Intelligent attribution method for suspension problem of continuous integration assembly line
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-10
- Publication Date
- 2026-03-24
AI Technical Summary
Existing technologies are insufficient in their accuracy for addressing the termination of continuous integration pipelines, have stringent applicable conditions, and low levels of automation, leading to reduced development efficiency and lengthy attribution time.
By employing a large language model combined with a hierarchical termination case knowledge base, and through multidimensional data mining and the hierarchical termination case knowledge base, intelligent attribution of pipeline termination problems is achieved. The steps include: Step 1: Obtaining pipeline logs and context information; Step 2: Locating key intervals in the logs; Step 3: Recalling candidate cases and adjusting priorities; Step 4: Generating attribution analysis results and scoring; Step 5: Optimizing low-confidence results.
It achieves accurate attribution of pipeline termination problems, improves the automation and accuracy of attribution, simplifies the operation process, adapts to different programming languages and pipeline configuration scenarios, and provides an efficient and reliable solution.
Smart Images
Figure CN121722795A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of software development, in particular to an intelligent attribution method for continuous integration pipeline suspension problems. BACKGROUND
[0002] Continuous integration is a software development practice that enables automatic compilation, building and testing of source code. Continuous integration practice realizes frequent integration of code through automated tools, and automatically performs building and testing and other links at each integration, providing fast and high-quality feedback for the development process. The continuous integration pipeline, which is crucial, connects the building, testing and other links of the code through automated tools, realizes the integration verification of frequent code changes, and proves its positive impact on release frequency, software reliability and team efficiency.
[0003] Although the continuous integration pipeline has irreplaceable advantages in the software development environment, its successful implementation faces challenges. Among them, the continuous integration pipeline suspension problem is a major obstacle to the smooth progress of the development process, which is manifested as an abnormal phenomenon that the pipeline does not run as expected and reach the final successful state, but ends prematurely due to various abnormal factors during pipeline execution.
[0004] The continuous integration pipeline suspension problem is difficult to avoid and exists universally, and how to accurately and automatically complete the attribution of the continuous integration pipeline suspension problem in the actual enterprise application process faces many challenges. On the one hand, the continuous integration pipeline suspension problem is diversified, and this diversity makes it complex and difficult to accurately determine the problem type and suspension reason. A large amount of log information will be generated during the running process of the continuous integration pipeline, and these logs contain rich debugging information, but at the same time they are accompanied by a large amount of irrelevant and redundant noise data. Developers need to locate key failure statements and related context from the vast amount of text data, which is time-consuming and prone to errors. On the other hand, the skills required for attribution of continuous integration pipeline suspension problems are different from traditional program debugging. Effective problem attribution not only requires developers to have a deep understanding of the working mechanism of each link of the pipeline, but also requires developers to have solid domain knowledge and practical experience, which greatly increases the problem understanding cost of developers. Therefore, how to quickly and accurately complete the attribution of the continuous integration pipeline suspension problem and simplify its attribution operation process has become a difficult problem to be solved.
[0005] In search of an effective way to solve this problem, the software engineering community has provided some solutions. The main methods for attributing CI pipeline interruption problems are as follows: rule-based matching, machine learning, runtime environment, and information retrieval. However, these existing methods still have limitations such as insufficient accuracy, harsh application conditions, and low automation, and their effectiveness is limited in practical applications, making it difficult to meet the timeliness and accuracy requirements of problem attribution in CI pipeline interruption scenarios.
[0006] Accordingly, the present application provides an intelligent attribution method for continuous integration pipeline interruption problems, which is based on a large language model combined with retrieval enhancement generation technology and a hierarchical interruption case knowledge base, helping users quickly and accurately complete CI pipeline interruption problem attribution, and thus better implementing continuous integration practices. SUMMARY
[0007] The present application aims to provide an intelligent attribution method for continuous integration pipeline interruption problems, which solves the problem of low development efficiency, low attribution accuracy, and long time consumption caused by abnormal interruption of the continuous integration pipeline in the prior art. The method uses multi-dimensional data mining and a hierarchical interruption case knowledge base to solve the problems presented in the background art.
[0008] To solve the above technical problems, the present application provides the following technical solution: after detecting the latest interruption event of the target continuous integration pipeline, automatically triggering the problem attribution processing flow, generating the attribution result and pushing it to the user for review and entry.
[0009] An intelligent attribution method for continuous integration pipeline interruption problems, the method comprising the following steps:
[0010] Step 1: Real-time acquisition of pipeline job logs, code change context, pipeline configuration context, and domain knowledge templates from the continuous integration pipeline platform, construction of structured interruption context information;
[0011] Step 2: Comparison of the current interruption log and the latest passing log through a text difference algorithm, positioning of the log key interval using a pipeline job time window, and extraction of the event template set based on a Drain log analysis algorithm;
[0012] Step 3: Recalling candidate cases from the hierarchical interruption case knowledge base according to a two-stage retrieval mode, adjusting the priority of the candidate cases using a position-sensitive reverse ranking fusion algorithm, and outputting Top-K historical related cases;
[0013] Step 4: Taking the output of steps 1-3 as the input of retrieval enhancement generation, guiding the large language model to generate attribution analysis results through a thought chain prompt, and calculating the confidence score of the attribution results;
[0014] Step 5: When the confidence score is lower than the threshold, trigger the optimization and improvement of the attribution result; automatically extract the feature vector of the attribution result passed by manual review and update it to the hierarchical suspension case knowledge base.
[0015] As an improvement of the present application, the logic of the field knowledge template in step S1 includes: 1-1 six typical pipeline suspension problems and their subdivided suspension reasons, the pipeline suspension problems include code quality problems, test failure problems, dependency management problems, infrastructure problems, permission and security problems, and configuration error problems;
[0016] 1-2 examples of feature signals of suspension reasons.
[0017] As an improvement of the present application, the logic of the text difference algorithm in step S2 includes: 2-1 using the text difference algorithm Myers to compare the passing log and the suspension log to obtain unique segments in the passing log and unique segments in the suspension log;
[0018] 2-2 line-to-line denoising based on text similarity to further filter relevant suspension log segments.
[0019] As an improvement of the present application, the logic of the two-stage retrieval in step S3 includes: 3-1 the recall stage adopts hybrid retrieval, combines different dimensional features for multi-path recall, and ensures extensive coverage of similar candidate result sets in a space-time trade-off manner;
[0020] 3-2 the reordering stage is based on a deep semantic matching model, performs fine-grained feature cross analysis on the recall results, and performs fine reordering based on relevance.
[0021] As an improvement of the present application, the content of the confidence evaluation in step S4 includes: 4-1 historical case matching degree, which includes code change similarity, problem pattern matching degree, and log feature coincidence degree; 4-2 context consistency, evaluating whether the attribution result is inconsistent with the current context information (build environment, timing event); 4-3 field knowledge matching degree, including problem type matching degree, suspension reason matching degree, and feature signal matching degree;
[0022] 4-4 causal logic integrity, evaluating whether the reasoning result considers complex causal chains and multiple reasons, and whether the reasoning process logic is complete and self-consistent.
[0023] As an improvement of the present application, the confidence threshold in step S5 is set to β ∈ [0.7, 0.9], and when the confidence score is lower than β, the optimization and improvement of the attribution result is triggered.
[0024] An electronic device comprises a memory, a processor, and a computer program stored on the memory and executable on the processor, and the processor implements the intelligent attribution method for continuous integration pipeline abortion problems when executing the program.
[0025] A computer readable storage medium having stored thereon computer instructions that, when executed by a processor, implement the intelligent attribution method for continuous integration pipeline abortion problems.
[0026] Compared with the prior art, the present application has the following advantages:
[0027] The present application combines the reasoning ability of a large language model with the retrieval enhancement mechanism of a hierarchical abortion case knowledge base, automatically extracts code changes, log features and configuration context information using a multi-dimensional data processing pipeline, and realizes accurate attribution analysis of continuous integration pipeline abortion problems, effectively solving the problems of low attribution efficiency and limited application scope caused by strong manual dependence and redundant log information in the prior art. At the same time, through the attribution result optimization driven by confidence evaluation and the self-supervised review mechanism to continuously optimize the knowledge base, the adaptability and explainability of the attribution results in cross-programming language and different pipeline configuration scenarios are significantly improved, providing an efficient and reliable solution for automatic attribution of abortion problems in continuous integration pipelines. BRIEF DESCRIPTION OF DRAWINGS
[0028] Figure 1 is an application scenario flowchart of an embodiment of the present application,
[0029] Figure 2 is a whole use flowchart of an embodiment of the present application,
[0030] Figure 3 is a project information integrity check schematic diagram of an embodiment of the present application,
[0031] Figure 4 is a hierarchical abortion case knowledge base design diagram of an embodiment of the present application,
[0032] Figure 5 is a multi-dimensional data acquisition and key feature extraction schematic diagram of an embodiment of the present application,
[0033] Figure 6 is a two-stage retrieval mode schematic diagram of an embodiment of the present application. DETAILED DESCRIPTION
[0034] The present application will be further described in detail below in conjunction with the drawings and embodiments. It can be understood that the specific embodiments described herein are only used to explain the present application, and not to limit the present application. In addition, it should be noted that, in order to facilitate description, only the parts related to the present application are shown in the drawings, not all structures.
[0035] Example: An intelligent attribution method for continuous integration pipeline abort problems, the method comprising the following steps:
[0036] Step 1: Obtain pipeline job logs, code change contexts, pipeline configuration contexts, and domain knowledge templates in real time from the continuous integration pipeline platform, and construct structured abort context information;
[0037] Step 2: Compare the current aborted log with the most recently passed log using a text difference algorithm, locate the key interval of the log using a pipeline operation time window, and extract the event template set based on the Drain log parsing algorithm;
[0038] Step 3: Retrieve candidate cases from the hierarchical termination case knowledge base according to the two-stage retrieval mode, and adjust the priority of candidate cases by combining the position-sensitive reverse ranking fusion algorithm to output the Top-K historical relevant cases;
[0039] Step 4: Use the output of Steps 1-3 as input for retrieval enhancement generation, guide the large language model to generate attribution analysis results through mind chain prompts, and calculate the confidence score of the attribution results;
[0040] Step 5: When the confidence score is lower than the threshold, the attribution result optimization and improvement is triggered; feature vectors are automatically extracted from the attribution results that have been manually reviewed and approved, and updated to the hierarchical termination case knowledge base.
[0041] The logic of the domain knowledge template in step S1 includes:
[0042] 1-1 Six typical pipeline termination problems and their detailed termination reasons: pipeline termination problems include code quality issues, test failure issues, dependency management issues, infrastructure issues, permission and security issues, and configuration errors.
[0043] Examples of characteristic signals for reasons for termination (1-2).
[0044] The logic of the text difference algorithm in step S2 includes:
[0045] 2-1 Using the Myers text difference algorithm to compare the log and the aborted log, unique fragments in the log and the aborted log are obtained;
[0046] 2-2 Line-to-line denoising based on text similarity is used to further filter relevant aborted log segments.
[0047] The logic of the two-stage retrieval in step S3 includes:
[0048] The 3-1 recall stage employs a hybrid retrieval approach, combining features from different dimensions for multi-path recall, using a space-for-time trade-off to ensure broad coverage of similar candidate result sets;
[0049] The 3-2 re-ranking stage is based on a deep semantic matching model, which performs fine-grained feature cross-analysis on the recall results and performs refined re-ranking based on relevance.
[0050] The confidence assessment in step S4 includes:
[0051] 4-1 Historical case matching degree, which includes code change similarity, problem pattern matching degree, and log feature overlap; 4-2 Context consistency, assessing whether the attribution results contradict the current context information (build environment, time sequence events); 4-3 Domain knowledge matching degree, including problem type matching degree, abort reason matching degree, and feature signal matching degree.
[0052] 4-4 The integrity of causal logic: Evaluate whether the reasoning results take into account complex causal chains and multiple causes, and whether the reasoning process is logically complete and self-consistent.
[0053] In step S5, the confidence threshold is set to β∈[0.7,0.9]. When the confidence score is lower than β, the optimization and improvement of the attribution results are triggered.
[0054] Figure 1 This document presents a flowchart illustrating an application scenario of the intelligent attribution method for continuous integration pipeline abort issues, as described in this invention. It primarily consists of four parts: knowledge base construction, project preparation, continuous integration pipeline abort issue attribution, and verification testing. First, historical CI pipeline execution logs and other relevant contextual information need to be integrated to ensure comprehensive coverage of common abort issues. Next, a self-supervised knowledge base expansion method needs to be designed and implemented to automatically extract and update knowledge from abort cases, ensuring the real-time nature and completeness of the knowledge base. Subsequently, continuous integration pipeline abort events automatically trigger project information integrity checks, completing project preparation and data preparation. Then, a fully automated continuous integration pipeline abort issue attribution analysis is performed, and the attribution report is notified to enterprise engineers. Finally, to facilitate use and verification testing by enterprise engineers, the method provides flexible behavior management capabilities and an attribution review mechanism, helping them quickly get started and complete the attribution analysis of continuous integration pipeline abort issues.
[0055] Figure 2This document presents a flowchart illustrating the overall usage of an intelligent attribution method for continuous integration pipeline termination issues, as described in an embodiment of the present invention. First, enterprise engineers need to select a target knowledge base; if the knowledge base does not exist, it needs to be created, and knowledge base options such as retrieval strategy, similarity threshold, and maximum citation limit need to be configured. Then, enterprise engineers need to configure the issue attribution task, such as the full name of the project repository, project pipeline identifier, start scan time, and metadata filtering expression. A continuous integration pipeline termination event automatically triggers the issue attribution processing flow, and the method performs a project information integrity check to ensure that the project information has been imported. Next, data collection and preprocessing begin, using a text embedding model to generate embedding vectors. A two-stage retrieval mode is then used to obtain high-quality historically relevant termination cases, and the case positions are adjusted by optimizing the ranking fusion algorithm. Finally, by combining termination context information, historically relevant termination cases, and domain knowledge templates, a large language model is guided by thought chain prompts to complete intelligent issue attribution. The method identifies low-quality attribution results through a confidence assessment mechanism encompassing four core dimensions: historical case matching, contextual consistency, domain knowledge matching, and causal logic integrity. The assessment feedback then drives the LLM (Local Management Model) to refine the attribution results. Finally, a manual review mechanism intercepts abandoned cases awaiting entry, allowing enterprise engineers to view the attribution results and intermediate outcomes.
[0056] Figure 3 This diagram illustrates a project information integrity check using an intelligent attribution method for continuous integration pipeline aborts, as described in an embodiment of the present invention. For project information that has not yet been synchronized, the method searches for and synchronizes the information in the target repository via the GitHub API and binds it to the default knowledge base. Simultaneously, the method searches for and synchronizes pipeline information in associated pipelines and imports the pipeline configuration.
[0057] Figure 4This paper presents a hierarchical termination case knowledge base design diagram for an intelligent attribution method for continuous integration pipeline termination issues, as described in an embodiment of the present invention. The hierarchical termination case knowledge base is the core knowledge management system of the method, employing a two-layer knowledge base system of "user-customized layer + system general layer". In the problem attribution process, the system first accurately matches the user-maintained domain-customized termination case knowledge base to ensure that the attribution results conform to the specific environment of the project. When the retrieved data is insufficient, it automatically moves to the method's built-in continuous integration pipeline termination case meta-knowledge base, supplementing the analysis with general problem patterns. Through systematic collection, filtering, and annotation of real GitHub Actions termination cases from the open-source community, each case is abstracted into five core fields: key termination log, relevant context, problem type, termination reason, and detailed description, along with other auxiliary fields. The CI pipeline termination case meta-knowledge base construction process includes three main steps: data acquisition, data filtering and verification, and knowledge base construction. Through strict exclusion rules, records that do not meet quality or relevance requirements are eliminated, ultimately retaining over 900 high-quality termination cases for meta-knowledge base construction.
[0058] Figure 5 This illustration demonstrates a multi-dimensional data acquisition and key feature extraction schematic of an intelligent attribution method for continuous integration pipeline abort problems, as described in an embodiment of the present invention. It includes two main functions: data acquisition and data preprocessing. The data acquisition objects include runtime abort logs and abort context data, where the abort context covers code change context, pipeline configuration context, pipeline job context, etc. The data preprocessing process includes steps such as DiffWithPass, key interval localization, and event template extraction. First, the Myers text difference algorithm is used to compare and obtain unique segments from the logs and abort logs, and line-to-line denoising is performed based on text similarity. Then, the start and end times of the first failed step are obtained based on the pipeline job context, and the complete key pipeline log interval is located by combining the command group's specific format and the program failure statement. Finally, the Drain log parsing algorithm is used to extract the log event template set.
[0059] Figure 6This illustration shows a two-stage retrieval model of an intelligent attribution method for continuous integration pipeline termination problems, as described in an embodiment of the present invention. The recall-reordering two-stage retrieval model, by decoupling the computational objectives of similarity matching and relevance inference, can alleviate the performance degradation problem of traditional single retrieval models when data scales up. Specifically, the recall stage employs hybrid retrieval, combining features of different dimensions for multi-path recall, ensuring broad coverage of similar candidate result sets by trading space for time. The reordering stage, based on a deep semantic matching model, performs fine-grained feature cross-analysis on the recall results and performs refined reordering based on relevance. This two-stage retrieval model enables the system to transform data scale growth into information gain: as the hierarchical termination case knowledge base expands self-supervised, the recall layer obtains richer similar case references, while the reordering layer captures more dimensional contextual evidence through a powerful cross-encoder model. The two work together to achieve a positive correlation between retrieval accuracy and data scale.
[0060] It should be noted that the above embodiments are not intended to limit the scope of protection of the present invention. Equivalent transformations or substitutions made based on the above technical solutions all fall within the scope of protection of the claims of the present invention.
Claims
1. An intelligent attribution method for continuous integration pipeline termination problems, characterized in that, Includes the following steps: S1. Multidimensional data acquisition: Real-time acquisition of pipeline job logs, code change contexts, pipeline configuration contexts, and domain knowledge templates from the continuous integration pipeline platform to construct structured termination context information; S2. Key Feature Extraction: The current aborted log is compared with the most recently passed log using a text difference algorithm. The key intervals of the log are located using a pipeline operation time window. Event template sets are extracted based on the Drain log parsing algorithm. S3. Hierarchical Case Retrieval: Based on the two-stage retrieval model, candidate cases are retrieved from the hierarchical termination case knowledge base. The priority of candidate cases is adjusted by combining the position-sensitive reverse ranking fusion algorithm, and the Top-K historical relevant cases are output. S4. Collaborative Reasoning with Large Language Model: The outputs of steps S1-S3 are used as inputs for retrieval enhancement generation. The large language model is guided to generate attribution analysis results through mind chain prompts, and the confidence score of the attribution results is calculated. S5. Dynamic optimization and knowledge update: When the confidence score is lower than the threshold, the attribution results are optimized and improved. Feature vectors are automatically extracted from the attribution results that have been manually approved and updated to the hierarchical termination case knowledge base.
2. The intelligent attribution method for continuous integration pipeline termination problems according to claim 1, characterized in that, The logic of the domain knowledge template in step S1 includes: 1-1 Six typical pipeline termination problems and their detailed termination reasons: pipeline termination problems include code quality issues, test failure issues, dependency management issues, infrastructure issues, permission and security issues, and configuration errors. Examples of characteristic signals for reasons for termination (1-2).
3. The intelligent attribution method for continuous integration pipeline termination problems according to claim 1, characterized in that, The logic of the text difference algorithm in step S2 includes: 2-1 Using the Myers text difference algorithm to compare the log and the aborted log, unique fragments in the log and the aborted log are obtained; 2-2 Line-to-line denoising based on text similarity is used to further filter relevant aborted log segments.
4. The intelligent attribution method for continuous integration pipeline termination problems according to claim 1, characterized in that, The logic of the two-stage retrieval in step S3 includes: The 3-1 recall stage employs a hybrid retrieval approach, combining features from different dimensions for multi-path recall, using a space-for-time trade-off to ensure broad coverage of similar candidate result sets; The 3-2 re-ranking stage is based on a deep semantic matching model, which performs fine-grained feature cross-analysis on the recall results and performs refined re-ranking based on relevance.
5. The intelligent attribution method for continuous integration pipeline termination problems according to claim 1, characterized in that, The confidence assessment in step S4 includes: 4-1 Historical case matching degree, which includes code change similarity, problem pattern matching degree, and log feature overlap degree; 4-2 Contextual consistency: Evaluate whether the attribution results contradict the current context information (construction environment, time sequence events); 4-3 Domain knowledge matching degree: including problem type matching degree, termination reason matching degree, and feature signal matching degree; 4-4 The integrity of causal logic: Evaluate whether the reasoning results take into account complex causal chains and multiple causes, and whether the reasoning process is logically complete and self-consistent.
6. The intelligent attribution method for continuous integration pipeline termination problems according to claim 1, characterized in that, The confidence threshold in step S5 is set to β∈[0.7,0.9]. When the confidence score is lower than β, the optimization and improvement of the attribution results are triggered.
7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements an intelligent attribution method for continuous integration pipeline termination problems as described in any one of claims 1 to 6.
8. A computer-readable storage medium storing computer instructions thereon, characterized in that, When executed by the processor, the computer instruction implements an intelligent attribution method for continuous integration pipeline abort problems as described in any one of claims 1-5.