[Openstack-devel] Debstack instead of Devstack?

[Daniel Pocock]

On 08/02/13 14:33, Thomas Goirand wrote:
> On 02/08/2013 08:41 PM, Daniel Pocock wrote:
>   
>> I've had a look over devstack and also the Folsom howto on the Debian wiki
>>     
> I wouldn't recommend using devstack. That's for developers only, using
> sources, with everything setup in /opt. That's really not suitable for
> production, and it will be hard to debug problems that you *will* face
> (it's using the latest version from github, which must be filled with
> bugs at this point of the release cycle (eg: in the middle of it)).
>
>   

That was my impression - hence my suggestion that I would prefer a
`DebStack' instead

>> Can anyone comment on how to get a devstack-like single box up and
>> running the Debian way, using Xen/wheezy/Folsom? Can the approach in the
>> howto be used?
>>     
> If you want a fast setup on a single box, simply do:
>
> apt-get install openstack-toaster
>
> using Folsom and what's on the wiki (eg: my unofficial Debian repo). And
> if you don't want to bother answering the Debconf questions, then you
> can do some pre-seeding. I've started writing a preseeds script here:
>
> http://archive.gplhost.com/misc/automagical_openstack.txt
>
> but I haven't finished it. Though for anyone doing a bit of Debian
> packaging that includes some debconf and dbconfig-common stuff, it
> shouldn't be hard to understand. I'm sure you will! :)
>
> Also, this is for KVM, not for Xen, which is a lot more complicated to
> setup. If you're planning on using XCP and Openstack, then you should
> read what's in /usr/share/doc/nova-xcp-plugins/README.xcp_and_openstack
> which will only give you an overview of it, and which I wrote a year
> ago, so it must be outdated for Folsom (for example: no need to bother
> with setting-up databases and MySQL access rights, since that part is
> now automated with dbconfig-common ...).
>
>   
Ok, I will have a look at that and see which way to go

>From my initial impression of XCP: it appears to be more than just
tools, even the disk images are different.  Can VMs still run from raw
LVs on the dom0?  Or does it insist on having them wrapped in the XVA
format?  Or can that be worked around by using iSCSI or other abstractions?

I found a few mailing list posts about converting from xend/xm to XCP,
they mention things like converting the filesystems and also changing
device names to /dev/xvd[abcd] - can you make any Debian-specific
comments on this, particularly in the context of upgrades from lenny or
squeeze to wheezy?

>> The box I have in mind is a DL360 G6, 32GB and 4 fast disks - should
>> this be sufficient for a combined proxy node + compute node + cinder +
>> glance all-in-one? Anything I learn from doing this I'm happy to
>> contribute back in scripts or the wiki.
>>     
> Yes, that's more than enough. For a test machine, just to try Openstack,
> I believe that a single HDD, and 4GB of RAM is enough.
>
> If you plan on using Cinder, make sure you do smart partitioning though.
> Cinder uses LVM, while compute and Glance will write images in /var,
> which needs to be big enough. So, with 4 disks, I'd go for LVM over
> RAID10, and have your /var on the LVM, so you can resize everything as
> you wish / as needed.
>
>   

I saw one comment suggesting the Cinder needs it's own VG, is that
true?  In a dev/test server, is it OK to just put all disks in a single
VG and give some of the space to Cinder as required?

> Last word: yes! packaging contributions, wiki enhancement, etc. are more
> than welcome! Thanks. :)
>
>   
Ok, great.  I'll have to see whether I make it work first, I don't want
to lead people over the edge of a cliff.