Hey there,

On 06/03/2023 22:34, Nguyễn Hữu Khôi wrote:
>
> I am looking for instance backup solution. I am using Cinder backup
> with nfs backup but it looks not too fast. I am using a 10Gbps
> network. I would like to know experience for best practice for
> instance backup solutions on Openstack.


On 13/03/2023 12:46, Eugen Block wrote:
> We use Ceph as back end for all services (nova, glance, cinder), and
> the most important machines are backed up by our backup server
> directly via rbd commands:

There is RBD and "the other" drivers. While RBD uses the native export /
import feature of Ceph, all other drivers (file, NFS, object storages
like S3) are based on the abstract chunked driver
(https://opendev.org/openstack/cinder/src/branch/master/cinder/backup/chunkeddriver.py).
This driver reads the volume / image and treats it as chunks before
making use of a concrete driver (e.g. NFS or S3) to send those chunks
off somewhere to be stored. Restore works just the opposite way. The
performance of the chunked driver based back-ends is not (yet)
comparable to what RBD can achieve due to various reasons.

But again, while "RBD" uses Ceph's mechanisms internally all other
"targets" for backup storage work differently.
We ourselves were looking into using and S3-compatible storage and thus
I started a dicsussion about the state of those other drivers at
https://lists.openstack.org/pipermail/openstack-discuss/2022-September/030263.html

This then led to a discussion at the Cinder PTG
https://etherpad.opendev.org/p/antelope-ptg-cinder#L119 with many
observations.

There also are changes in the works, like restore into sparse volumes
(https://review.opendev.org/c/openstack/cinder/+/852654) when going via
the chunked driver.
But also features like "encryption"
(https://review.opendev.org/c/openstack/cinder-specs/+/862601) are being
discussed.



Regards


Christian