Balancing redundancy and HA with costs: did you really need all N replicas?AKA We were running what and it cost us how much?! With Ren Lee SRE at Arista Networks

Key takeaways:

“Lazy but Simple” vs. “Proactive but Expensive” methods of scaling: knowing when to pay the seemingly scarier price of running infrastructure than costing engineering time, and vice versa

Hidden costs: cost of bad deployments and things that just don’t work

When autoscaling becomes the demon: especially in public cloud environments when access to pools of resources is no longer your barrier

Abstract:
In an engineer’s ideal world we would love all the resources and redundancies we can possibly get for our services and infrastructure that supports them for sanity and of course, HA. However, how do you balance between “enough” redundancy and the actual operational costs of supporting such engineering choices, and what are some of the tough engineering decisions that need to be made? This talk focuses primarily on services being run on Kubernetes (or public cloud offering of Kubernetes), but the principles can be extended to any infrastructure environment.

Key Topics: capacity planning, cost management, distributed services

Bio:
Ren is an SRE at Arista Networks for CloudVision services team. Deeply passionate about fixing broken things without anyone noticing and using effective monitoring to preempt potential disasters. Wrangler of services that run on Kubernetes to keep the zoo running any day, every day.

Join our slack:
https://join.slack.com/t/dokcommunity/shared_invite/zt-g3ui5r0g-jDKz5dhh2W1ayElqwKYYAg
Follow us on Twitter: @dokcommunity

Connect with Demetrios on LinkedIn:
https://www.linkedin.com/in/dpbrinkm/

Connect with Ren on Linkedin:
https://www.linkedin.com/in/therendeye/

This meetup is sponsored by MayaData, which helped start the DOK.community and remains an active supporter. MayaData sponsors two Cloud Native Computing Foundation (CNCF) projects, OpenEBS - the leading open-source container attached storage solution - and Litmus - the leading Kubernetes native chaos engineering project, which was recently donated to the CNCF as a Sandbox project. As of June 2020, MayaData is the sixth-largest contributor to CNCF projects. Well-known users of MayaData products include the CNCF itself, Bloomberg, Comcast, Arista, Orange, Intuit, and others. Check out more info at https://mayadata.io/

#4 DoK community: The problem of stateful workloads - balance of keeping data HA vs. costs // Ren Lee 59:25 Share

Data on Kubernetes Community

Shownotes Transcript

#4 DoK community: The problem of stateful workloads - balance of keeping data HA vs. costs // Ren Lee