OpenStack Compute (nova)

libvirt: better performance for resizing backing files

Bug #1087031 reported by Vish Ishaya on 2012-12-05

This bug affects 1 person

Affects		Status	Importance	Assigned to	Milestone
	OpenStack Compute (nova)	Fix Released	Low	Michael Still	OpenStack Compute (nova) 2013.1 "grizzly"

Bug Description

Currently backing files are resized (from <uuid> to <uuid>_40 for example) using a copy, which is extremely slow. It would be nice to either:

a) allow resizing by making the <uuid>_40 a qcow image with <uuid> as a backing file and performing a resize on the qcow.

b) allow the disk file to directly back to <uuid> and perform the resize on disk directly

both of these have some concerns when relating to image cache cleanup, but overall option b) seems far better. It also makes resizing the image later much easier.

Vish Ishaya (vishvananda) on 2012-12-05

Changed in nova:
assignee:	nobody → Michael Still (mikalstill)
status:	New → Triaged
importance:	Undecided → Low

OpenStack Infra (hudson-openstack) on 2012-12-17

Changed in nova:
status:	Triaged → In Progress

Revision history for this message

OpenStack Infra (hudson-openstack) wrote on 2012-12-22: Fix merged to nova (master)

Reviewed: https://review.openstack.org/18276
Committed: http://github.com/openstack/nova/commit/c82be9f4b997d252cfc84350a067a64d6d4b380b
Submitter: Jenkins
Branch: master

commit c82be9f4b997d252cfc84350a067a64d6d4b380b
Author: Michael Still <email address hidden>
Date: Mon Dec 17 14:56:19 2012 +1100

libvirt: Skip intermediate base files with qcow2.

    Instead of having two files in _base (the original and a resized
    copy), let's just keep the originals and resize with the qcow2
    image in the instance's directory. This will reduce the size of
    _base and simplify cleanup. This also simplifies resizing of
    instance disk images later.

Resolves bug 1087031.

Change-Id: Id91426e3cb9f75f31339b5156785e3782a4cb98f

Changed in nova:
status:	In Progress → Fix Committed

Thierry Carrez (ttx) on 2013-01-09

Changed in nova:
milestone:	none → grizzly-2
status:	Fix Committed → Fix Released

Revision history for this message

Pádraig Brady (p-draigbrady) wrote on 2013-02-25:

I've a query about the change.

Now resizes are always done on qcow file in the instance directory.
That will introduce a possibly unexpected change when using
unpartitioned ext[234] images, as they'll no longer have their
file system resized up (as resize2fs will fail on the qcow2 file).

Maybe we document this in release notes, that the limited
file system resizing is no longer supported?
Or perhaps we key this new behaviour on whether the
image in base could be resized (disk.can_resize_fs(_base_image, size))

Thierry Carrez (ttx) on 2013-04-04