On 10/03/10 19:24, Douglas Garstang wrote:
> On Wed, Mar 10, 2010 at 10:18 AM, Brice Figureau
> <brice-pup...@daysofwonder.com> wrote:
>> On 10/03/10 18:58, Douglas Garstang wrote:
>>> We have puppet 0.24.8 running on multiple EIGHT core 3.16Ghz servers
>>> with 32Gb of RAM, and in each case puppet is taking longer and longer
>>> to run, as we have it control more. Currently it's taking up to 20
>>> minutes to perform a run.
>>>
>>> What approaches can I take to significantly reduce the time it takes
>>> puppet to run? It's ALSO sucking up an inordinate amount of CPU while
>>> it performs a run. The server is using passenger.
>>
>> Where do you experience the issue: on the clients or on the master?
>> 0.25 highly improved the master performance and file serving.
>
> The issue is on the clients. The master seems fine. I'd like to avoid
> 0.25 for now, as I simply could not get the SSL keys to work with it
> the last time I tried and I can't risk production seems not being able
> to receive updates for days on end.
>
>>
>> High cpu usage on the client is highly dependent on what you are
>> managing (ie most of the time is usually spent in other processes than
>> puppet, like package manager). Something that also can stress clients is
>> managing deep file hierarchies.
>
> We probably have some deep file hierarchies.
Try to comment those in your manifests, just to see if that is the root
cause.
Also make sure you don't have a default like this:
File { checksum => md5 }
or other checksum in your non-sourced/non-content file{} resource.
Because that means all your local not sourced file management will have
to md5 every managed file. If you combine this with deep hierarchies and
recursion, then you'll have some CPU consumption troubles.
If you have those recursive not sourced file {} resources, make sure to use:
checksum => undef
in them to at least not md5 everything.
I think this problem doesn't exist in 0.25.
>>
>> For high cpu usage on the master, you can try to:
>> * disable storeconfigs or use thin_storeconfigs (0.25)
>> * make sure your clients sleep longer than the default or use splay
>> times so they don't ask their catalogs at the same time
>> * use a different ruby interpreter, and/or passenger
>> * if you're doing tons of file serving, offload those to a static
>> server (see my last blog article in my signature). This will free your
>> masters to serve more catalogs per unit of time.
>
> The main isssue isn't even really the high CPU usage... it's just that
> the client takes 20 minutes to run. That's the really inconvenient
> bit. We aren't using storeconfigs. Putting config in a db is crazy. We
> only have a total of maybe a dozen machines running the client, so I
> doubt increasing the time between runs will make any difference. We
> ARE using passenger on the server (said that in my original post). Not
> doing tons of file serving... the master is not working anywhere near
> as hard as the clients.
You can ignore my above advice since only your clients are affected.
--
Brice Figureau
My Blog: http://www.masterzen.fr/
--
You received this message because you are subscribed to the Google Groups
"Puppet Users" group.
To post to this group, send email to puppet-us...@googlegroups.com.
To unsubscribe from this group, send email to
puppet-users+unsubscr...@googlegroups.com.
For more options, visit this group at
http://groups.google.com/group/puppet-users?hl=en.
