I love when engineering teams share their tricks of trade for other organizations to benefit. While this might seem counter-intuitive, sharing knowledge makes the entire ecosystem better.
Etsy‘ engineering team does a great job of publishing their architecture, methodologies and code at https://codeascraft.com.
This particular article on how they optimize their caching infrastructure (https://codeascraft.com/2017/11/30/how-etsy-caches/) is pretty enlightening. I always thought the best method to load balance objects (app hits, cache requests, queues etc) to hosts was to use mod operations. In this blog post Etsy’ team talk about using consistent hashing instead of modulo hashing.
At a high level, it allows cache nodes to fail and not impact the overall performance of the application drastically in addition to making it easy to scale the number of nodes. This method is useful when you have a large amount of cache nodes.
More reference links
- http://www.tom-e-white.com/2007/11/consistent-hashing.html
- https://www.toptal.com/big-data/consistent-hashing
- https://en.wikipedia.org/wiki/Consistent_hashing
