining the frequency of occurrence and historical failure rate was calculated for each rule. Finally, these rules were converted into tiered warnings in natural language, such as a high-risk warning: Historical data shows that when the target library is version V12.x and the migration data volume is extremely large, the failure rate of this solution exceeds 70%, thus generating the final applicable condition boundary description. This description will be bound to the recommended path as key metadata and presented in the tool interface in the form of a collapsible panel, hover tooltip, etc., providing in-depth and interpretable risk warnings.

[0050] If no experience knowledge unit with a similarity greater than the first preset threshold exists, a bypass inference layer is activated. This layer constructs downgraded recommendation results using a three-dimensional downgrade strategy, instead of returning empty results. Specifically, the first-level downgrade matching involves discarding the version range and size range dimensions and recalculating similarity using the remaining four dimensions. If candidates exist after the first-level downgrade, the confidence weight of the matched result is multiplied by a decay coefficient of 0.85 before displaying it, along with a downgraded label to alert the engineer. The second-level downgrade matching involves further discarding the task type dimension and retaining only the library type combination, anomaly category, and occurrence stage dimensions for matching, with the confidence weight multiplied by a decay coefficient of 0.65. The third-level downgrade matching involves performing the broadest fallback matching if the second-level downgrade still fails to find a matching experience knowledge unit: only the anomaly category and occurrence stage are retained, and cross-library type combination searches are performed, with the confidence weight multiplied by a decay coefficient of 0.45, and the recommendation results are displayed with a broad fallback label.

[0051] In one implementation of this application, if the handling suggestion is adopted and the task succeeds, the confidence weight of the experience knowledge unit corresponding to the handling suggestion is increased. If the handling suggestion is adopted but the task fails, the confidence weight of the experience knowledge unit corresponding to the handling suggestion is decreased, and the applicable condition boundary description corresponding to the handling suggestion is updated based on the failure scenario. If the handling suggestion is not adopted but the task succeeds, the currently executed handling path is stored as a new experience knowledge unit in the experience knowledge base.

[0052] Specifically, this process is automatically executed after each task's final state node is triggered. By sensing recommendation biases, absorbing differing paths, and dynamically adjusting confidence weights, the knowledge base continuously evolves without any manual intervention. When an engineer adopts a recommended path and the task succeeds, the confidence weight of the corresponding knowledge unit is positively adjusted (+0.02, maximum 1.0), and the number of verifications is accumulated. When an engineer adopts a recommended path but the task fails, the confidence weight is negatively adjusted (-0.10), and the scenario features of this failure are extracted and added to the condition boundary description of that knowledge unit. When an engineer uses a different path than the recommendation and the task succeeds, the differing path is added to the knowledge base as a new experiential knowledge unit, with an initial confidence weight set to 0.7, and is listed alongside the original recommended path for subsequent retrieval. Every 30 days, knowledge units with a confidence weight below 0.3 and no successful records within 90 days undergo soft archiving, retaining historical data but reducing recommendation priority to prevent the knowledge base from being polluted by low-quality paths.

[0053] Figure 2 This is a schematic diagram of a database migration engineering experience reuse device provided in an embodiment of this application. Figure 2As shown, the database migration engineering experience reuse device 200 includes: at least one processor 201; and a memory 202 communicatively connected to the at least one processor 201; wherein the memory 202 stores instructions executable by the at least one processor 201, which, when executing the at least one processor 201, enable the at least one processor 201 to: collect operational events during the task lifecycle and generate an original operational event sequence during the execution of a database migration task; wherein the operational events include at least anomaly triggering events and corresponding manual intervention events; and, upon detecting the completion of the task, execute a diagnostic path based on the original operational event sequence. The process involves reverse path reconstruction to generate a handling path. This handling path is then associated with the multi-dimensional scenario fingerprint and anomaly description information corresponding to the current task to generate structured experiential knowledge units, which are stored in the experiential knowledge base. The scenario fingerprint is a multi-dimensional feature vector defined based on the database migration scenario. When an anomaly is detected during the execution of a new database migration task, multi-dimensional weighted matching is performed in the experiential knowledge base based on the scenario features corresponding to the anomaly. Based on the multi-dimensional weighted matching results, a main path recommendation or bypass recommendation strategy is executed to determine a reference handling path in the experiential knowledge base, and this reference handling path is output as a handling suggestion for the current anomaly.

[0054] This application provides a non-volatile computer storage medium storing computer-executable instructions. These instructions are configured to: collect operational events within the task's lifecycle during database migration task execution, generating an original operational event sequence; wherein the operational events include at least an anomaly triggering event and a corresponding manual intervention event; upon detection of task completion, perform reverse reconstruction of diagnostic paths based on the original operational event sequence to generate a handling path; associate the handling path with the multi-dimensional scenario fingerprint and anomaly description information corresponding to the current task to generate structured experience knowledge units, which are then stored in an experience knowledge base; wherein the scenario fingerprint is a multi-dimensional feature vector defined based on the database migration scenario; when an anomaly is detected during the execution of a new database migration task, perform multi-dimensional weighted matching in the experience knowledge base based on the scenario features corresponding to the anomaly; based on the multi-dimensional weighted matching result, execute a main path recommendation or bypass recommendation strategy to determine a reference handling path in the experience knowledge base, and output the reference handling path as a handling suggestion for the current anomaly.

[0055] The various embodiments in this application are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments of apparatus, devices, and non-volatile computer storage media are basically similar to the method embodiments, so the descriptions are relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0056] The above descriptions are merely embodiments of this application and are not intended to limit the scope of this application. For those skilled in the art, various modifications and variations can be made to the embodiments of this application. These modifications or substitutions do not cause the essence of the corresponding technical solutions to depart from the spirit and scope of the technical solutions in the embodiments of this application.

Claims

1. A method for reusing engineering experience in database migration, characterized in that, The method includes: When performing a database migration task, the operation events within the task lifecycle are collected to generate the original operation event sequence; After the task is detected to have ended, based on the original operation event sequence, a diagnostic path reverse reconstruction process is performed to generate a disposal path. Specifically, this includes: taking the abnormal triggering event as the starting anchor point, extracting all operation events from the anchor point to the end of the current task in the original operation event sequence to generate a candidate disposal event sequence; scanning the candidate disposal event sequence and marking operation events belonging to a preset intervention type as decision nodes; wherein, the preset intervention type includes at least one of parameter modification, object skipping, manual SQL execution, and tool repair function call; merging and recording multiple status query operations issued consecutively for the same target object within a preset time window into a single status confirmation operation, and filtering out intermediate decision operations whose operation results are covered by the operations of subsequent decision nodes to obtain a disposal action chain; based on the final state of the database migration task, labeling the disposal action chain with disposal information to obtain the disposal path; wherein, the disposal information includes at least one of the following: task success or failure status, total path execution time, and residual side effect information; The processing path is associated with the multi-dimensional scene fingerprint and anomaly description information corresponding to this task to generate structured experience knowledge units, which are then stored in the experience knowledge base; wherein, the scene fingerprint is a multi-dimensional feature vector defined based on the database migration scenario. The step of associating the handling path with the multi-dimensional scene fingerprint and anomaly description information corresponding to the current task to generate a structured experience knowledge unit specifically includes: extracting multi-dimensional scene feature vectors based on the overall context information of the current task to construct a multi-dimensional scene fingerprint; wherein the multi-dimensional scene fingerprint includes at least one of the following: database type combination, version range, task type, anomaly category, migration scale range, and anomaly occurrence stage; extracting anomaly description information from the anomaly triggering event that triggers the handling path; wherein the anomaly description information includes at least one of anomaly identifier and error information; associating and combining the handling path, the multi-dimensional scene fingerprint, the anomaly description information, and the handling information to generate the structured experience knowledge unit; and setting an initial confidence weight for the structured experience knowledge unit based on the handling information; wherein a path shown as successful in the handling information is assigned a first initial confidence value, and a path shown as failed in the handling information is assigned a second initial confidence value, wherein the first initial confidence value is higher than the second initial confidence value. When an anomaly is detected during the execution of a new database migration task, multi-dimensional weighted matching is performed in the experience knowledge base based on the scenario features corresponding to the anomaly. Based on the multidimensional weighted matching results, a main path recommendation or bypass recommendation strategy is executed to determine a reference handling path in the experience knowledge base, and the reference handling path is output as a handling suggestion for the current anomaly. The step of executing a main path recommendation or bypass recommendation strategy based on the multi-dimensional weighted matching results to determine a reference handling path in the experience knowledge base specifically includes: if there are experience knowledge units with similarity greater than a first preset threshold, then start the main path recommendation, obtain the historical failure case information corresponding to the experience knowledge units that meet the confidence weight conditions, generate an applicable condition description, and generate a recommendation panel based on the experience knowledge units that meet the confidence weight conditions and the applicable condition description; if there are no experience knowledge units with similarity greater than the first preset threshold, then start bypass inference processing, sequentially filter out one or more dimensions in the scene fingerprint, re-execute similarity calculation and matching with the downgraded dimension combination, and apply a corresponding decay coefficient to the confidence of the matching result to obtain a downgraded recommendation reference handling path; The process of sequentially filtering out one or more dimensions from the scene fingerprint, re-performing similarity calculation and matching with the downgraded dimension combinations, and applying a corresponding attenuation coefficient to the confidence of the matching results specifically includes: filtering out the version range dimension and migration scale range dimension from the scene fingerprint, matching with the remaining four dimensions, and multiplying the confidence of the matching results by a first attenuation coefficient to complete the first-level downgrade; if the first-level downgrade does not match any experience knowledge units that meet the similarity conditions, then further filtering out the task type dimension, matching with the remaining three dimensions, and multiplying the confidence of the matching results by a second attenuation coefficient to complete the second-level downgrade; if the second-level downgrade does not match, then matching is performed only based on the anomaly category dimension and the occurrence stage dimension across database types, and the confidence of the matching results is multiplied by a third attenuation coefficient to complete the third-level downgrade; wherein, the first attenuation coefficient is greater than the second attenuation coefficient, and the second attenuation coefficient is greater than the third attenuation coefficient.

2. The method for reusing engineering experience in database migration according to claim 1, characterized in that, When performing a database migration task, the operation events within the task's lifecycle are collected to generate a raw operation event sequence, specifically including: When performing database migration tasks, operation events are collected at stage switching nodes, exception triggering nodes, manual intervention nodes, and task final state nodes to generate the original operation event sequence. The stage switching node is a node where the task undergoes a state transition at least between the pre-inspection stage, the full transmission stage, the incremental synchronization stage, the data comparison stage, the repair stage, and the completion stage. The abnormal triggering node is the node where the tool detects an error, alarm, or verification inconsistency; The manual intervention node is the node where the engineer performs operations including pausing or resuming a task, modifying configuration parameters, skipping a specified object, manually executing SQL commands, and calling tool repair functions; The task final state node is the node where the task flow transitions to a success, failure, or manual termination state.

3. The method for reusing engineering experience in database migration according to claim 1, characterized in that, The step of performing multi-dimensional weighted matching based on the scene features corresponding to the anomaly in the experience knowledge base specifically includes: In the task environment corresponding to the current anomaly, feature values ​​of multiple dimensions are extracted to form the fingerprint of the current scene. In the aforementioned experience knowledge base, a hot track knowledge base and a cold track knowledge base are determined; wherein, based on the confidence weight and the call record, the experience knowledge units in the hot track knowledge base and the cold track knowledge base are automatically transferred. The current scene fingerprint and the target scene fingerprint of the pre-stored experience knowledge unit in the hot track knowledge base are weighted and similar in multiple dimensions to obtain the first set of similarity. The current scene fingerprint and the target scene fingerprint of the pre-stored experience knowledge unit in the cold track knowledge base are weighted and similar in multiple dimensions to obtain a second set of similarity. The first set of similarities and the second set of similarities are combined in descending order of weighted similarity to select one or more empirical knowledge units with similarity scores higher than a preset matching threshold, which are then used as the matched candidate recommendation set.

4. The method for reusing engineering experience in database migration according to claim 1, characterized in that, After outputting the reference handling path as a handling suggestion for the current anomaly, the method further includes: If the proposed action is adopted and the task is successful, the confidence weight of the experience knowledge unit corresponding to the proposed action is increased. If the proposed action is adopted but the task fails, the confidence weight of the experience knowledge unit corresponding to the proposed action is reduced, and the applicable condition boundary description corresponding to the proposed action is updated based on the failure scenario. If the proposed action is not adopted but the task is successful, the currently executed action path will be stored as a new knowledge unit in the knowledge base.

5. A device for reusing engineering experience in database migration, characterized in that, The device includes a memory for storing computer program instructions and a processor for executing the program instructions, wherein when the computer program instructions are executed by the processor, the device is triggered to perform the method described in any one of claims 1-4.

6. A non-volatile computer storage medium storing computer-executable instructions, characterized in that, The computer-executable instructions are capable of performing the method described in any one of claims 1-4.

Images