Stuart,

It's covered here too:
http://ceph.com/docs/master/faq/#how-can-i-give-ceph-a-try

That comment only applies to the quick start--e.g., someone spinning up a
Ceph cluster on their laptop to try it out. One of the things we've tried
to provide to the community is a way to try Ceph out on the minimum number
of machines possible so that you can get a feel for how it works without
building a production-worthy cluster. The issue with trying Ceph out on
only one machine is that if you have ceph-mon and ceph-osd daemons running
on a host, you really shouldn't try to mount rbd as a kernel object or
CephFS as a kernel object on the same host. It's not related to Ceph, but
rather to the Linux kernel itself. You'd never do this in production. The
admonishment only applies to the quick start.

You can run ceph-gateway on the same host as the OSDs. It's only kernel
mounted clients on older versions of the kernel that have the potential to
deadlock. As always, with running multiple daemons on a host, be cognizant
of the memory and CPU requirements of the various daemons.

Regards,


John



On Tue, Apr 16, 2013 at 12:27 AM, Stuart Longland <[email protected]>wrote:

> Hi all,
>
> I've been doing quite a bit of research and planning for a new virtual
> computing cluster that my company is building for their production
> infrastructure.
>
> We're looking to use OpenStack to manage the virtual machines across a
> small cluster of nodes.
>
> Currently we're looking at having 3 storage nodes each with 2 3TB drives
> running dual Gigabit, 3 management nodes with dual 10GbE network cards
> and about 16 (or more) compute nodes.
>
> The plan is that the management nodes will provide services such as
> Rados Gateway for object storage and volume management with Cinder.  The
> reasoning; using rbd means that we have redundancy of volumes and
> images, something plain LVM (which Cinder would otherwise use) can't
> provide on its own.
>
> What I'm not clear on, is just where each daemon can exist.  ceph-ods
> will obviously be running on each of the storage nodes.  I'd imagine
> ceph-mon will run on the management nodes.  However I read in the
> documentation:
>
> > We recommend using at least two hosts, and a recent Linux kernel. In
> > older kernels, Ceph can deadlock if you try to mount CephFS or RBD
> > client services on the same host that runs your test Ceph cluster.
>
> I presume that the actual rbd client for volume/block storage would
> in-fact be running on the compute node and thus be separate from the ods
> and mon daemons.  The one I'm not clear on is the ceph-gateway.
>
> Does this operate as a client, and thus should run separate from
> ceph-mon/ceph-ods?  Or can it reside on one of the monitor hosts?  Do
> the clients potentially deadlock with ceph-mon, ceph-ods or both?
>
> Apologies if this is covered somewhere, I have been looking on-and-off
> over the last few months but haven't spotted much on the topic.
>
> Regards,
> --
> ##   -,-''''-. ###### Stuart Longland, Software Engineer
> ##.  :  ##   :   ##   38b Douglas Street
>  ## #  ## -'`   .#'   Milton, QLD, 4064
>  '#'  *'   '-.  *'    http://www.vrt.com.au
>      S Y S T E M S    T: 07 3535 9619    F: 07 3535 9699
> _______________________________________________
> ceph-users mailing list
> [email protected]
> http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
>



-- 
John Wilkins
Senior Technical Writer
Intank
[email protected]
(415) 425-9599
http://inktank.com
_______________________________________________
ceph-users mailing list
[email protected]
http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com

Reply via email to