The second edition of the International Workshop DevOps@MODELS devoted explicitly to DevOps and Model-Driven Engineering will be held virtually on October 11 as part of the satellite events of the ACM/IEEE 24nd International Conference on Model Driven Engineering Languages and Systems (MODELS 2021).
To attend the workshop, you need to register through the MODELS 2021 registration portal. Registering gives you access to DevOps@MODELS, as well as all the other co-located events and main conference tracks.
Program (October, 11th 2021)
The workshop is organized into two acts of 75 minutes each, each part containing one invited presentation (30’ presentation + 10’ questions), one paper presentation (10’ presentation + 10’ questions) and one lightning talk (5’ presentation + 10’ questions). See details of the logistics and abstracts below. After the workshop, we plan to have an additional one-hour session of discussions at the end of the workshop.
In the end, the workshop will be 90 minutes of presentations and 120 minutes of discussions, so bring an open mind and be ready to discuss!
The workshop will start at 11 PM in Tokyo, 4 PM in Berlin, 3 PM in London, 10 AM in Toronto and 7 AM in Vancouver. In case of doubts, please use the official MODELS webpage.
- Opening (10 minutes)
- Act I: DevOps & Modelling (75 minutes)
- Break (15 minutes)
- Act II: Models for migrating to DevOps (75 minutes)
- Invited: Three real world uses cases of AI operationalization to illustrate AI DevOps challenge (Eric Charton, from Banque Nationale)
- Paper: From Monolithic to Microservice Architecture: The Case of Extensible and Domain-Specific IDEs (slides)
- Lightning: Feedback on the development of a software system for automatic online selection of ML pipelines (slides)
- Break (30 minutes)
- Extra session for discussions (60 minutes)
Video recording (registered participant only): Whova session
Invited Industrial Presentations
- Ali Tizghadam (Principal Technology Architect, TELUS): Opportunities in the midst of challenges
- DevOps journey has been a challenging yet fascinating experience for different organizations. In this talk we zoom into DevOps practices in Telco domain, the challenges we have faced in TELUS as a Telecommunications service provider and the opportunities arose. The talk will explore the shifts we had to experience in architecture, processes, mindset, execution, and daily operations habits, together with the role of abstraction, models and automation in addressing the challenges.
- Dr Ali Tizghadam is Principal Technology Architect at TELUS (the 2nd largest telco operator in Canada) since 2013, and Senior Research Associate at the University of Toronto since 2012. After his M.Sc. degree in Tehran (1994), he graduated from the University of Toronto (PhD) in 2009.
- Eric Charton (Senior Director, National Bank of Canada): Three real world uses cases of AI operationalization to illustrate AI DevOps challenge
- Widely recognized and promoted DevOps methodologies or tools used to push AI systems in production are not a 100% recipe for success. With three use cases as an illustration, we will explain how the question of change management can be much more important than the DevOps framework to successfully deploy in production. The three uses cases we will use as an illustration are real life ones: introducing state of the art research in a high traffic search engine, deploying successfully and multiple time an open source dialog platform in a bank, and delivering a new machine learning based risk modelling algorithm. Those use cases were applied in various industries (banking, media), between 2016 and now.
- Dr Eric Charton is Senior Director (AI Science) at the National Bank of Canada. He graduated from Université d’Avignon et des Pays du Vaucluse (M.Sc. – 2007, Ph.D. – 2010).
- Alessandro Colantoni, Benedek Horváth, Ákos Horváth, Luca Berardinelli and Manuel Wimmer (Johannes Kepler University Linz, IncQuery Labs): Towards Continuous Consistency Checking of DevOps Artefacts
- DevOps processes are often scattered over a multitude of technologies, and thus, their integration is a challenging endeavour. In order to tackle this problem, several integration platforms have been proposed in the industry, e.g., Keptn to mention a current open-source initiative, which aims to provide an integration platform to specify and execute DevOps processes. They often employ a family of languages for this purpose. However, as we have learnt from UML, SysML, and many others, a family of languages requires inter-model constraints to be checked in order to guarantee a high consistency between the different artefacts. This is of special importance for DevOps artefacts, as they are critical ingredients for the success of modern projects. In this work-in-progress paper, we propose a Model-Driven Engineering (MDE) approach for the continuous consistency checking of DevOps artefacts. First, we explicitly represent each DevOps artefact as a model, second we establish links across them to set a navigable network of models elements, third we enable services on top of this network. We envision the possibility of using GitOps to pull the DevOps artefacts, execute services for checking consistency and performing model repairs, uploading the changes to the DevOps tools, and finally the DevOps tool pushing the artefacts to Git, thus resulting in a continuous consistency checking process in practice.
- Romain Belafia, Pierre Jeanjean, Olivier Barais, Gurvan Le Guernic and Benoit Combemale (Univ Rennes, Inria, CNRS, IRISA ): From Monolithic to Microservice Architecture: The Case of Extensible and Domain-Specific IDEs
- Integrated Development Environments (IDEs) are evolving towards cloud-native applications with the aim to relocate the language services provided by an IDE on distant servers. Existing research works focus on the overall migration process to handle more efficiently their specific requirements. However, the microservicization of legacy monolithic applications is still highly dependent on the specific properties of the application of interest. In this paper, we report our experiment on the microservicization process of the Cloud-Based graphical modeling workbench “Sirius Web”. We aim to identify the technical challenges related to applications with similar properties and provide insights for practitioners to migrate their similar applications towards microservices. We discuss the main lessons learned and identify the underlying challenges to be further addressed by the community.
- Nicolas Hili (Univ. Grenoble Alpes, CNRS, LIG): An Integrated CI/CD Workflow for eXecutable Domain-Specific Modelling Languages
- With the growing interest in eXecutable Domain-Specific Modelling Languages (xDSMLs), there is now a need to address the complexity of their development in the same way as the software artifacts they intend to model. It is especially true since dedicated languages are more and more complex, fast-evolving and that specifying their execution semantics is error-prone. This extended abstract focuses on the continuous integration and continuous delivery (CI/CD) aspects of DevOps and their automation to support faster and smaller releases. It details a first experience report on integrating CI/CD workflows with modelling environments to ease the development of xDSMLs. This report intends to fulfill four objectives: i) it depicts the adaptation of a Test-Driven Development (TDD) approach to modelling to automate the execution of test cases to validate the correctness of the abstract syntax and execution semantics of xDSMLs; ii) it describes the setup of a GitLab CI workflow coupled with the Mocha test framework to facilitate the development, test, and delivery of xDSMLs; iii) it discusses the implementation of a proof-of-concept tool to facilitate the development and the test of xDSMLs, including model coverage reporting; and iv) it discusses how to adapt other code-specific activities (e.g., code linting) from a model perspective. An application to the definition of an executable version of UML for Real-Time (a dedicated language for the development of reactive systems) is shown to evaluate the suitability of the workflow for taming the complexity of medium-to-large xDSML definitions.
- Mireille Blay-Fornarino and Frédéric Precioso (UCA, CNRS, I3S, Inria): Feedback on the development of a software system for automatic online selection of ML pipelines
- This presentation aims to share the lessons learned from the implementation of the ROCKFLows software system dedicated to the automatic selection of online Machine Learning (ML) pipelines. Focusing on the critical phases of the DevOps lifecycle, we will outline some of the mistakes made while designing the system (2015-2020, and highlight the challenges of integrating and leveraging ML techniques in a DevOps ecosystem. The research behind ROCKFlows is motivated by the observation that there is no single best pipeline for structured data classification. It extends the known problem of automatic algorithm selection to algorithm composition selection. The rock flows system combines multiple pipelines into a portfolio. Faced with a specific instance of a classification problem, it determines the appropriate pipelines and predicts their performance. This software system integrates ML (meta-learning), software product line, and graph analysis techniques to reduce the space of possibilities. We will present our efforts to better integrate ROCKFlows-like systems in DevOps ecosystems for data scientists. We aim at a new architecture more aware of resource management, integrating the elicitation of learning processes, and adapting to the evolution of our knowledge. We advocate a more intelligent marriage between context-driven research, metamodeling to limit abstraction debts, and reasoned use of machine learning.
- Nicolas Hili, Grenoble Alpes University (France)
- Wesley K. G. Assunção, University of Rio de Janeiro (Brasil)
- Romina Eramo, University of L’Aquila (Italy)
- Alfredo Capozucca, University of Luxembourg (Luxembourg)
- Benjamin Benni, Instant Systems (France)
- Aitor Arrieta, Mondragon Goi Eskola Politeknikoa (Spain)
- Grischa Liebel, Reykjavik University (Iceland)
- Henry Muccini, University of L’Aquila (Italy)
- Manuel Wimmer, Johannes Kepler University Linz (Austria)
- Candy Pang, MacEwan University (Canada)
- Nicolas Ferry, University Côte d’Azur (France)
- Matthias Tichy, Ulm University (Germany)
- Mark van den Brand, Eindhoven University of Technology (Netherlands)
- Andreas Wortmann, University of Stuttgart (Germany)
- Francis Bordeleau, École de Technologie Supérieure (ETS), Canada.
- Jean-Michel Bruel, Université de Toulouse, France.
- Juergen Dingel, Queen’s University, Canada.
- Sébastien Mosser, Université du Québec à Montréal (UQAM), Canada.