Open Source Architecture for the Data Mesh by Ia Mgvdliashvili

Ia Mgvdliashvili explains what a data mesh is and why it emerged: a single team with a single warehouse becomes a bottleneck for business intelligence, so the domain-driven approach known from microservices and platform engineering is applied to data. She walks through the principles of a data mesh, the components it needs, and the open-source stack Canonical has chosen for it: Trino and Ranger, DataHub, Airbyte, Temporal and dbt. The last part covers how the team operates these services and their dependencies on self-hosted infrastructure with Juju and charms, followed by a Q&A. What you'll learn: Why a central data team and a single warehouse stop scaling, and how ownership and control address this What a data mesh is: an organizational structure, an approach to data as a product, and a platform - a federation, not a replacement for warehouses The building blocks: data and metadata catalog, versioned schemas and pipelines, ingestion, global identity and access management, visualization How Trino and Ranger provide query federation with centralized access policies, and how DataHub covers discovery, ownership and lineage for impact analysis How Airbyte replicates data from enterprise systems, with Temporal as the runtime for ingestions and pipelines How Canonical deploys and integrates the stack on Kubernetes with Juju and charms, and what "juju integrate" does Speaker: Ia Mgvdliashvili, Software Engineer at Canonical. Recorded at Thessaloníki DevOps Meetup #101 ("Season kick-off: DevOps meets Data") on 1 October 2026 at OK!Thess, Thessaloniki. Slides: https://iyamg.github.io/slides/2609-okthes... Speaker (LinkedIn): https://www.linkedin.com/in/ia-mgvdliashvi... Event: https://www.meetup.com/thessaloniki-devops... Thessaloníki DevOps Meetup: https://www.meetup.com/thessaloniki-devops... 0:00 Introduction 1:49 Why a single data team becomes a bottleneck 3:54 What a data mesh is and is not 5:21 Data mesh principles and data as a product 8:21 Components needed to implement a data mesh 10:45 Trino and Ranger: query federation and access 12:05 DataHub: data catalog and lineage 14:38 Airbyte ingestion and Temporal execution 16:22 Operating the stack and digital sovereignty 19:25 Juju and charms: operations as software 22:35 Q&A: when to use a data mesh, Juju integrate TAGS #datamesh #datagovernance #thessdevops