we’ve integrated LXC into Kubernetes to handle stateful services in a way which works for us on a bigger scale. We called that implementation “LXE”, cause the management daemon of LXC is calles LXD.
There are 2 main reasons to do that:
- we have the policy: always online - restart is not an acceptable side effect (i.e. in case of updating docker daemon or bug in docker daemon)
- we want muteable containers to avoid container restarts as much as possible (see 1)
My question is, if there is a community need for that and on the other side if a contribution for the CRI spec itself is welcome. For us it’s the only way to handle stateful services on a bigger scale in a good way.