Catalog compile times > 40s - Puppet 2.7.18

[treydock]

I am finding that the puppetlabs-apache module is somehow adding 30-60 seconds onto a host's catalog compile time when the puppetmaster has no other hosts contacting or generating catalogs.  The Puppetmaster is setup to use Puppet-2.7.18 - Apache & Passenger.  RIght now only 2 hosts are even configured to use this new PM, the PM itself and a Foreman host.  With neither hosts contacting the PM, and disabling the "include apache" and "require apache::mod::dev" in puppetlabs-passenger, I get "Config retrieval: 2.45".

Catalogs had hostnames replaced, puppetmaster.tld is the CNAME for Puppet master and puppetmaster-host.tld is the actual hostname.

Catalog: http://pastebin.com/V29JmB8J

# puppet agent --{summarize,test,debug,evaltrace,noop} | perl -pe 's/^/localtime().": "/e'

<snip>

Sun Dec  9 13:05:35 2012: notice: Finished catalog run in 2.43 seconds

Sun Dec  9 13:05:35 2012: Changes:

Sun Dec  9 13:05:35 2012: Events:

Sun Dec  9 13:05:35 2012:             Total: 1

Sun Dec  9 13:05:35 2012:              Noop: 1

Sun Dec  9 13:05:35 2012: Resources:

Sun Dec  9 13:05:35 2012:       Out of sync: 1

Sun Dec  9 13:05:35 2012:             Total: 22

Sun Dec  9 13:05:35 2012:           Skipped: 6

Sun Dec  9 13:05:35 2012: Time:

Sun Dec  9 13:05:35 2012:         Resources: 0.00

Sun Dec  9 13:05:35 2012:        Filebucket: 0.00

Sun Dec  9 13:05:35 2012:           Package: 0.00

Sun Dec  9 13:05:35 2012:              File: 0.00

Sun Dec  9 13:05:35 2012:          Firewall: 0.01

Sun Dec  9 13:05:35 2012:              Exec: 0.14

Sun Dec  9 13:05:35 2012:          Last run: 1355079935

Sun Dec  9 13:05:35 2012:    Config retrieval: 2.45

Sun Dec  9 13:05:35 2012:             Total: 2.61

Sun Dec  9 13:05:35 2012: Version:

Sun Dec  9 13:05:35 2012:            Config: 1355077701

Sun Dec  9 13:05:35 2012:            Puppet: 2.7.18

Uncommenting "include apache" and "require apache::mod::dev" in puppetlabs-passenger about 40 seconds are added to the compile time.  I went through disabled all modules and parameters, and found that enabling the passenger module alone with Apache included caused the huge delay.

Catalog: http://pastebin.com/wwcKgX5b

# puppet agent --{summarize,test,debug,evaltrace,noop} | perl -pe 's/^/localtime().": "/e'

<snip>

Sun Dec  9 13:13:08 2012: notice: Finished catalog run in 2.23 seconds

Sun Dec  9 13:13:08 2012: Changes:

Sun Dec  9 13:13:08 2012: Events:

Sun Dec  9 13:13:08 2012:             Total: 4

Sun Dec  9 13:13:08 2012:              Noop: 4

Sun Dec  9 13:13:08 2012: Resources:

Sun Dec  9 13:13:08 2012:       Out of sync: 4

Sun Dec  9 13:13:08 2012:           Skipped: 6

Sun Dec  9 13:13:08 2012:             Total: 80

Sun Dec  9 13:13:08 2012: Time:

Sun Dec  9 13:13:08 2012:        Filebucket: 0.00

Sun Dec  9 13:13:08 2012:         Resources: 0.00

Sun Dec  9 13:13:08 2012:           Package: 0.00

Sun Dec  9 13:13:08 2012:          Firewall: 0.01

Sun Dec  9 13:13:08 2012:              File: 0.01

Sun Dec  9 13:13:08 2012:             A2mod: 0.03

Sun Dec  9 13:13:08 2012:           Service: 0.07

Sun Dec  9 13:13:08 2012:              Exec: 0.14

Sun Dec  9 13:13:08 2012:          Last run: 1355080388

Sun Dec  9 13:13:08 2012:    Config retrieval: 42.79

Sun Dec  9 13:13:08 2012:             Total: 43.06

Sun Dec  9 13:13:08 2012: Version:

Sun Dec  9 13:13:08 2012:            Config: 1355080261

Sun Dec  9 13:13:08 2012:            Puppet: 2.7.18

The install is via RPM.  I'm using the Fedora SRPMs rebuilt for CentOS 6.  The Puppet master has two environments, development and production with this as the puppet.conf

# cat /etc/puppet/puppet.conf

---

[main]

server                    = puppetmaster.tld

logdir                    = /var/log/puppet

vardir                    = /var/lib/puppet

ssldir                    = /var/lib/puppet/ssl

rundir                    = /var/run/puppet

factpath                  = $vardir/lib/facter

pluginsync                = true

templatedir               = $confdir/templates

[agent]

report                          = true

environment                     = production

preferred_serialization_format  = yaml

[master]

modulepath                = /etc/puppet/environments/$environment/modules:/etc/puppet/modules

manifest                  = /etc/puppet/manifests/site.pp

environment               = production

autosign                  = false

ssl_client_header         = SSL_CLIENT_S_DN

ssl_client_verify_header  = SSL_CLIENT_VERIFY

certname                  = puppetmaster.tld

dns_alt_names             = puppetmaster.tld,puppetmaster-host.tld

report                    = true

reports                   = store

#reports                   = store, foreman

[production]

manifest                  = /etc/puppet/manifests/site.pp

Has anyone run into this issue with puppetlabs-apache or know why my compile times are so long?  I have a Puppet 2.6.17 server (Also Apache+Passenger) that has one host, for example, with 524 resources , and only a 21.36 second catalog compile time.

Thanks

- Trey

[treydock]

Continuing to debug this issue it seems the apache::mod::default is the bulk of the compile time.  I'm curious if this is a symptom of the module itself or something with 2.7.x.  I moved this puppet master to the EPEL 2.6.17 RPMs and now catalogs that previously took 40-60 seconds are now down to 1-2 seconds.

I noticed that during the long compile times this VM's (2 vCPU , 2GB RAM) CPU usage is at 100% from a single Ruby process that links to the PM Rack process.  My other PM running same exact setup with about 50 managed hosts never has had performance issues always running 2.6.17.

I'd like to move to 2.7.x for functionality of more modern modules but this PM is for an HPC and I'm already struggling to convince my predecessors that Puppetizing the HPC is not going impact performance.  Is there possibly a specific version of 2.7.x or maybe even 3.0.x that may address the performance issues I'm seeing ?

Thanks

- Trey

[Michael Stahnke]

Puppet 3 has lots of performance improvements. In many cases it's
faster than 2.6.x was. Some of those fixes were back-ported to 2.7.20,
but not all of them could be.

For best performance, try out something in the Puppet 3 series.



On Wed, Dec 19, 2012 at 8:30 AM, shurik <a.gr...@gmail.com> wrote:
> Hi
>
> We are tried to update puppet master from 2.6.6 to 2.7.18 and saw similar
> picture. Catalog
> compile time took ~60-120 seconds. Before this update compilation took about
> ~1-5 second.
> After upgrade average CPU usage increased from 10-20% up to 20-40%, during
> some activity
> (compilation?) ruby process puppetmasterd take 100% too.
>
> All, is there are some magic to solve or this is expected behaviour?
>
> Thanks!
>
> ---
> Alexander

> --
> 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/-/cWlBauzx5hwJ.
>
> 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.

[Trey Dockendorf]

In addition to upgrading to Puppet 3, would any other changes help improve performance?  I've seen mention of switching to ruby enterprise edition for example.  We will likely have to disable Puppet automatic runs entirely on compute nodes and have it only run as a administrative queue job or have it run as part of every job's pre or post execution, so minimal catalog compile and run times are critical.

Thanks
- Trey