Bug 1243971

I tried this today and it worked really well for me. On my deploy I mistakenly omitted the class (I used the ram_allocation_ratio example) in the hieradata I passed. I then used the same mechanism and did a stack update. The update completed and the changes were effected in the conf file(s).  Some notes in case useful for anyone else looking to test/verify:

SETUP
=====
All of the changes mentioned above are merged upstream but afaics, none downstream (on openstack-tripleo-heat-templates-0.8.6-45.el7ost.noarch none are applied). I cherry picked all the tht changes in the order above - mild merge conflict for the last one (/#/c/197908/, git mergetool). I then grabbed the tripleo-puppet-change like

sudo su; cd /usr/share/tripleo-puppet-elements/hiera/os-refresh-config/configure.d/
mv 40-hiera-datafiles 40-hiera-datafiles.OLD
curl "https://review.openstack.org/gitweb?p=openstack/tripleo-puppet-elements.git;a=blob_plain;f=elements/hiera/os-refresh-config/configure.d/40-hiera-datafiles;h=9be07a92fc7bc5ef7dded8bff1f1415ec64f2785;hb=5d2d14f69085b3a7dac8474aa00b197e3521e4d0" > 40-hiera-datafiles
exit

Rebuilt the overcloud-full image and reloaded the tuskar roles using the procedure at https://github.com/rdo-management/instack-undercloud/blob/c072ac1e16f3f75dc229c7cffae8acab0a52c1c9/doc/source/advanced_deployment/reload_roles_and_plan.rst

TEST
====
Set some ExtraConfig for Compute and Controller:

openstack management plan set 7cfaa92c-26c3-49a6-85dc-76d1317143bf -P 'Controller-1::ControllerExtraConfig={"neutron::dhcp_agents_per_network": 2}'

openstack management plan set 7cfaa92c-26c3-49a6-85dc-76d1317143bf  -P 'Compute-1::NovaComputeExtraConfig={"nova::scheduler::filter::ram_allocation_ratio": 1.8}'

Note here the second example is wrong, since I don't also pass the required class (we don't define/include nova::scheduler::filer otherwise). This is important because I was successfully able to update this after the create_complete of the deploy using this same arbitrary hiera data passing mechanism.

After CREATE_COMPLETE:

[stack@instack ~]$ node_ips=`nova list | grep Running | awk '{print $12}' | tr "ctlplane=" "\n"`
[stack@instack ~]$ for ip in ${node_ips[@]}; do ssh heat-admin@$ip "hostname; sudo grep -rni 'ram_allocation\|dhcp_agents' /etc/*"; done

on the controllers:
/etc/neutron/neutron.conf:243:dhcp_agents_per_network = 2

on computes:
overcloud-compute-0.localdomain
/etc/nova/nova.conf:1597:#ram_allocation_ratio=1.5
/etc/puppet/hieradata/compute_extraconfig.yaml:2:nova::scheduler::filter::ram_allocation_ratio: 1.8

Note ^^^ the value is passed in the hierdata correctly, but not effected in nova.conf since I didn't include the class. 

Update using the same mechanism:

openstack management plan set 7cfaa92c-26c3-49a6-85dc-76d1317143bf  -P 'Compute-1::NovaComputeExtraConfig={"nova::scheduler::filter::ram_allocation_ratio": 1.8, "compute_classes": ["::nova::scheduler::filter"] }'

openstack overcloud deploy --plan overcloud

after UPDATE_COMPLETE, on the computes:

overcloud-compute-1.localdomain
/etc/nova/nova.conf:1604:ram_allocation_ratio=1.8
/etc/puppet/hieradata/compute_extraconfig.yaml:5:nova::scheduler::filter::ram_allocation_ratio: 1.8

We are tracking this outside of bugzilla with other features, so closing this RFE.