Building a data mesh platform for self-service analytics

About

Founded in Barcelona in 2015, Glovo is a pioneering multi-category app connecting users with businesses and couriers to offer on-demand services from local restaurants, grocers and supermarkets, and high-street retail stores. Glovo’s vision is to give everyone easy access to everything within their city. 

The company serves over 15 million customers in 25 countries across South West Europe, Eastern Europe, Central Asia, and Sub-Saharan Africa. As this quick-commerce startup grew, it wanted to replace its monolithic data warehouse with a data mesh solution to scale its data platform through decentralization.

Challenge

Glovo has a complex, four-sided marketplace that connects customers, couriers, advertisers, and brands. All of these participants need constant, reliable access to data for many analytical and operational use cases. The company is continually optimizing its platform to find the best methods of creating shared value for all participants. 

To support this, Glovo decided to shift to a data mesh architecture and needed a high-performing query engine on top of its Amazon S3 data lake. The company started using open source Trino, but the engineering team required enterprise-grade features that Trino couldn’t provide.

In addition, Glovo was looking for these features:

• Cluster autoscaling. 
• Data products dashboard.
• Integration with Privacera.

Solution

After evaluating other options like Amazon Athena and Google BigQuery, Glovo saw the most value in Starburst Enterprise. It offers enterprise-grade features such as autoscaling and security, 24×7 support from Trino experts, and query optimizations from Starburst Enterprise’s pre-built connector library, including Delta Lake support. 

In addition, Starburst Enterprise offered Glovo these features:

• The simplicity of deploying with Kubernetes.
• Elastic autoscaling for many types of workloads.
• A user interface allowing both admins and data consumers a visual overview to monitor clusters and optimize queries.
• A connectivity layer enabling direct access datasets where they reside.
• Integration with Apache Ranger, allowing Glovo to standardize authorization through role-based access control (RBAC).
• Integration with preferred Business Intelligence tools, including Looker and Tableau.

Results

Glovo’s data platform team deployed Starburst Enterprise, which enables self-service analytics for its data mesh. The company has seen significant business improvements, including:

• 50% increased query speed. Query time improved 50% — from 10 minutes to 5 minutes — compared to open-source Trino.
• Insights for self-service analytics. Starburst Insights give users a visual overview of important metrics and the ability to monitor clusters and optimize queries. 
• Reduced total cost of ownership (TCO). Glovo can separate compute from storage, reducing costs by storing its data in the most affordable storage layer (Amazon S3) and only using compute resources as needed.
• Improved Parquet performance. With Starburst Enterprise, Glovo achieves an average of 20% improvement in Parquet read performance over Trino.

Choosing Starburst Enterprise allows non-technical data users to query and consume data easily from their favorite data tools. With federated data access, Glovo can run tests on all of its data for continual improvement of its 4-sided marketplace. “Starburst unlocks easier and faster experimentation, and that’s something we needed to optimize our platform,” says Grandi.