Open goern opened 3 years ago
an interim goal could be to deploy this on a temp OCP/ppc64 deployment, but the target needs to be to deploy this to Op1st.
@durandom
power 9 builds need power9 systerm, it means we deploy a pipeline on a power9 system.
Right, there is an activity on Op1st to set up OCP4.6/ppc64 and Marvin and Sebastian have access to P9 too, so we can test.
Kubeflow consists of several components - so let's start with a first (simple) one and then add one after the other. I'd go for kfctl first, as it is simple and we already committed upstream (https://github.com/kubeflow/kfctl/pull/459).
Essentially, building it is super easy; roughly like this:
git clone --branch v1.2.0 https://github.com/kubeflow/kfctl.git
make -f kfctl/Makefile
Hence, I think this is a good first example to be tested end-to-end.
power 9 builds need power9 systerm, it means we deploy a pipeline on a power9 system.
I think this depends on what you actually want to compile. Golang is pretty nice in that way as it allows cross-compilation out-of-the-box for several architectures (including ppc64le). The example Sebastain mentioned can be run on x86 and will still produce an ppc64le artifact if GOARCH=ppc64le
is set.
Kubeflow consists of several components - so let's start with a first (simple) one and then add one after the other. I'd go for kfctl first, as it is simple and we already committed upstream (kubeflow/kfctl#459).
Essentially, building it is super easy; roughly like this:
git clone --branch v1.2.0 https://github.com/kubeflow/kfctl.git make -f kfctl/Makefile
Hence, I think this is a good first example to be tested end-to-end.
This is a good example of which was an easy port for ppc64le, but the PR is not backported. Therefore the branch v1.2.0 will just produce Linux/Darwin/Windows binaries for x86 & Linux for ARM. The latest commit at the master branch #486 has the flag GOARCH=ppc64le
enabled.
For C & Python it can vary from just recompiling the source code on a Power System up to changing the source code if there is x86 specific parts in it (e.g. MKL exists only for x86)
--branch master
) and release builds (--branch v1.3.0
once the release comes out)hey all, do we have some movement on this?
We'd be ready once the OpenShift cluster on Power is stable and we got access to it. To our knowledge, this has not happened yet. What's the status there? Anything we can do before?
(sadly) very true, see https://operatefirst.slack.com/archives/C01TSGYT0R4/p1621515928005000
Issues go stale after 90d of inactivity.
Mark the issue as fresh with /remove-lifecycle stale
.
Stale issues rot after an additional 30d of inactivity and eventually close.
If this issue is safe to close now please do so with /close
.
/lifecycle stale
Stale issues rot after 30d of inactivity.
Mark the issue as fresh with /remove-lifecycle rotten
.
Rotten issues close after an additional 30d of inactivity.
If this issue is safe to close now please do so with /close
.
/lifecycle rotten
/lifecycle frozen
waiting on power9 systems
Is your feature request related to a problem? Please describe. As an Operator/User of OpenShift on Power9 I want to run a deployment of AICoE-CI so that I can build ppc64 artifacts/container images/golang binaries
Describe alternatives you've considered manual building on OCP/ppc64
Additional context This will mainly be used for Kubeflow on Power9 work by IBM @lehrig @mgiessing could you provide a little bit of info/script on how you build artifacts?
/kind feature /priority important-soon