Service abstractions to cloud service providers: A tale of trade-offs

 

Mercado Libre, the largest e-commerce, fintech & logistics company in LATAM with over 110 million active users, more than 26K microservices, and 50K data services, heavily relies on core platform services such as databases, pub-sub messaging, data streams, big data, and search engines. For most of these, we built abstractions to cloud services providers that play an essential role regarding our priorities, like uptime, compliance, and efficiency. This extra layer shortens the learning curve and allows the development teams to focus on running their business rather than delving into the peculiarities of a cloud vendor and its idiosyncratic interfaces, improving their productivity and reducing the overall time-to-market. However, there are always trade-offs, and, for some scenarios, we chose to use the providers' services directly.Our abstractions deliver excellent flexibility in supporting maintenance and evolution since it maximizes the reuse potential of these interfaces and keeps the implementation details isolated in one place. So, for example, although we had thousands of microservices relying on our key-value store abstraction, we could evolve it to take full advantage of new technologies without requiring changes to any of those services. In addition, they also leverage the adoption of strategies like Multi-Availability Zones, Multi-Region, Multi-Account, and Multi-Cloud, increasing the resiliency to outage events and helping the organization comply with the broad range of regulatory and governance mandates. For example, during an incident with one of our cloud providers, we could transparently move our message broker abstraction, which is responsible for delivering billions of messages per minute to thousands of microservices, to another provider and thus mitigate the impact that had on our business. Cloud service provider abstractions brought significant competitive advantages to Mercado Libre regarding governance, resiliency, flexibility and efficiency. Considering the current potential economic volatility, when companies are bracing for a harsh reality ahead and freezing or reducing staffing for low-priority projects, efficiency has become an even more critical concern.n this session, we will present the different strategies we utilized to build our cloud providers' abstractions, discuss the trade-offs associated with each of them, and tell some of the unique challenges we faced and the lessons we learned along the way. In this session, we will present the different strategies we utilized to build our cloud providers' abstractions, discuss the trade-offs associated with each of them, and tell some of the unique challenges we faced and the lessons we learned along the way.


Speaker

Oscar Mullin

Sr. Tech Director and Head of Core Platform Services, Databases, Operational Excellence, and SRE @Mercado Libre

Sr. Tech Director and Head of Core Platform Services, Databases, Operational Excellence, and SRE at MercadoLibre. He has 20+ years of experience working in Technology in different size organizations around the world, working with many different technologies and business domains.

Read more

Session Sponsored By

Mercado Libre is the #1 e-commerce technology and financial services company in Latin America

Date

Monday Oct 24 / 01:40PM PDT ( 50 minutes )

Share

From the same track

Session

Solutions Track Session 1

Details coming soon.

Session

Our journey into high performance and reliable document databases with RavenDB

Monday Oct 24 / 04:10PM PDT

When I started at Kobo, we needed to look beyond the relational and into document databases.

Trevor Hunter

Chief Technology Officer @Kobo Inc.

Session

Optimal Data Storage Choices - Data Lakes vs Databases

Monday Oct 24 / 05:25PM PDT

 

Greg Hamer

Data and Application Architecture Specialist @Backblaze

Session

Building Agile Data Architectures in Support of Digital Twins and Data Products

Monday Oct 24 / 02:55PM PDT

Agile software development and elastic cloud foundations have enabled on-demand expansion of compute functions from real-time processing to Machine Learning at scale but Data has been left behind.

Stuart Sim

Leader @Build by McKinsey

Session

Is Web3 Here to Stay?

Monday Oct 24 / 11:50AM PDT

You may be familiar with the current reputation of web3, but are you up to date on the advantages of distributed ledgers applied in the real world? Join us for a deeper dive into how companies are using this innovative technology today.

Richard Bair

VP of Software Engineering @Hedera

Session

Bringing green, sustainable software solutions into the enterprise

Monday Oct 24 / 10:35AM PDT

 

Adam Jordan

Distinguished Software Engineer @Shell