"Why is this such shit?"
"oh it's 'fixed' in the newest version of kubernetes"
Cries remember how much of a pain upgrading our clusters and migrating resources is.
@groved the major k8s is designed to assume all applications are self contained and self healing so you can reboot/upgrade/destroy any node at any time and have 0 downtime.
K8s with applications designed to be pinned to a single snowflake box is the worst of both worlds.
@lrvick that only works if you have fully stateless applications. If you keep any data in clusters it's always a pain to move. Especially when you're talking about databases
@groved This is why data should not be kept directly in the clusters but on external mounted volumes that you can detach and re-attach wherever you need them, even to new servers after the old ones were burned down. EBS, S3, NFS, SSHFS, etc etc.
@lrvick EBS doesn't support "ReadWriteMany" and is the default for the EKS CSI (including the previous iteration). This is part of the problem. It can only be mounted to a single node and can only be accessed by services on th nodeoses. This makes rolling upgrades impossible and can also lead to scaling issues
@groved So what you are saying is that EKS is badly designed, and can't properly take advantage of the core value propositions of k8s that rely on stateless/disposable nodes? That checks out.
@lrvick I mean, this problem plagues all stateful applications across most providers. DOKS runs into the same problem as they also only support RWO. You can set up your own RWX by making an NFS server IN your cluster or on a different droplet, but you're still dealing with manual migration issues none the less.
This Mastodon instance is for people interested in technology. Discussions aren't limited to technology, because tech folks shouldn't be limited to technology either!