|
|
|
|
|
|
by majewsky
3035 days ago
|
|
|
The short answer is: It's pretty ugly. I don't know the details, but we PXE-boot and install CoreOS, then run kubelet using rkt (using a systemd service). The other k8s components (etcd, apiserver, controller manager etc.) are managed by the kubelet using static manifests. Persistent volumes are backed by a separate storage appliance via NFS. Our team built this entire process ~2 years ago, when we started using k8s. We would probably use some off-the-shelf parts today, but practically nothing existed back then. |
|
|