Explore how Yelp.com leverages Apache Mesos and AWS Spot Fleet to optimize infrastructure costs and operational efficiency in this 36-minute Linux Foundation conference talk. Dive into the intricacies of balancing significant cost savings with potential operational risks associated with AWS Spot Instances. Learn about Yelp's custom configuration tweaks for Mesos, Marathon, Chronos, and autoscalers that ensure reliable infrastructure. Discover the power of Terraform in managing Spot Fleet, best practices for diversification, bidding strategies, and maintenance primitives. Gain insights into the cost-benefit analysis of using Spot Fleet and understand the key considerations for implementing this approach in your own infrastructure. Follow along as Kyle Anderson, a Site Reliability Engineer at Yelp, shares valuable lessons and practical tips for running a major web platform on theoretically unstable yet highly cost-effective infrastructure.
How Yelp Runs on Apache Mesos in AWS Spot Fleet - Balancing Cost Savings and Operational Risk
Overview
Syllabus
Introduction
Agenda
Spot Fleet Definitions
Spot Instances
Spot Price
What is Spot Fleet
Why is Spot Fleet important
Spot Fleet Web Interface
Spot Fleet CLI
The JSON Problem
Terraform
Why use Terraform
Spot Fleet Instance Types
Why We Use Terraform
Terraform Demo
Best Practices
Diversification
Diversification vs Lowest Price
Diversification vs Rebalance
AZ Instance Types
Diversification Weighting
Diversity in Action
How to Achieve Diversity
Easy Balancing
Bidding
Choosing the Right Bid
Dealing with Terminations
Outbidding Yelp
Maintenance Primitives
Marathon
Outbid
Maintenance
Recap
Is it worth it
I ran the numbers
Spot Fleet is saving us tons
Spot Fleet is saving us 50
Shout out to the early adopters
What you need to know
GitHub
Costbenefit analysis
Taught by
Linux Foundation
