Puppet Master VMs

167 views
Skip to first unread message

Scott Merrill

unread,
Mar 1, 2012, 4:18:56 PM3/1/12
to puppet...@googlegroups.com
Is anyone running their Puppet Master server(s) as virtual guests? If
so, how big are those VMs in terms of memory and virtual CPUs, and how
many Puppet clients are they serving?

Thanks!
Scott

Jeffrey Watts

unread,
Mar 1, 2012, 4:33:02 PM3/1/12
to puppet...@googlegroups.com
I am. I run ours on a 1GB RAM VM with 2 vCPUs. We manage about 500 Linux servers, each checking in every 30 mins. It's pretty much idle, though I'm gradually rolling out our modules so our catalog currently compiles fast. I'm using Webrick, and plan on moving to Passenger/Apache whenever it starts slowing down.   I doubt I'll ever need to go to a physical server. I run Dashboard on another VM. 

Jeffrey
--
You received this message because you are subscribed to the Google Groups "Puppet Users" group.
To post to this group, send email to puppet...@googlegroups.com.
To unsubscribe from this group, send email to puppet-users...@googlegroups.com.
For more options, visit this group at http://groups.google.com/group/puppet-users?hl=en.

Alexander Swen

unread,
Mar 1, 2012, 4:41:08 PM3/1/12
to puppet...@googlegroups.com

Hi scott, we do.
our puppet master runs as a KVM guest with 2 cpu,s 1G mem and 1g swap. it currently serves ~30 puppet agents. because we restart all those agents using cron every night (because of mem leak in ruby 1.8) the mem load is a bit high on this master when all those agents try to fetch their catalog after startup. this might be reason for us to increase mem size.
other details:
We run the master as passenger behind apache and use a separate mysql server to run the storeconfig db. this off course gives the master some relief. We have no complaints about performance. (but with only ~30 modules we don't have a very heavy config for now)
normal catalog run on an avarage server takes 15-20 seconds. only the nagios host exceeds this time to more than a minute sometime because of the enormous amount of tests (we test ~2500 things on nearly 90 hosts).

in the future we might start using Hiera (we suspect this to increase load)
and we will place another master in our other datacenter. this will be a VM as well. that will lower the load for as long as each DC has less servers each than we have in total now.
we will add more and more modules.

kinds regards,
Alex

Peter Bukowinski

unread,
Mar 1, 2012, 4:47:35 PM3/1/12
to puppet...@googlegroups.com
I run puppetmaster (2.7.11) and dashboard (1.2.3) with Passenger/Apache (3.0.11/2.2.3) and mysql db (5.0.77) on a single rhel5.5 VM. I service about 700 clients with hourly check-ins. I used to also run mcollective on this VM but have since moved it. The VM is *slightly* overprovisioned with 4 vCPUs and 20GB of memory.

-- 
Peter M. Bukowinski
Sr. Systems Engineer
Janelia Farm Research Campus, HHMI

Gary Larizza

unread,
Mar 1, 2012, 4:51:41 PM3/1/12
to puppet...@googlegroups.com
On Thu, Mar 1, 2012 at 1:33 PM, Jeffrey Watts <jeffrey...@gmail.com> wrote:
I am. I run ours on a 1GB RAM VM with 2 vCPUs. We manage about 500 Linux servers, each checking in every 30 mins. It's pretty much idle, though I'm gradually rolling out our modules so our catalog currently compiles fast. I'm using Webrick, and plan on moving to Passenger/Apache whenever it starts slowing down.   I doubt I'll ever need to go to a physical server. I run Dashboard on another VM. 

Jeffrey


Hey Jeffrey,

You should really look at moving to Passenger from webrick.  Webrick is great for development purposes, but is NOT ideal for production environments.  The fact that you're running webrick with 500 nodes makes me think that your catalogs might be fairly small.  If it's been working for you, then that's awesome, but I'd definitely look towards Passenger :)


 
On Thursday, March 1, 2012, Scott Merrill wrote:
Is anyone running their Puppet Master server(s) as virtual guests? If
so, how big are those VMs in terms of memory and virtual CPUs, and how
many Puppet clients are they serving?

Thanks!
Scott

--
You received this message because you are subscribed to the Google Groups "Puppet Users" group.
To post to this group, send email to puppet...@googlegroups.com.
To unsubscribe from this group, send email to puppet-users...@googlegroups.com.
For more options, visit this group at http://groups.google.com/group/puppet-users?hl=en.

--
You received this message because you are subscribed to the Google Groups "Puppet Users" group.
To post to this group, send email to puppet...@googlegroups.com.
To unsubscribe from this group, send email to puppet-users...@googlegroups.com.
For more options, visit this group at http://groups.google.com/group/puppet-users?hl=en.



--

Gary Larizza
Professional Services Engineer
Puppet Labs

Markus Falb

unread,
Mar 1, 2012, 5:33:49 PM3/1/12
to puppet...@googlegroups.com
On 1.3.2012 22:41, Alexander Swen wrote:
> On Thursday, March 1, 2012 10:18:56 PM UTC+1, Scott Merrill wrote:

> because we restart all those
> agents using cron every night (because of mem leak in ruby 1.8) the mem
> load is a bit high on this master when all those agents try to fetch
> their catalog after startup. this might be reason for us to increase mem
> size.

$ puppet agent --genconfig
...
# The maximum time to delay before runs. Defaults to being the same as the
# run interval.
# The default value is '$runinterval'.
splaylimit = 1800
...
# Whether to sleep for a pseudo-random (but consistent) amount of time
before
# a run.
# splay = false
...

--
Kind Regards, Markus Falb

signature.asc

Alexander Swen

unread,
Mar 1, 2012, 6:01:43 PM3/1/12
to puppet...@googlegroups.com
Thanks Markus, that's a very usefull tip. will follow!

 

# Whether to sleep for a pseudo-random (but consistent) amount of time
before
# a run.
# splay = false

best regards, Alex

Brian Gallew

unread,
Mar 1, 2012, 8:17:07 PM3/1/12
to puppet...@googlegroups.com
Also, there are some known issues with certain Ruby builds that cause the puppet daemon to randomly hang.  Sometimes it's best to have cron run your puppet jobs.

--
You received this message because you are subscribed to the Google Groups "Puppet Users" group.
To view this discussion on the web visit https://groups.google.com/d/msg/puppet-users/-/nwAfDpOddOUJ.

Jeffrey Watts

unread,
Mar 5, 2012, 10:16:57 AM3/5/12
to puppet...@googlegroups.com
I am. I run ours on a 1GB RAM VM with 2 vCPUs. We manage about 500 Linux servers, each checking in every 30 mins. It's pretty much idle, though I'm gradually rolling out our modules so our catalog currently compiles fast. I'm using Webrick, and plan on moving to Passenger/Apache whenever it starts slowing down. 

Jeffrey

On Thursday, March 1, 2012, Scott Merrill wrote:
>
> --
> You received this message because you are subscribed to the Google Groups "Puppet Users" group.
Reply all
Reply to author
Forward
0 new messages