During the last year, at Spotify we’ve been building a Cloud distributed cache service with sharding and replication running in Kubernetes. Client services use our own custom Java library on top of the open source spotify/folsom Java memcached client.
The aim of this project was twofold: To reduce costs and improve the developer experience and productivity. Now the service is serving literally millions of requests per second for some tens of microservices… but the road has been bumpy at times. In this talk I’ll share some surprising learnings about cloud-distributed systems at scale.