Hi,

thanks for your information and sharing of your experience! One time we had a powerloss for 1 hour. After that the cloud as well as the Ceph cluster where up with absolutely no issues. I think a corrupted Ceph cluster might be result of hardware failure, e.g. using Erasure Coding with to many broken hosts/OSDs. Some other Ceph admins told me they use Object Storage, RBD and CephFS on the same cluster with nearly no effect on performace in their environment.

We also used simple linux machines as NFS servers for Cloudstack. Since we need them to shut down, I am going to setup Ceph FS and NFS on our existing cluster.

Regards,

Mevludin

Am 04.11.2022 um 13:58 schrieb Wido den Hollander:


Op 03-11-2022 om 12:00 schreef Mauro Ferraro - G2K Hosting:
It is a very promising technology, but in our experience, ceph made our life a hell. We had an energy event in our Datacenter and ceph couldnt support it, we can understand that these events may not happend, but shits happends.


That's a totally different experience! We are using multiple Petabytes of Ceph clusters behind our CloudStack environments and this works flawlessly.

A powerloss which causes Ceph to go corrupt is probably a hardware related issue and not Ceph.

Our traditional infrastructure (servers with hardware raid and nfs) supported this event with no major problems. With ceph we had a lot of data loss, we were searching data in the "garbage" (blocks) with rbd tools, months of work, stress of the team and lots of furious clients.

Ceph can support some contingency but if the problem is big it will take a lot of time to recover and in this time you can't put your clusters and VMs to work.

If you will use ceph make sure that you have a mirror working in another site that can run when have a big problem.


El 3/11/2022 a las 06:58, Mevludin Blazevic escribió:
Hi all,

is there someone who is using a ceph cluster both for secondary storage (NFS service) and primary storage (rbd pool)? In my test environment I have experienced some issues starting VMs after updating cloudstack not able to convert images to RBD.


Yes, as said, we have.

Our Primary Storage is all RBD with KVM behind CloudStack.

Secondary Storage comes from simple Linux machines just exposing NFS towards CloudStack.

Wido

Regards,

Mevludin

--
Mevludin Blazevic, M.Sc.

University of Koblenz-Landau
Computing Centre (GHRKO)
Universitaetsstrasse 1
D-56070 Koblenz, Germany
Room A023
Tel: +49 261/287-1326

Reply via email to