Kubernetes Everywhere: Lessons Learned From Going Multi-Cloud


Authors:   Niko Smeds


Lessons learned from deploying workloads on a new cloud provider
  • Use infrastructure as code and version control from early on in the project
  • Expect to iterate and plan around limitations
  • Don't blindly copy solutions from previous providers, look for possible improvements
  • Load test applications to discover issues before customers do
  • Be aware of quotas when scaling up nodes
The speaker discovered poor disk performance and Docker Hub rate limits during load testing of a deployed product. They fixed the issues by updating storage classes and installing a Docker registry mirror. They also emphasized the importance of load testing to discover issues before customers do and to benchmark performance between different providers.


Many companies are interested in deploying their products across multiple cloud providers, but few actually see it through. While benefits like avoiding provider lock-in and increased uptime during provider outages are attractive, several factors are important to consider. Grafana Labs successfully deployed across AWS, Azure, DigitalOcean, GCP, and Linode in more than 30 regions; that includes inter-cloud network connections. This talk will explore some of the large and subtle differences in networking and managed Kubernetes services between said providers. We’ll discuss the approaches we took while scaling our infrastructure across multiple environments, the challenges we faced, and what worked in the end.Click here to view captioning/translation in the MeetingPlay platform!


Post a comment