This causes problems to @python/python3.8 project, where is too
heavy re-use of VMs (several hundreds of builds on one VM).
This reverts commit 9988126185.
We re-use the mock profile several times during one package build,
and it doesn't make sense to re-download the repo metadata all the
time.
We have nowadays namespaced directory under /var/cache/mock
per-build ID, so there's no risk that we'll use some very old
caches because of 'max_age_days' or 'max_metadata_age_days' config
options.
Put 'elfutils-libelf', otherwise
devtoolset-3-elfutils-libelf-0.161-1.el7.x86_64 from scl
repositories might be installed by dnf into bootstrap chroot.
This one is fedora-copr-only hack, and as such the ethtool package
isn't supposed to be put as Requires: into copr-builder. To not
C&P between x86_64/ppc64le playbooks, put the ethtool hack into
separate tasks file.
.. so we don't fail if yum-deprecated isn't available (F30 builder
case). If the 'yum.rpm' package that provides /bin/yum-deprecated
script _is_ available => mock would use it anyways (it's the case
on F28 builder).
The dnf-utils package is not installable together with yum-utils
on Fedora 28, so even though it's needed for F30 - it would break
F28 spinup scripts.
To be precise - from F30 we should move to `copr-builder` package
anyways, which will bring all the necessary deps with itself; so
we'll remove the list entirely.
There is a problem with our current playbooks, that they can be
executed automatically without us knowing about it. That is an issue
particularly during release process because we can prepare new
packages into infra-tags repo or bodhi and a nightly reprovision
can upgrade to them outside of an outage window or any of us being
prepared for it.
Therefore `groups/copr-*.yml` playbooks *should not* upgrade any
packages, but only ensure, that those packages are installed. For
upgrade, there should be separate `manual/copr/copr-*-upgrade.yml`
playbooks. Because they are located under `manual` directory, it
is secured, that they can't be run automatically.
See https://pagure.io/copr/copr/issue/636
By default only root can run the `sign` command. This
check is applied within obs-signd code. We need to
allow regular user in the config, see `man sign.conf`.
Also /usr/bin/sign is owned by root:obsrun with
-rwsr-x--- hence we need to add a user to the obsrun group.
This seems to be caused by some asynchronous update of ansible package
on copr backend:
The error appears to have been in
'/home/copr/provision/provision_builder_tasks.yml': line 45,
column 7, but may
be elsewhere in the file depending on the exact syntax problem.
The offending line appears to be:
with_items:
- @copr/copr
^ here
historically proxy01 worked the best for us but now after
https://pagure.io/fedora-infrastructure/issue/7553, the situation
could be better.
alternative would be to use proxy01 and proxy10 which are both
in phx2 that would require something else than /etc/hosts alias.
