Zero-Downtime Live Migration of Stateful VMs on Kubernetes - Felicitas Pojtinger, Loophole Labs

Zero-Downtime Live Migration of Stateful VMs on Kubernetes - Felicitas Pojtinger, Loophole Labs

If you’ve worked with VMs on Kubernetes, you’re familiar with how far they are behind the rest of the container ecosystem. This is caused by the different requirements around the way that VMs are started and migrated, which doesn’t match up with way we handle these things in the Kubernetes world. Outside of Kubernetes, existing legacy hypervisor solutions require VMs to be shutdown before migration, resulting in significant downtime. Even if they support live migration, that process is usually slow, error-prone, requires extensive preparation, and cuts networking during migration. These challenges make it difficult for legacy, stateful applications to leverage Kubernetes. This talk will dive into how we use r3map to solve these issues by providing stateful applications with the same UX that developers expect from containers by speeding up migrations, enabling persistent network connections, and guaranteeing zero downtime.

[tomaszsuchorowski]

when I saw VM live-migration the first time, it was done between the racks in a DC and myself not losing ssh session while it was executed, I thought it is some sort of magic ;-) But this.. this is just wow, a completely next level of sorcery^N ! How is this even possible? that whitepaper mentioned at the end of the talk, does not explain how on earth the clients didn't lose network connectivity over Internet to the server migrated between the continents?