Microservices offer flexibility and scalability, but their decentralized and fast-changing nature makes it difficult to maintain consistent and meaningful test coverage, particularly at the level of service endpoints. Services are developed independently, logic is spread across multiple layers, and execution paths vary widely based on input and control flow. As a result, automated testing is hard to scale, and manual testing is time-consuming and error-prone. Recent advances in large language models present a promising opportunity for generating tests automatically. However, existing approaches often rely on providing the entire source code as input, which can exceed model limitations and include unrelated logic that reduces test quality. This paper proposes a structured approach that uses interprocedural control flow graphs to guide language models in generating accurate, maintainable endpoint tests that better reflect the complexity of modern microservice systems.
Microservices decompose systems into small, independent services, offering benefits that have led to widespread adoption. However, their distributed nature, diverse technology stacks, and independent development introduce significant challenges in ensuring system reliability. While numerous automated testing approaches have been proposed, only a few manually implemented test suites are openly available for comparison. To the best of our knowledge, no extensive benchmark test suites for component and integration testing of microservices are publicly accessible. This work addresses this gap by providing two benchmark test suites - one on the component and one on the integration level - developed within a case study on the well-established TrainTicket microservice system. The test suites were initially implemented by four independent teams and later unified by the authors, yielding 1,365 component and 210 integration tests in total. They uncover 51 faults, 8 untypical design decisions, and 1 fallback-related issue within the system. We further manually analyze the detected faults and compare them with existing fault taxonomies, identifying 11 gaps in current classifications. Additionally, we provide insights into the relation between the test case type and fault type it detected. Our insights can support future evaluations of new approaches and foster the development of more comprehensive fault models for microservice systems.
In distributed edge computing, effective resource management is key. While orchestration offers centralized control, choreography provides a decentralized alternative more suited to dynamic, resource-constrained environments-especially within emerging serverless paradigms. This study investigates choreography techniques for serverless edge computing, analyzing coordination strategies, communication models, and application domain. We conducted a Multivocal Literature Review (MLR), selecting 12 peer-reviewed studies from 332 sources (200 gray, 122 white literature). Three main choreography approaches emerged: structured, scheduling-based, and hybrid. End-to-end communication was twice as common as broadcasting. IoT was the most prevalent domain, consistent with the dominant use of lightweight edge devices. Choreography offers performance and flexibility benefits but remains underexplored. Key limitations include resource management challenges, underscoring the need for further research into scalable, decentralized coordination models for the edge.
Proactive resource scaling in container orchestration platforms like Kubernetes is essential to maintain application responsiveness under fluctuating workloads. Our study aims to explore predictive auto-scaling by forecasting CPU utilization from service request volumes using statistical and neural-networkbased time series models. We applied 29 forecasting techniques to production traces from a FinTech system, evaluating each model's accuracy using nine metrics: MAE, RMSE, SMAPE, MASE, RMSSE, $R^{2}$, NMAE, and NRMSE, and assessing the actual prediction distance. Experimental results show that neural network models, particularly Transformer and GRU, consistently exhibited strong accuracy performance with high explanatory power ($R^{2}$ values above 0.88) and relatively low mean absolute errors (MAE). However, several statistical models, specifically AutoTheta, FFT, and Exponential Smoothing, achieved higher accuracy than neural approaches, and in particular Exponential Smoothing, was the best performing model scoring the lowest MAE (332.824) and highest $R^{2}$ (0.961). Our findings demonstrate the viability of lightweight forecasting-driven scaling and suggest practical improvements for auto-scaling reliability in Kubernetes environments. The source code we used in this study is available as open-source on github.
This empirical study investigates industry awareness about the adoption of green practices across five modern software architectures: Microservices, Event-Driven, Serverless, Machine Learning, and Generative AI. Using an approach that combined surveys (109 responses) and interviews (19 professionals) from the industry, we gathered data to understand awareness of green practices throughout the design, development and maintenance of software architectures. Our findings reveal that while software professionals understand green software concepts, practical implementation lags due to higher priority given to performance, cost, accuracy, and reliability. Each architecture has specific energy challenges: Microservices struggles with resource allocation and excessive communication; Serverless faces over-invocation and cold-start issues; Event-Driven systems struggle with excessive triggering; Machine Learning systems have challenges with inadequate maintenance of model health and lack of awareness in compute requirements between training and inference phases; and Generative AI consumes high energy due to model size and algorithmic complexity. We identify both architecture-specific and common operational strategies for reducing energy consumption, establishing a foundation for future work in sustainable development in resource-intensive architectures.
Manual security policy validation of Infrastructure-as-Code (IaC) creates bottlenecks in enterprise CI/CD pipelines, with 90% of cloud breaches involving misconfigured IaC. Traditional static analyzers struggle with evolving cloud services and custom policies. We propose a production-ready framework augmenting conventional scans with Large Language Models (LLMs) for Kubernetes, IAM, and Terraform validation. Our evaluation on 500 synthetic IaC cases shows ensemble methods achieve F1 = 0.95 at 3.1s latency. LLMs detect complex violations missed by rule-based tools, reducing manual review by 60% and maintenance by 70%. Real-world testing in Jenkins and Bamboo confirms cross-platform compatibility. We provide: (i) privacy-preserving CI/CD architecture, (ii) safeguards against prompt injection and hallucination, and (iii) phased rollout strategy for regulated enterprises balancing security and velocity.
Cloud-native architectures face escalating security challenges that traditional approaches cannot address at scale. This paper presents an AI-augmented DevSecOps framework integrating machine learning models into security pipelines for realtime threat detection and automated response. The framework achieves 95% attack detection rates with sub-2 second latency at 10 k events/sec. Key contributions include LSTM-based threat detection embedded in CI/CD workflows, adaptive model training with 98% accuracy retention over 6 months, and complete opensource implementation. Experimental validation across multiple attack scenarios demonstrates effectiveness while maintaining operational efficiency in hybrid Kubernetes-serverless environments.
Advanced data science and AI technologies, such as Digital Twins, offer new avenues for innovation in addressing complex scientific problems. However, developing an effective Virtual Research Environment (VRE) to realize these innovation potentials is technically challenging. Incorporating the emerging technologies into the research lifecycle as a new problem-solving paradigm requires assembling diverse software and technological components, which are often not yet fully production-ready. Additionally, successful implementations are often limited to demonstrator cases that are difficult to generalize, leading to significant development risks when creating mature services within a VRE intended for large scientific communities. While agile practices can support development by promoting close engagement with end users throughout the process, their iterative nature does not inherently guarantee the evolution of a technically sustainable solution, potentially increasing risks in project delay or even failure. This paper introduces a Virtual Lab Maturity Framework designed to facilitate the co-development process between end users and development teams, ensuring a more coordinated and sustainable approach to building effective VREs.
This paper introduces an adaptive test and runtime healing approach that delivers resolver-level resilience for GraphQL service-oriented architectures by unifying three telemetry streams: semantic log embeddings obtained from large language models, structural dependencies encoded via graph neural networks, and statistically grounded operational metrics. These signals are fused into a single reinforcement learning state vector, enabling a deep Q-network to learn context-aware recovery actions including selective retry, safe skip, dependency reordering, and escalation without obscuring root causes. The approach is evaluated in a production-grade case study involving a real-world lifestyle coaching platform used by thousands of active users. The application's asynchronous, cloud-native architecture with complex resolver interactions and AI-powered personalization provides a realistic and challenging environment for assessing the system's robustness. Across more than one thousand simulated failure episodes that inject realistic cloud uncertainty, the approach improves test and runtime success rates from 68.7% to 92%, reduces mean-time-to-recovery from 687 ms to 203 ms, and trims CI compute time by 61% using a KL-stability early-stop rule. It also preserves tail-latency accuracy within a 5% error bound while incurring only 11.8 ms median inference overhead per healed request. These results demonstrate that statistically principled, reinforcement-learning-driven healing offers a practical, fine-grained self-recovery solution for serviceoriented systems deployed in modern, real-world cloud applications.
Background: The rapid growth of IoT and edge devices demands low-latency, scalable computing solutions. Traditional cloud architectures introduce bottlenecks, while serverless computing, particularly Function-as-a-Service (FaaS), offers an event-driven, on-demand execution model. However, existing FaaS platforms rely on centralized orchestration, leading to cold start delays, resource inefficiencies, and limited adaptability in edge environments. This study examines containerized workload management for serverless edge computing, evaluating orchestration, choreography, and offloading techniques to improve function execution efficiency and system scalability. Objective: The aim of this work is to conduct a systematic review the state-of-the-art practices in orchestration and offloading techniques for containerized environments. Method: We conducted a Multivocal Literature Review, selecting 94 works from a pool of 771. Results: The offloading strategies rely on a central orchestrator and on the use of well-established tools such as Kubernetes and Docker. Few studies propose alternatives, highlighting a lack of diversity in offloading tools. These studies underscore a significant lack of diversity in the tools used for offloading in containerized systems. Conclusion: We have identified key strengths and limitations of the existing approaches. These limitations are primarily related to response time, performance, and resource utilization.
According to Forbes, eight of the top ten digital companies are from the U.S., which demonstrates the country's technological advancements. Yet, one does not need to be a rocket scientist to notice something amiss with the U.S. software engineering workforce. Women account for just 23% of the U.S. software engineering workforce, and the figures are no better when assessing the computer and information sciences bachelor's degrees in the U.S., where the same proportion of 23% goes to women. Rather than importing the workforce, there is an urgent need to engage women in software engineering careers, but what do we do to make it happen, and what are we doing wrong? This paper presents a critical perspective with pointers to current issues in software engineering education. It illustrates innovative perspectives that educators could employ to transform the education environment into more women-friendly. Narrowing gender equality in a traditionally male-dominant field like software engineering would lead to female workforce development and economic rewards, with 51% of the population more actively included in the field.
This paper presents a case study conducted at a manufacturing enterprise focused on modernizing mobile application integration by transitioning from SOAP-based web services to RESTful APIs using the JAX-RS Reference Implementation (RI). The objective was to expose existing enterprise services in JSON/REST format to support mobile platforms, specifically iOS and Android. The study demonstrates that adopting JAX-RS RI for RESTful service development not only improves developer productivity and service maintainability but also enhances runtime performance and scalability. By utilizing JAX-RS RI, the development team was able to implement stateless, resource-oriented APIs that significantly simplified client-side development while maintaining backend compatibility with existing enterprise systems. Our findings indicate that RESTful services provide significant advantages over traditional SOAP-based architectures for mobile applications, including quantifiable improvements in performance, developer productivity, and system maintainability. The implementation resulted in a 65% reduction in response payload size, 40% improvement in response times, and 50% reduction in development effort for mobile client applications. Additionally, the RESTful approach facilitated easier debugging, testing, and cross-platform support through standard HTTP methods and human-readable JSON formats. The results confirm that RESTful services, implemented via JAX-RS RI, offer a scalable and effective alternative to SOAP for enterprise mobile integration scenarios.
The advent of large language models has ushered in a new era of agentic systems, where artificial intelligence programs exhibit remarkable autonomous decision-making capabilities across diverse domains. This paper explores agentic system workflows in the financial services industry. In particular, we build agentic crews with human-in-the-loop orchestrator that can effectively collaborate to perform complex machine learning modeling tasks. The modeling crew consists of a judge agent and multiple agents who perform specific tasks such as exploratory data analysis, feature engineering, model selection/hyperparameter tuning, model training, model evaluation, and writing documentation. We demonstrate the effectiveness and robustness of modeling crews by presenting a comparative experiment applied to the detection of credit card fraud and card portfolio credit risk. Our fraud detection experiment achieved a recall of 81.6 % and an F1-score of 88.9 %, outperforming AutoML's 70.4 % and 82.1 %, respectively.
This demonstration paper presents, a framework that showcases the integration of Google's Agent-to-Agent (A2A) Protocol and Model Context Protocol (MCP) within multi-agent AI systems. Using IT incident response as our demonstration example, we show how these protocols enable effective communication, context sharing, and problem-solving among AI agents with diverse capabilities. Our demonstration illustrates how Agent-to-Agent Protocol handles structured message passing and task delegation, while Model Context Protocol ensures consistent contextual understanding. Through simulated IT incidents of varying complexity, we demonstrate how this integrated approach improves response efficiency, reduces resolution time, and enhances automated remediation quality. This work provides insights into protocol interoperability challenges and presents a reference implementation adaptable for other multi-agent collaborative scenarios beyond IT operations.
Microservice architectures improves scalability but introduces complex communication challenges when breaking changes occur. Identifying affected teams and coordinating responses often depend on manual processes and fragmented knowledge. I propose a framework for automating coordination workflows by combining Incremental Software Architecture Reconstruction, developer identification using Git blame, refinement through Social Network Analysis, and cross-project issue generation. The aim is to reduce coordination overhead and improve response through targeted, architecture-aware notifications.
The growing adoption of industrial robots has spurred interest in combining Industrial IoT and cloud computing technologies to create unified management platforms. Traditional systems focus mainly on analyzing robot-status data to infer maintenance cycles and monitor operational health, yet often overlook the complex process optimization tasks that many robots must execute. In addition, the heterogeneity among industrial robot operating systems further complicates the management and execution of process optimization. To address these challenges, we propose CIRPO-a Cloud-edge collaborative platform for Industrial Robots Process Optimization, a framework designed to unify both the execution and optimization of robotic processes across cloud and edge tiers. It provides a unified command interface by abstracting vendor-specific controller differences, and implements process dispatching and execution through a workflow-oriented approach, thereby supporting diverse industrial operations. We validate the framework by integrating two real-world applications-palletizing and polishing-demonstrating CIRPO's ability to deliver closed-loop, cloud-based process optimization for heterogeneous robot fleets. The CIRPO source code, helm charts, and anonymised experiment traces will be release on github upon acceptance.
The widespread adoption of Microservices Architecture (MSA) across industries has amplified trust, privacy, and security challenges. Due to its distributed nature, the growth in data-sharing rate among microservices increases privacy concerns. These are implicit privacy risks that users are unaware of, and they rarely realize when or how their private data has been exposed. For instance, fitness trackers can leak location data via third-party app installation, while photo-sharing apps may expose it through weak APIs. Existing distributed tracing technologies provide runtime monitoring, security solutions that rely on them are retrospective. The massive trace data with excessively redundant traces further complicates real-time threat detection, especially when data can travel beyond the intended destination. Our lightweight, preplanned path traversal platform ensures authentication without exposing private data, monitors service paths, and decentralizes trust. We demonstrate experimentally that the platform enhances privacy protection and guarantees absolute path certainty. Load testing on a microservices-based system with 750 simultaneous users achieved (100 %) accuracy in blocking unauthorized path deviation. Our theoretical analysis confirms robust protection, precise path planning, and effective monitoring-delivering privacy, security, and data integrity.
Decentralized identities take a user-centric approach, which means that users are in control of their identity information. This allows users to share only selected information with specific service providers. One challenge associated with decentralized identities is managing the issuance and verification of digital identities within the software development process. There are standards and systems for decentralized identities, but an engineering approach is still missing. In this article, we focus on microservice-based applications. Several vendors offer decentralized identity systems that support the issuance and verification of digital credentials. During the development of an application, the concepts of decentralized identities need to be applied in a systematic way. The core contribution of this publication are the engineering processes for issuance and verification that address the requirements and architectural challenges associated with decentralized identities. During development, we consider how decentralized identities can be integrated into a microservice architecture in a loosely coupled manner. We illustrate the applicability by demonstrating how the engineering concepts of decentralized identities can be applied within a proof of concept, the DrivingLicensePoC. This PoC consists of two microservice-based applications, one for the issuance of a digital driving license and one for the verification of this digital driving license in the context of a car rental. In addition, we address the interoperability aspects of DI integration that need to be addressed.
Managing costs across multiple cloud service providers (CSPs) is challenging due to each platform's unique data formats and billing models. We propose a service-oriented Multi-Cloud Cost Management Platform (MCCMP) that treats each CSP's cost exports as independent services, unified under the FOCUS specification. This approach simplifies cross-provider cost comparisons and fosters modular integration, allowing new CSPs or features to be added with minimal disruption. We further illustrate an anomaly detection (AD) capability using MLOps to demonstrate the system's extensibility for advanced analytics. We present our initial findings on the system architecture, data normalization process, and partial AD results, including early validation showing an AUC of 0.885 for anomaly detection and ingestion latencies under one minute.
This paper presents Graph-RAG and Self-learning LLM-based Agent Services Framework for structured reasoning and knowledge-driven analysis. The proposed approach integrates graph-enhanced retrieval mechanisms with self-learning Large Language Models (LLMs) to improve critical analysis and domain-specific decision-making. The framework is evaluated using Air Accidents Investigation Branch (AAIB) Publications Reports, which provide structured, investigative narratives aimed at preventing future aviation incidents rather than assigning blame. By leveraging graph-based knowledge learning, the framework enhances causal reasoning, multimodal response generation, and retrieval accuracy, demonstrating its capability to support structured problem analysis based on real-world investigative experiences. Experimental results show significant improvements in hallucination mitigation, retrieval precision, and real-time performance when compared to standard Retrieval-Augmented Generation (RAG) models. The findings highlight the potential of graph-augmented self-learning LLMs in transforming automated analytical workflows, paving the way for enhanced visual knowledge exploration and structured decisionsupport systems.
