Closed trown closed 8 years ago
Have been running with this for past few weeks. plan to move it to here as a nugget, or in it's own split out role (more likely) https://review.gerrithub.io/#/c/270709/
WIP here (in this project): https://review.gerrithub.io/#/c/273532/
This is getting close. Gates running now and targeting a merge this week (tomorrow ideally)
https://review.gerrithub.io/#/c/277420/
@weshayutin fyi
This issue has been resolved / merged.
@trown do you need to close (since you opened?)
moved from: https://github.com/redhat-openstack/tripleo-quickstart/issues/64
halcyondude commented 18 days ago
Background: the existing publish role in oooq uses fetch to copy everything from ansible target (virthost in this case) where the images are built --> localhost (or the jenkins slave in the CI use case). They are then copied with a shell block calling rsync, from localhost to publishing destination.
On our jenkins slaves this is problematic as the images are single-digit GB in size, and the default home disk for the slaves is 20G . The existing playbook does clean up localhost after the copy, but already I've had a few failures in POC jobs due to space issues on the slaves.
This issue is to track creating a simple 1-hop (vs. existing 2-hop) variant of the publish role. Goal is for it to be small, simple, no frills. It will be useful for CI + developers using tripleo-quickstart image building to cook their own custom images.
It should use delagate to (from) --> publishing_dest using the synchronize standard module (instead of shell: rsync).
The existing playbook will be left alone. We still do require the 2 hop, rsync protocol only (vs ssh) implementation for RDO CI, as there is not direct routes between image build location --> publishing endpoint.