Which deployment method for ceph (rook|ceph-ansible|tripleo)

Goutham Pacha Ravi gouthampravi at gmail.com
Mon Nov 9 16:39:10 UTC 2020

On Tue, Nov 3, 2020 at 12:11 PM Oliver Weinmann <oliver.weinmann at me.com>

> Hi Tom,
> Thanks a lot for your input. Highly appreciated. John really convinced me
> to deploy a
> Standalone cluster with cephadm and so I started to deploy It. I stumbled
> upon a few obstacles, but mostly because commands didn’t behave as expected
> or myself missing some important steps like adding 3 dashes in a yml file.
> Cephadm looks really promising. I hope that by tomorrow I will have a
> simple 3 node cluster. I have to look deeper into it as of how to configure
> separate networks for the cluster and the Frontend but it shouldn’t be too
> hard. Once the cluster is functioning I will re-deploy my tripleo POC
> pointing to the standalone Ceph cluster. Currently I have a zfs nfs Storage
> configured that I would like to keep. I have to figure out how to configure
> multiple backends in tripleo.

Is the ZFS NFS storage going to be managed by the ZFSOnLinux driver in
Manila? Or is this Oracle ZFSSA?
Configuring integrated backends [2] in TripleO is fairly straightforward -
but neither of the drivers above are integrated into TripleO, you'll have
to use *ExtraConfig steps to shove the config into manila.conf:

Something along these lines: http://paste.openstack.org/show/799832/

> Cheers,
> Oliver
> Von meinem iPhone gesendet
> > Am 03.11.2020 um 16:20 schrieb Tom Barron <tpb at dyncloud.net>:
> >
> > On 01/11/20 08:55 +0100, Oliver Weinmann wrote:
> >> Hi,
> >>
> >> I'm still in the process of preparing a OpenStack POC. I'm 100% sure
> that I want to use CEPH and so I purchased the book Mastering CEPH 2nd
> edition. First of all, It's a really good book. It basically explains the
> various methods how ceph can be deployed and also the components that CEPH
> is build of. So I played around a lot with ceph-ansible and rook in my
> virtualbox environment. I also started to play with tripleo ceph
> deployment, although I haven't had the time yet to sucessfully deploy a
> openstack cluster with CEPH. Now I'm wondering, which of these 3 methods
> should I use?
> >>
> >> rook
> >>
> >> ceph-ansible
> >>
> >> tripleo
> >>
> >> I also want to use CEPH for NFS and CIFS (Samba) as we have plenty of
> VMs running under vSphere that currently consume storage from a ZFS storage
> via CIFS and NFS. I don't know if rook can be used for this. I have the
> feeling that it is purely meant to be used for kubernetes? And If I would
> like to have CIFS and NFS maybe tripleo is not capable of enabling these
> features in CEPH? So I would be left with ceph-ansible?
> >>
> >> Any suggestions are very welcome.
> >>
> >> Best Regards,
> >>
> >> Oliver
> >>
> >>
> >
> > If your core use case is OpenStack (OpenStack POC with CEPH) then of the
> three tools mentioned only tripleo will deploy OpenStack.  It can deploy a
> Ceph cluster for use by OpenStack as well, or you can deploy Ceph
> externally and it will "point" to it from OpenStack.  For an OpenStack POC
> (and maybe for production too) I'd just have TripleO deploy the Ceph
> cluster as well.  TripleO will use a Ceph deployment tool -- today,
> ceph-ansible; in the future, cephadm -- to do the actual Ceph cluster
> deployment but it figures out how to run the Ceph deployment for you.
> >
> > I don't think any of these tools will install a Samba/CIFs gateway to
> CephFS.  It's reportedly on the road map for the new cephadm tool. You may
> be able to manually retrofit it to your Ceph cluster by consulting upstream
> guidance such as [1].
> >
> > NFS shares provisioned in OpenStack (Manila file service) *could* be
> shared out-of-cloud to VSphere VMs if networking is set up to make them
> accessible but the shares would remain OpenStack managed.  To use the same
> Ceph cluster for OpenStack and vSphere but have separate storage pools
> behind them, and independent management, you'd need to deploy the Ceph
> cluster independently of OpenStack and vSphere. TripleO could "point" to
> this cluster as mentioned previously.
> >
> > I agree with your assessment that rook is intended to provide PVs for
> Kubernetes consumers.  You didn't mention Kubernetes as a use case, but as
> John Fulton has mentioned it is possible to use rook on Kubernetes in a
> mode where it "points" to an external ceph cluster rather than provisioning
> its own as well.  Or if you run k8s clusters on OpenStack, they can just
> consume the OpenStack storage, which will be Ceph storage when that is used
> to back OpenStack Cinder and Manila.
> >
> > -- Tom Barron
> >
> > [1]
> https://www.youtube.com/watch?v=5v8L7FhIyOw&list=PLrBUGiINAakNCnQUosh63LpHbf84vegNu&index=19&t=0s&app=desktop
> >
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack-discuss/attachments/20201109/0a3b5434/attachment-0001.html>

More information about the openstack-discuss mailing list