Re: [Puppet Users] passenger used in puppet 3.0?

[Matthew Burgess]

On Tue, Nov 27, 2012 at 2:27 PM, Dominic James <domini...@gmail.com> wrote:
> Question: on my current puppet setup 2.7.3-0 rhel 6.2 centos - passenger is
> used to provide the throughput for the heavy delivery that puppet could not
> do by itself, was this fix in version 3.0 puppet? I will be upgrading to 3.0
> and will like puppet master to push to multiple of clients without being
> clobbered, Can I just use puppet transport agent by itself instead of using
> apache/passenger?

You'll have to stick with Apache/Passenger for any non-trivial number
of nodes. Webrick (the HTTP server used by the puppetmaster) is
single-threaded, so can only handle one client checking in at a time.

Regards,

Matt.

[DJames]

ahh i was hoping the puppet master on 3.0 could handle the transport on its own, well dealing with 1800+ plus production systems this mechanism could be an issue...

[DJames]

ran across this, what do you think about Mongrel? http://projects.puppetlabs.com/projects/1/wiki/Using_Mongrel

On Tuesday, November 27, 2012 9:45:28 AM UTC-5, Matthew Burgess wrote:

[Ramin K]

On 11/28/2012 6:11 AM, DJames wrote:
> ran across this, what do you think about Mongrel?
> http://projects.puppetlabs.com/projects/1/wiki/Using_Mongrel

It's a more complex config and system for no benefit. Are you having
any problems with Passenger/Apache? It's the simplest and most common
way to run a Puppet master. I would however recommend going through the
extra work to install Passenger 3.x if your distro does not ship with it.

Ramin

[DJames]

current dev system is using passenger, but production (babystage) is using 3.0 version, and we have 1800+ servers (prod) that will be clients....
is passenger the way to go still?

[Jason Slagle]

On 11/28/2012 09:53 AM, DJames wrote:
> current dev system is using passenger, but production (babystage) is
> using 3.0 version, and we have 1800+ servers (prod) that will be clients....
> is passenger the way to go still?
>

I was previously using passenger and just recently switched to using
nginx/unicorn. I like the setup a lot better - I can't really quantify
why other than the architecture of the way it does it seems better to
me, and getting passenger working was a pain under RHEL6.

General instructions here:

http://projects.puppetlabs.com/projects/1/wiki/using_unicorn

Some notes:

1) I ended up symlinking /etc/puppet to ~puppet/.puppet to fix the issue
where puppet master not running as root will not look in the right place
for it's config.

2) I couldn't find unicorn packages for RHEL - I ended up using gem2rpm
and tweaking it to build.

3) I used supervisord instead of god because I'm more familiar with it.
Again you'll likely need to build packages for it - I just updated the
spec here: https://github.com/easel/supervisor-rpm There's a good
puppet module to manage supervisord here:
https://github.com/plathrop/puppet-module-supervisor

Thanks,

Jason

[DJames]

we have f5's load balancer can it be used instead of Nginx?

[Daniele Sluijters]

That depends.

If you start Unicorn and make it listen to a Unix socket then no, you'll need something in front of it to accept the HTTP connections for you and then proxy that to the unicorn socket.

You can also tell Unicorn to bind against an IP, in which case you won't need an nginx in front and you can just tell f5 where to find your unicorn(s).

You will, in both cases, need something like God, supervisord or something else to check and restart the Puppet Master in case it stops/crashes.

-- 

Daniele Sluijters

[DJames]

The setup can be starts off with a internal VIP created on an internal f5 (behind FW) set up pointing to Nginx then pointing to passenger then ---> puppet master...

again i have a large scales og client/hosts.... this should work.