Size Them

Out of the box, OpenVox Server uses 2 GB of memory and at most four workers, however big the machine is. Each worker, a JRuby instance, compiles one catalog at a time, so a large compiler with the defaults leaves most of its CPUs and memory idle. openvox_tune measures each server and recommends a size for it.

What you’ll learn:

• Running openvox_tune’s plan against your servers
• Putting its recommendations in each server’s node data
• Applying them with Puppet

Prerequisites

Before you start, you’ll need:

• Both servers managed by openvox_server, as in Hand Them to Puppet
• openvox_tune in your OpenBolt project: bolt module add miharp-openvox_tune

Get a recommendation

Run the plan against both servers together, from your workstation:

bolt plan run openvox_tune::tune --targets puppet.example.com,compiler01.example.com hiera=openvox_server

hiera=openvox_server prints the recommendations as Hiera data for openvox_server. For the machines in the recording, a main server with 4 CPUs and 8 GB and a compiler with 8 CPUs and 16 GB:

# For puppet.example.com:
openvox_server::max_active_instances: 2
openvox_server::heap_mb: 1536
openvox_server::reserved_code_cache_mb: 512
# For compiler01.example.com:
openvox_server::max_active_instances: 7
openvox_server::heap_mb: 4096
openvox_server::reserved_code_cache_mb: 1024

The compiler goes from four workers and 2 GB to seven workers and 4 GB. The plan leaves a quarter of each machine’s memory to the operating system and gives each worker 512 MB.

The main server goes down, from three workers to two and from 2 GB to 1.5 GB. (Out of the box, OpenVox Server runs one worker fewer than the machine has CPUs, up to four.) Run against both servers, the plan knows the main server has a compiler beside it, and the compiler does the compiling.

Put the values in node data

Each server’s values go in its node data, next to what is already there:

# data/nodes/compiler01.example.com.yaml
openvox_server::ca: false
openvox_server::ca_server: 'puppet.example.com'

# OpenVox Server sized for this machine by openvox_tune's plan.
openvox_server::max_active_instances: 7
openvox_server::heap_mb: 4096
openvox_server::reserved_code_cache_mb: 1024

# data/nodes/puppet.example.com.yaml
# OpenVox Server sized for this machine by openvox_tune's plan.
openvox_server::max_active_instances: 2
openvox_server::heap_mb: 1536
openvox_server::reserved_code_cache_mb: 512

Apply them

Commit, push, deploy with r10k, and run Puppet on each server:

ssh root@compiler01 'puppet agent -t'
ssh root@puppet 'puppet agent -t'

Each run changes the memory settings in JAVA_ARGS, writes the number of workers to conf.d/openvox_tune.conf, and restarts OpenVox Server before the run ends, so the server already runs at its new size when the run finishes:

$ ssh root@compiler01 'puppet agent -t'
...
-JAVA_ARGS="-Xms2g -Xmx2g -Djruby.logger.class=com.puppetlabs.jruby_utils.jruby.Slf4jLogger"
+JAVA_ARGS="-Xms4096m -Xmx4096m -Djruby.logger.class=com.puppetlabs.jruby_utils.jruby.Slf4jLogger -XX:ReservedCodeCacheSize=1024m"
...
Notice: /Stage[main]/Openvox_tune/File[/etc/puppetlabs/puppetserver/conf.d/openvox_tune.conf]/ensure: defined content as '{sha256}c84d111504...'
Notice: /Stage[main]/Openvox_server::Service/Service[puppetserver]: Triggered 'refresh' from 1 event

Watch it

Next, Same Code on Every Server gets your code onto the compiler.