[Gluster-infra] Configuration management, part 2

[Michael Scherer]

Le lundi 01 septembre 2014 à 06:15 +0100, Justin Clift a écrit :
> On 01/09/2014, at 5:44 AM, Joe Julian wrote:
> > Saltstack is python. 😀
> 
> Yep I know.  I'm not against the idea.
> 
> Ansible has a reputation for being pretty reliable. 

Problem is that ansible is a bit like a toolbox, you need to construct
your own system with it. I did one for manageiq, but I prefer when stuff
are already setup.

>  Several people that
> use Salt a lot (~12 months ago) have said it's very buggy, and advised
> not to use it (then) for stuff where reliability is key.
> 
> It may have improved significantly since then.  I really don't know. :)

My main problem with salt is that the fedora package move a lot. While I
have nothing against fast moving target, I would prefer a more stable
things. 

I did use salt, albeit mostly as POC in the past.
So the contender we have are :
- puppet
- ansible
- salt

The admins who spoke on this thread are:
- Louis
- Justin
- Michael
- Joe

( and James was mentioned, but I do not know who is it )

So we have:
Knowing Puppet:
Louis, Michael, Joe, James

Knowing Ansible:
Michael

Knowing Salt:
Joe, Michael ( albeit not much )

I tend to share the concern of Justin regarding Ruby, especially since
we have a bit old stuff to manage like RHEL 5 so despites having a
strong team around puppet, I am not sure we should follow this road. 
The transition on puppet3/ruby2 was quite problematic, and I think may
still cause issues with the diverse set of servers we have, regarding
the need to have some sort of synced server/client ( a problem that
ansible do avoid, and salt too in some way due to salt-ssh, albeit being
alpha )

Not to mention that puppet do not provides remote command orchestration,
and I think mcollective setup on RHEL/Fedora is a bit buggy at the
moment ( due to activemq package being just a jar dump without any
service files ). I would prefer something that do both out of the box,
and for that, both ansible and salt would work fine.

So as I am the only one that know ansible, I guess pushing for it
wouldn't be wise.

I think salt would be a good middle ground, since that satisfy 3 out of
4 people who expressed in this thread. Louis, do you have a significant
concern on saltstack ?

We can also decide to just make a POC for it, and decide after a while
( like 6 months ) if this is the good solution or not ?

( and for people that want to see ansible, I plan to open the infra of
another project I am working one, manageiq.org, even if the infra is
much less exciting than gluster )

-- 
Michael Scherer
Open Source and Standards, Sysadmin
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 836 bytes
Desc: This is a digitally signed message part
URL: <http://www.gluster.org/pipermail/gluster-infra/attachments/20140901/7b7fdb44/attachment.sig>