Bug 949072

Description of problem:

We are currently shipping rubygem-openshift-origin-node-1.0.13-1.el6op
and openshift-origin-msg-node-mcollective-1.0.3-1.el6op.  These versions are incompatible for the following reasons:

• node-1.0.13 has an up-to-date version of /etc/openshift/resource_limits.template that no longer has the max_apps and max_active_apps settings (they are replaced by max_gears and max_active_gears).

• On a typical OSE installation, /etc/openshift/resource_limits.conf is usually a symlink to, or a copy of, /etc/openshift/resource_limits.template.

• mcollective-1.0.3 has an old version of /usr/lib/ruby/site_ruby/1.8/facter/openshift_facts.rb that still reads max_apps and max_active_apps from /etc/openshift/resource_limits.conf.

• The old openshift_facts.rb uses max_apps and max_active_apps to compute active_capacity, which is computed by dividing a quantity by max_active_apps, but if max_active_apps is not set in resource_limits.conf, then it defaults to 0, and thus active_capacity is set to NaN (resulting from division by zero).

• Because the node's facts say that active_capacity is NaN, no gears can be created on the node.


Version-Release number of selected component (if applicable):

The latest released bits.  The system is yum-updated, and the following channels are enabled:

jb-eap-6-for-rhel-6-server-rpms
jb-ews-1-for-rhel-6-server-rpms
rhel-6-server-cf-tools-1-rpms
rhel-6-server-rhev-agent-rpms
rhel-6-server-rpms
rhel-ha-for-rhel-6-server-rpms
rhel-lb-for-rhel-6-server-rpms
rhel-rs-for-rhel-6-server-rpms
rhel-server-ose-infra-6-rpms
rhel-server-ose-jbosseap-6-rpms
rhel-server-ose-node-6-rpms
rhel-server-ose-rhc-6-rpms

How reproducible:

• This bug will break all new installations of OSE.

• The bug may also break existing installations on yum update if those existing installations symlink /etc/openshift/resource_limits.conf to resource_limits.template.

• The bug will not break existing installations on yum update if those existing installations use a copy of resource_limits.conf (yum will not touch resource_limits.conf itself).


Steps to Reproduce:
1. Install a new OSE installation with one node.
2. Try `rhc app create testapp php` with your new OSE installation.
3. Try `mco inventory node.example.com | grep capacity`.


Actual results:

After step 2, you will see "Creating application 'testapp2' ... Unable to create gear on node"

After step 3, you will see that active_capacity and capacity are both NaN.


Expected results:

After step 2, you the application should be created successfully.

After step 3, you should see that active_capacity and capacity are numbers.


Additional info:

The following commit adds the new _gear settings to resource_limits.template and updates openshift_facts.rb so that it uses both the old _app and the new _gear settings:

https://github.com/openshift/enterprise-server/commit/556a9efaf786f5d225de4f2c23432d2b28f724f8

The following commit deletes the old settings from resource_limits.template and updates openshift_facts.rb to use only the new _gear settings:

https://github.com/openshift/enterprise-server/commit/fa38702b762989c09af1c3a5f1324b13a8c0918f

We ship openshift-origin-msg-node-mcollective-1.0.3-1.el6op, which still has an old openshift_facts.rb that uses the old _app settings.  We need to ship openshift-origin-msg-node-mcollective-1.0.4 or newer to get the updated openshift_facts.rb.