Presentation: Anti-Entropy Using CRDTs on HA Datastores @Netflix

Track: Modern CS in the Real World

Location: Pacific DEKJ

Duration: 10:35am - 11:25am

Day of week: Tuesday

Share this on:

Abstract

Entropy that causes replicas to diverge on Highly Available datastores is unavoidable in practice. To deal with this, many systems have anti-entropy mechanisms to offer an eventually consistent view of the data. We will be talking about Dynomite, an open-source distributed datastore primarily backed by Redis, built to be highly available, and how it uses CRDTs to make anti-entropy possible with minimal coordination between replicas, as compared to other mechanisms. This talk will briefly introduce Dynomite, offer a deep dive on how anti-entropy is implemented and talk about the underlying principles of CRDTs that make this possible.

Question: 

What is the work you’re doing today?
 

Answer: 

I work on distributed datastores at Netflix, primarily as a developer on Dynomite, a highly available globally distributed datastore backed by Redis. Dynomite is typically used by globally distributed services within Netflix that have require low-latency with varying levels of consistency.

Question: 

What are your goals for the talk?

Answer: 

A fundamental challenge with a highly available datastore is data consistency across replicas. Traditionally, systems have had a number of ways of dealing with it, the most important of them being having a repair process that takes care of repairing replicas out of sync. This talk aims to show how we can use CRDTs to implement a self healing system for data-structures natively supported by Redis and the practical challenges involved.

Question: 

What do you want people to leave the talk with?

Answer: 

I hope to leave the audience with a better understanding of CRDTs after the talk, how they may identify opportunities where the application of CRDTs are a "good fit", and hopefully with this understanding, how they may apply it to their own use cases.

Question: 

What do you think is the next big disruption in software?

Answer: 

Having worked as a developer on databases and also supporting managed databases for a while, I believe the next big disruption in this space is having self-managed databases that gets rid of a lot of the mundane administration tasks typically done by humans today. However, this will definitely take a while to achieve with possible incremental progress; just as self-driving cars may slowly be able to take over mundane driving tasks but not be adequate for complicated scenarios, so will databases slowly shed the need for human intervention as systems become smarter and learn to better manage themselves. Professor Andy Pavlo is the leading researcher in this space and it's quite exciting to track his group's progress in this regard.

Speaker: Sailesh Mukil

Senior Software Engineer @Netflix

Sailesh Mukil is a Senior Software Engineer at Netflix where he works on Dynomite, a highly available in-memory database backed by Redis. Prior to Netflix, he worked at Cloudera where he contributed to the Apache Impala and the Apache Kudu projects, in his capacity as a PMC and a committer to both. He received his graduate degree from Carnegie Mellon University and is an open source enthusiast.

Find Sailesh Mukil at

Similar Talks

Evolution of Edge @Netflix

Qcon

Engineering Leader @Netflix

Vasily Vlasov

Mistakes and Discoveries While Cultivating Ownership

Qcon

Engineering Manager @Netflix in Cloud Infrastructure

Aaron Blohowiak

Monitoring and Tracing @Netflix Streaming Data Infrastructure

Qcon

Architect & Engineer in Real Time Data Infrastructure Team @Netflix

Allen Wang

Future of Data Engineering

Qcon

Distinguished Engineer @WePay

Chris Riccomini

Coding without Complexity

Qcon

CEO/Cofounder @darklang

Ellen Chisa

Holistic EdTech & Diversity

Qcon

Holistic Tech Coach @unlockacademy

Antoine Patton

Tracks

Monday, 11 November

Tuesday, 12 November

Wednesday, 13 November