Diskless Nodes

Registered by Julian Edwards

Support for network-mounted root filesystems on nodes

Blueprint information

Status:
Complete
Approver:
Daniel Westervelt
Priority:
Undefined
Drafter:
None
Direction:
Approved
Assignee:
None
Definition:
Obsolete
Series goal:
None
Implementation:
Informational Informational
Milestone target:
None
Completed by
Adam Collard

Related branches

Sprints

Whiteboard

Diskless nodes
==============

Prior work: https://docs.google.com/a/canonical.com/document/d/1ux5-jSct0ZSCye3s4SHD739ABZV9Ycd8897mNgKbRX0/edit

Network-mounted root filesystems.

Ceph now has a tgt frontend. But that's slow.

Why?
 * Because ARM
 * Use existing nodes as compute nodes without wiping out the disks.
 * Simple root disk re-use
 * Less hardware in nodes

Options:
 * Use cinder as per "prior work" above (cinder is a provisioning api, still need to use tgt)
 * Boot with ephemeral image r/o and then use cinder in overlayfs mode to store write deltas

Questions:
 * can we mount ceph volumes natively as a root filesystem? (smoser: yes, via userspace)

Cool stuff:
 * could use ceph base images for ephemeral booting which saves cluster from having to download images

(?)

Work Items