[libvirt PATCH v2] ci: Shuffle builds between Debian 10 and Debian sid
Debian sid is currently broken on mips64el and s390x, so use Debian 10 for those specific builds and move other architectures (armv7l and ppc64le) from Debian 10 to Debian sid to maintain the rough 1:1 split. Signed-off-by: Andrea Bolognani <abologna@redhat.com> --- Test pipeline: https://gitlab.com/abologna/libvirt/-/pipelines/355069168 Still running, but all Debian builds have already completed successfully. Changes from [v1]: * also swap s390x and ppc64le. [v1] https://listman.redhat.com/archives/libvir-list/2021-August/msg00515.html .gitlab-ci.yml | 40 ++++++++++++++++++++-------------------- 1 file changed, 20 insertions(+), 20 deletions(-) diff --git a/.gitlab-ci.yml b/.gitlab-ci.yml index d1609c260d..a7197b557c 100644 --- a/.gitlab-ci.yml +++ b/.gitlab-ci.yml @@ -225,7 +225,7 @@ armv6l-debian-10-container: NAME: debian-10-cross-armv6l armv7l-debian-10-container: - extends: .container_job + extends: .container_optional_job variables: NAME: debian-10-cross-armv7l @@ -240,7 +240,7 @@ mips-debian-10-container: NAME: debian-10-cross-mips mips64el-debian-10-container: - extends: .container_optional_job + extends: .container_job variables: NAME: debian-10-cross-mips64el @@ -250,12 +250,12 @@ mipsel-debian-10-container: NAME: debian-10-cross-mipsel ppc64le-debian-10-container: - extends: .container_job + extends: .container_optional_job variables: NAME: debian-10-cross-ppc64le s390x-debian-10-container: - extends: .container_optional_job + extends: .container_job variables: NAME: debian-10-cross-s390x @@ -270,7 +270,7 @@ armv6l-debian-sid-container: NAME: debian-sid-cross-armv6l armv7l-debian-sid-container: - extends: .container_optional_job + extends: .container_job variables: NAME: debian-sid-cross-armv7l @@ -280,7 +280,7 @@ i686-debian-sid-container: NAME: debian-sid-cross-i686 mips64el-debian-sid-container: - extends: .container_job + extends: .container_optional_job variables: NAME: debian-sid-cross-mips64el @@ -290,12 +290,12 @@ mipsel-debian-sid-container: NAME: debian-sid-cross-mipsel ppc64le-debian-sid-container: - extends: .container_optional_job + extends: .container_job variables: NAME: debian-sid-cross-ppc64le s390x-debian-sid-container: - extends: .container_job + extends: .container_optional_job variables: NAME: debian-sid-cross-s390x @@ -473,12 +473,12 @@ armv6l-debian-10: NAME: debian-10 CROSS: armv6l -armv7l-debian-10: +armv7l-debian-sid: extends: .cross_build_job needs: - - armv7l-debian-10-container + - armv7l-debian-sid-container variables: - NAME: debian-10 + NAME: debian-sid CROSS: armv7l i686-debian-sid: @@ -497,12 +497,12 @@ mips-debian-10: NAME: debian-10 CROSS: mips -mips64el-debian-sid: +mips64el-debian-10: extends: .cross_build_job needs: - - mips64el-debian-sid-container + - mips64el-debian-10-container variables: - NAME: debian-sid + NAME: debian-10 CROSS: mips64el mipsel-debian-10: @@ -513,20 +513,20 @@ mipsel-debian-10: NAME: debian-10 CROSS: mipsel -ppc64le-debian-10: +ppc64le-debian-sid: extends: .cross_build_job needs: - - ppc64le-debian-10-container + - ppc64le-debian-sid-container variables: - NAME: debian-10 + NAME: debian-sid CROSS: ppc64le -s390x-debian-sid: +s390x-debian-10: extends: .cross_build_job needs: - - s390x-debian-sid-container + - s390x-debian-10-container variables: - NAME: debian-sid + NAME: debian-10 CROSS: s390x mingw32-fedora-rawhide: -- 2.31.1
On a Wednesday in 2021, Andrea Bolognani wrote:
Debian sid is currently broken on mips64el and s390x, so use Debian 10 for those specific builds and move other architectures (armv7l and ppc64le) from Debian 10 to Debian sid to maintain the rough 1:1 split.
This still marks sid-based containers as mandatory, so any improvement is only temporary Jano
Signed-off-by: Andrea Bolognani <abologna@redhat.com> --- Test pipeline: https://gitlab.com/abologna/libvirt/-/pipelines/355069168
Still running, but all Debian builds have already completed successfully.
Changes from [v1]:
* also swap s390x and ppc64le.
[v1] https://listman.redhat.com/archives/libvir-list/2021-August/msg00515.html
.gitlab-ci.yml | 40 ++++++++++++++++++++-------------------- 1 file changed, 20 insertions(+), 20 deletions(-)
On Wed, Aug 18, 2021 at 01:42:28PM +0200, Ján Tomko wrote:
On a Wednesday in 2021, Andrea Bolognani wrote:
Debian sid is currently broken on mips64el and s390x, so use Debian 10 for those specific builds and move other architectures (armv7l and ppc64le) from Debian 10 to Debian sid to maintain the rough 1:1 split.
This still marks sid-based containers as mandatory, so any improvement is only temporary
sid, just like Rawhide and Tumbleweed, will break from time to time. This is something that's simply inevitable because they're such fast moving targets. The upside is that we also get to catch actual failures ahead of time. If the price to pay is that we have to shuffle a couple of jobs around once every few months (last time was in February, and the previous one was last June) I'd say the benefit greatly outweighs the cost. -- Andrea Bolognani / Red Hat / Virtualization
On a Wednesday in 2021, Andrea Bolognani wrote:
On Wed, Aug 18, 2021 at 01:42:28PM +0200, Ján Tomko wrote:
On a Wednesday in 2021, Andrea Bolognani wrote:
Debian sid is currently broken on mips64el and s390x, so use Debian 10 for those specific builds and move other architectures (armv7l and ppc64le) from Debian 10 to Debian sid to maintain the rough 1:1 split.
This still marks sid-based containers as mandatory, so any improvement is only temporary
sid, just like Rawhide and Tumbleweed, will break from time to time. This is something that's simply inevitable because they're such fast moving targets.
Yes, and we accept that. As of commit 42f3ec750027ee55c8f98a487cc0f183dd9b429d rawhide container builds are optional. It might not be a good first impression for people who freshly cloned the repo, but neither is them not getting FreeBSD coverage.
The upside is that we also get to catch actual failures ahead of time. If the price to pay is that we have to shuffle a couple of jobs around once every few months (last time was in February, and the previous one was last June) I'd say the benefit greatly outweighs the cost.
The cost is also decreased signal-to-noise ratio, which reduces the contributors' faith in CI and makes it more likely for actual bugs to slip in. Also, to me, not all jobs are equally important. Knowing whether the build passed for mingw and there are no unused variables left for clang to find is more important than yet another cross build for Linux. But if you're willing to reroll the dice in a few months time and make yet another version of this patch: Reviewed-by: Ján Tomko <jtomko@redhat.com> </thread> Jano
participants (2)
-
Andrea Bolognani -
Ján Tomko