qemu-nbd-volume.md

Overview

Components

1. (1) trusted storage connector pod
2. (Many) VM pods connection to storage using shared directory

Flow

1. Get the credentials from a PVC
2. Create a qcow2 pointing to the storage described by PV
3. Export the qcow2 (and thus the backing storage) using nbd over a socket
4. (Other pod) connect to the socket booting the VM

Trusted Storage Connector Pod

# Generate an image pointing to the backing store: http://docs.ceph.com/docs/giant/rbd/qemu-rbd/
# A backing store can be any storage system supported by qemu.
$ qemu-img create -f qcow2 -b rbd:$WHATVER_COMES_FROM_PV qcow-for-my-pvc.qcow2

# Now share the disk all day long
$ qemu-nbd --persistent -k /shared-nbd/my-pvc.sock $PWD/qcow-for-my-pvc.qcow2

VM

Any client can launch a VM using:

$ qemu-system-x86_64 --drive file=nbd:unix:/shared-nbd/my-pvc.sock

vmi.yaml

kind: VMI
spec:
  domain:
    devices:
      disks:
      - name: plugdisk
  volumes:
    plug:  # This is added by a user at RUNTIME using an UPDATE rest call
      name: plugdisk
      persistentVolumeClaim:
        claimName: the-plugged-pv

# The USER modifies the VMI
# The NDB POD is listening for volume/disk changes THEN reserves the PVC, the connects to the PV, then exports the PV over socket
# The launcher/handler is listening for volume/disk changes THEN waits for the socket to come up, then attaches (hot plugs) it to qemu