
On Tue, 2020-06-02 at 12:23 +0100, Daniel P. Berrangé wrote:
On Tue, Jun 02, 2020 at 01:10:08PM +0200, Andrea Bolognani wrote:
On Tue, 2020-06-02 at 11:33 +0100, Daniel P. Berrangé wrote:
I don't think we should be building container images that we're not going to be using in any of the jobs, as it can only ever slow down the build overall.
These same containers are also available for use outside of CI, eg. with 'make ci-build', so I think we should keep building them.
That only needs them built on the master branch of the main repo though, not every branch in every fork
Fair enough. So what you're suggesting is something like .container_optional_job_template: &container_optional_job_definition <<: *container_job_definition allow_failure: true except: variables: - $CI_PROJECT_NAMESPACE != libvirt only: - master correct?
As for slowing down builds, that still only applies to the first build after Dockerfiles are updated, so I don't think it ultimately matters very much.
I'd expect a rebiuld if the distro base image changes which could be fairly often for the rawhide like distros.
This advantage might be cancelled out by the fact that only a limited number of shared runners is available, so if for example we have access to 5 runners, whether we run 6 or 10 jobs will make no difference in terms of total pipeline completion time. -- Andrea Bolognani / Red Hat / Virtualization