[Deepsea-users] RFC: Experimental import of existing cluster

Larry Morris lmorris at suse.com
Thu May 25 07:36:01 MDT 2017


The PM was also worried that we would not have enough time to get to this.  Thanks Tim for making the time to look more into this.

While I don’t have the customer data I would like, it is a safe assumption that 1/3 to ½ of our current customers are running clusters that will need this type of process.

Remember, it is all about making SES easier to use than any other Ceph system.

From: deepsea-users-bounces at lists.suse.com [mailto:deepsea-users-bounces at lists.suse.com] On Behalf Of Craig Gardner
Sent: Thursday, May 25, 2017 7:19 AM
To: Discussions about the DeepSea management framework for Ceph <deepsea-users at lists.suse.com>; Tim Serong <TSerong at suse.com>; deepsea-users at lists.suse.com
Subject: Re: [Deepsea-users] RFC: Experimental import of existing cluster

On May 25, 2017 12:48:50 PM GMT+02:00, Tim Serong <tserong at suse.com<mailto:tserong at suse.com>> wrote:

Hi All,

We want DeepSea to be able to "import" existing ceph clusters, for
example if you've used ceph-deploy in the past, and want to migrate to
DeepSea.  I've implemented what you might call a rough outline of this
functionality at https://github.com/SUSE/DeepSea/commit/7a60715 (i.e.
the implementation is incomplete), and would appreciate feedback on the
approach in general.

There's more detail in the commit, but essentially you use it like so:

1) Install salt on every ceph node.
2) Run the first two DeepSea stages, plus my extra importer bit.
3) The importer checks what ceph services are running on all the minions
(mon, mds, osd, rgw), and generates a policy.cfg reflecting the
currently running cluser.
4) Now you can keep using DeepSea as usual.

WARNING: Do *NOT* try step 4 with the current code.  You WILL BREAK YOUR
CLUSTER.  I'm just trying to get feedback on whether there's any large
holes in the general shape of this thing (I had an earlier experiment
where I was interrogating ceph.conf to find the MONs, then realised I'd
end up having to talk to the cluster to ask it where the OSDs were, so I
decided that checking for running services was going to be simpler).

Thanks,

Tim

Oh, this looks very promising. Eric was quite worried that we wouldn't have enough time and too many other priorities to get to this. Thanks for looking into it, Tim.

--
Sent from my Android device with K-9 Mail. Please excuse my brevity.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.suse.com/pipermail/deepsea-users/attachments/20170525/6e775174/attachment.htm>


More information about the Deepsea-users mailing list