Bug 2123341

Summary:	ODF CSI RBD VolumeSnapshot are taking more space than what they're using
Product:	[Red Hat Storage] Red Hat OpenShift Data Foundation	Reporter:	Pablo Rodriguez Guillamon <parodrig>
Component:	csi-driver	Assignee:	Niels de Vos <ndevos>
Status:	CLOSED DEFERRED	QA Contact:	krishnaram Karthick <kramdoss>
Severity:	medium	Docs Contact:
Priority:	unspecified
Version:	4.8	CC:	assingh, madam, ndevos, ocs-bugs, odf-bz-bot, rar
Target Milestone:	---
Target Release:	---
Hardware:	Unspecified
OS:	Unspecified
Whiteboard:
Fixed In Version:		Doc Type:	If docs needed, set a value
Doc Text:		Story Points:	---
Clone Of:		Environment:
Last Closed:	2022-09-13 13:58:05 UTC	Type:	Bug
Regression:	---	Mount Type:	---
Documentation:	---	CRM:
Verified Versions:		Category:	---
oVirt Team:	---	RHEL 7.3 requirements from Atomic Host:
Cloudforms Team:	---	Target Upstream Version:
Embargoed:

Comment 6 Niels de Vos 2022-09-09 15:32:14 UTC

Trying to explain things a little differently, maybe that makes things clearer.

Copy-On-Write (COW) works like this:
1. a snapshot is taken at a certain time, and refers to the blocks that an rbd-image consumes (takes not much space)
2. the rbd-image changes, the snapshot is now the only owner of the blocks that were changed (takes up more space)

So, when the rbd-image makes modifications to its blocks, the space consumption of the snapshot grows.


This initial snapshot requires almost no space, even if the rbd-image contained lots of data already. The snapshot is not a self-containing full copy of the rbd-image (it becomes one when it is flattened). A snapshot tracks what changes to the original rbd-image were done, and stores those changes. If over time, the number of changes increase, and the snapshot grows. You can see a snapshot as a large diff between the state of the rbd-image when the snapshot was made, and the latest version of the rbd-image with all its modifications.

Comment 10 Niels de Vos 2022-09-13 13:58:05 UTC

The space consumption is expected behaviour. This comes from the Kubernetes/CSI requirement that VolumeSnapshots must be an independent object; it should be possible to delete the original volume, and restore from the snapshot.

The current design and implementation in Ceph-CSI is documented at https://github.com/ceph/ceph-csi/blob/39b1f2b4d3b79237ffa956ff16a3459cf21bf63f/docs/design/proposals/rbd-snap-clone.md#create-a-snapshot-from-pvc


As a feature request, it should be possible (with added complexity) to create dependent RBD-snapshots (only storing the diff between time of snapshot and current RBD-image), without the cloning step. This should reduce the data consumption quite a bit. Ceph-CSI will need to get extended to do accounting of RBD-image consumers, so that an RBD-image only is deleted if no PersistentVolume or VolumeSnapshotSource references it (indirectly).

As this is a feature request, and we have it tracked at https://issues.redhat.com/browse/RHSTOR-3799 , I will close this BZ now. For further release planning and progress, check the linked issue.