container-image-prepare is building too many containers

[Thomas Herve]

I'm trying to track down the root cause, but I'm filling this in the mean time.

In a simple containers-multinode CI job, tripleo-container-image-prepare takes more than 20 minutes. Of the 50+ containers built, at least 20 aren't used during the deployment (aodh, ceilometer, panko, cinder, barbican, etc).

It's my understanding that we should use role date to find out which services are enabled, but the detection doesn't seem to work.

[Emilien Macchi (emilienm)]

So the problem is on the Undercloud, we pull all containers from services that are in the default role:
https://github.com/openstack/tripleo-heat-templates/blob/master/roles/Undercloud.yaml#L14-L94

No matter the resource is set to OS::Heat::None or not. They are pulled here:
https://github.com/openstack/tripleo-heat-templates/blob/master/puppet/services/container-image-prepare.j2.yaml#L109

And that's why we pull them all.

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/619177

[Bogdan Dobrelya (bogdando)]

I'm not sure the problem is CI specific, but rather that's something to be fixed in t-h-t DF

[OpenStack Infra (hudson-openstack)]

Change abandoned by Thomas Herve (<email address hidden>) on branch: master
Review: https://review.openstack.org/619177