Tech Preview - OpenShift Origin on Fedora 19 ARM

ARM build scripts and misc for building/installing/running OpenShift Origin on Fedora 19 ARM

Installing Origin on ARM

#####Requirements

• Fedora 19 ARM (armhfp) and compatible device
• Spare time
• Patience (not verified to be ready for prime time, there may be dragons)

Most of these instructions are from the OpenShift Puppet Install Docs with adaptations where needed. The goal is to no eventually no longer require custom install documentation.

Install puppet and bind

yum install -y puppet bind git

Remove firewalld (This should be temporary, a new change recently happened in the Origin code base that is incompatible with firewalld and this doc will be updated once that is resolved).

 yum remove firewalld

Install dependency puppet modules because we're going to install using the Origin puppet master branch so puppet won't auto install the deps for us. Then clone the master repo. (Need ntp and stdlib, but ntp pulls in stdlib for us).

puppet module install puppetlabs/ntp
git clone https://github.com/openshift/puppet-openshift_origin.git /etc/puppet/modules/openshift_origin

Now we need to generate the BIND TSIG Key (we will need to put this in our puppet config).

#Using example.com as the cloud domain
/usr/sbin/dnssec-keygen -a HMAC-MD5 -b 512 -n USER -r /dev/urandom -K /var/named example.com
cat /var/named/Kexample.com.*.key  | awk '{print $8}'

Set the hostname.

echo "broker.example.com" > /etc/hostname
hostname broker.example.com

Now we need to create our puppet config, I call mine configure_origin.pp but the name doesn't really matter. Note to change the YOUR_BIND_TSIG_KEY with the key output from the cat /var/named/Kexample.com.*.key | awk '{print $8}' command we ran above.

class { 'openshift_origin' :
  # Components to install on this host:
  roles          => ['broker','named','activemq','datastore','node'],

  # The FQDNs of the OpenShift component hosts; for a single-host
  # system, make all values identical.
  broker_hostname            => 'broker.example.com',
  node_hostname              => 'broker.example.com',
  named_hostname             => 'broker.example.com',
  datastore_hostname         => 'broker.example.com',
  activemq_hostname          => 'broker.example.com',

  # BIND / named config
  # This is the key for updating the OpenShift BIND server
  bind_key                   => 'YOUR_BIND_TSIG_KEY',
  # The domain under which applications should be created.
  domain                     => 'example.com',
  # Apps would be named <app>-<namespace>.example.com
  # This also creates hostnames for local components under our domain
  register_host_with_named   => true,
  # Forward requests for other domains (to Google by default)
  conf_named_upstream_dns    => ['8.8.8.8'],

  # Auth OpenShift users created with htpasswd tool in /etc/openshift/htpasswd
  broker_auth_plugin         => 'htpasswd',
  # Username and password for initial openshift user
  openshift_user1            => 'openshift',
  openshift_password1        => 'password',

  # To enable installing the Jenkins cartridge:
  install_method             => 'yum',
  repos_base                 => 'https://mirror.openshift.com/pub/origin-server/release/3/fedora-19/',
  jenkins_repo_base          => 'http://pkg.jenkins-ci.org/redhat',

  #Enable development mode for more verbose logs
  development_mode           => true,

  # Set if using an external-facing ethernet device other than eth0
  #conf_node_external_eth_dev => 'eth0',

  #If using with GDM, or have users with UID 500 or greater, put in this list
  #node_unmanaged_users       => ['user1'],
}

At this point we should be good to go to run our puppet deploy. Take note that this takes a long time. On a Calxeda HighBank quad-core with 4GB of RAM it took about 30 minutes. There's a lot going on here.

 puppet apply --verbose configure_origin.pp

Once this is complete we need to run out "fixup" script because at the time of this writing there are a couple issues that occur during the deploy that are still yet to be tracked down to their root cause but the end goal is to work these issues out and not require this step.

git clone https://github.com/maxamillion/openshift-origin-arm.git ~/openshift-origin-arm
~/openshift-origin-arm/fixup_arm_origin.sh

At this point you should be able to reboot your system and have a functional OpenShift Origin PaaS running on ARM. However if you would like, you don't have to reboot but you'd have to know the order in which to start services and it's just easier. (I know, I know ... rebooting linux is terrible, feel free to read the docs and learn the order to start services if you care enough)

From here we can install the rhc gem and setup using our local system (or if you have DNS setup properly you can do this on an external system)

gem install rhc
rhc setup --server=broker.example.com

From here you should consult the rhc documentation if you are not familiar with using OpenShift. There is also a lot of great material here.

For more information about OpenShift Origin install with puppet or in general please reference the following URLs:

• http://openshift.github.io/documentation/
• http://openshift.github.io/documentation/oo_deployment_guide_puppet.html

If you run into any issues or have found a resolution to any ARM specific issues, feel free to file an issue ticket on this repo or ping me on #openshift-dev on irc.freenode.net, nick: maxamillion

Known Issues

1. postgresql cartridge does not work at this time on ARM (due to pgdata arch specific optimizations, this pull request should resolve the issue) (fixup script handles this in the mean time)
2. /usr/lib64 gets created during deployment process but not sure why (fixup script handles fixing this)

Building Origin ARM packages

There was an old method that included a script from this repo but it's far more clean of a method to use mockchain.

Steps to build:

Step 1 - Prep the SRPMS

Download all SRPMS and place them somewhere (doesn't matter where, we'll tell mockchain where they are.

Optionally you can build SRPMS from source using the following (note, this requires you have the tito utility installed which you can install with yum -y install tito)

git clone https://github.com/openshift/origin-server.git
cd origin-server
for pkg in $(find . -name \*.spec); do
  pushd $(dirname $pkg)
    tito build --test --srpm # use --test if you want nightlies, omit for latest stable tag
  popd
done

Now you will find a bunch of SRPMS in /tmp/tito and you can either mockchain from there or move them elsewhere. Your choice.

Step 2 - Run the mockchain

The following command will log to the build directory output by the chainbuild command and result in the filename mockchain.log, we're also adding the Origin dependencies repository to the mock config for the chainbuild run. Also, pass --recurse so that mockchain will handle circular dependencies for us, which does exist because of the console.

mockchain -r fedora-19-armhfp \
    --log mockchain.log \
    -a https://mirror.openshift.com/pub/origin-server/release/3/fedora-19/dependencies/armhfp/ \
    --recurse /tmp/tito/*.src.rpm # Or where ever your SRPMS actually are.

From here mockchain will output it's progress as well as inform you of where you can find the resulting packages.