Re: [PATCH] make mkcephfs and init-ceph osd filesystem handling more flexible

[Xiaopong Tran <xiaopong.tran@gmail.com>]

On 08/10/2012 09:53 AM, Sage Weil wrote:
> On Thu, 9 Aug 2012, Tommi Virtanen wrote:
>> On Thu, Aug 9, 2012 at 10:03 AM, Danny Kukawka <danny.kukawka@bisect.de> wrote:
>>> So you mean chef?! Will there be an alternative to simply setup a
>>> cluster from console?
>>>
>>> We (SUSE) are already working on an own chef ceph cookbook. But from
>>> what I've seen till now it's really hard and more laborious to initially
>>> setup a cluster with chef than with mkcephfs.
>>
>> I've written about this on the mailing list several times. We see a
>> lot of demand for Chef, but don't want to tie our hands -- Canonical
>> is working on Juju Charms, and I would like to see a mkcephfs
>> replacement that relies on just SSH connections from a workstation
>> node. We've made an explicit effort to improve the product as a whole,
>> and to make the Chef cookbook as thin as possible.
>>
>> For some reason, the threading is broken on the archive, but this is a
>> fragment of the most recent thread that talked about this:
>>
>> http://thread.gmane.org/gmane.comp.file-systems.ceph.devel/8263/focus=8265
>
> I think the real question is whether the planned "mkcephfs 2.0" is going
> to capture the equivalent functionality of being able to enumerate up
> front which osds will exist and which disks/journals they will use, and to
> bring them up.  Assuming it will (IMHO it needs to), can we make that
> compatible with the current way that mkcephfs is invoked (i.e., a
> ceph.conf file and a few command line args)?  Not everyone (and I daresay
> probably a minority) of users will be using Chef/Juju/Puppet/whatever,
> regardless of whether or not we feel that is the right way to do things
> and try to push them in that direction.
>
> In any case, since the new osd hotplugging stuff isn't available now and
> is probably still a sprint or two off, I think we should consider applying
> this patch.  We aren't recommending btrfs across the board, and cluster
> bringup is currently painful on ext4/xfs/etc.  And, Danny already did the
> work.  :)
>
> sage
> --

+1.

Rgds,

Xiaopong