-
Notifications
You must be signed in to change notification settings - Fork 1.6k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
ARM64 CI #188
Comments
@lubinsz You might help on this? |
note that we will need to fix #166 first, however that is very doable. Dims previously made a quick patch that worked, but we haven't PRed anything yet. |
@BenTheElder @dixudx |
@lubinsz see my previous patch in #166 (comment) |
https://github.com/WorksOnArm/cluster/issues/154 gives us access to Packet hardware.
/cc @devaii |
I think running docker / SSH only is the most well understood path currently, we can treat these similar to a node or cadvisor e2e job, and put credentials in Prow to access them. Long term it might be interesting to be able to run prowjobs on these machines directly, but that will require more work to maintain the cluster and it will take figuring out how we want to handle distributing other credentials. |
When trying to kind build, we note that docker-ce and friends are not available directly from the same repos:
Probably need some debugging:
Version check:
|
It looks like the issue is due to the See: https://github.com/kubernetes-sigs/kind/blob/master/images/base/Dockerfile#L29 I am working on a patch. |
Yeah, there's a bunch of places marked TODO for handling this because I wasn't sure where / how to plumb it through, I think using runtime.GOARCH should be fine, Dims's previous patch is here: #188 (comment) |
@BenTheElder unfortunately the paste with patch expired |
we still need CI, #358 works well! |
Thanks to @ZhengZhenyu and other awesome folks at OpenLab (https://github.com/theopenlab) We now have a functional KinD on ARM CI !!! Please see: |
Now that we have the jobs running successfully for a few days we would like to know how the kind community would like to do the testgrid reporting. I believe there are two options:
I believe the second option is possible but would require setting up a user/auth acct for openlab and of course the other would be for openlab to resolve; using the existing bucket we use for cloud-provider-openstack or setup a new one I could be wrong but we are ready to get the reporting to the proper place so the community can work as expected on any issues surfaced. |
cc @BenTheElder - please see the question from Melvin ^^ |
either works! see also https://github.com/kubernetes/test-infra/tree/master/testgrid/conformance, we can setup a GCS bucket if we don't want to use any existing ones. |
[sorry for the huge delay, this slipped through my inbox :(] |
No problem @BenTheElder totally understand. Will create a new one just to keep things separated since that is possible |
Re-Reading through this.. |
ack @BenTheElder |
Hi @mrhillsman , @dims and @BenTheElder I add an issue in OpenLab side to trace this job theopenlab/openlab#257 |
@aojea Hmm, are you sure? 0.26 seems not ok + kind build node-image --base-image kindest/base:latest --type=bazel --kube-root=/home/zuul/src/k8s.io/kubernetes |
I'm building again for 0.24 |
@aojea Hi, sorry for the delay, I've tried serveral versions, and finally rolled back to 0.23.2 and manually tested(no log will be updated), you should be able to see the results after next periodic run. |
@ZhengZhenyu you did it, now is building the cluster. |
@ZhengZhenyu the e2e tests are running but is failing to upload the results because the script seems to need python > 3.6 but the node has python 3.5
The encoding argument is not present in python 3.5
Is it possible to use python > 3.6? |
@aojea sure, I will try |
we hit another problem , seems the account is not longer valid
|
@dims @ZhengZhenyu do you have an idea on what can be the problem with the service account? ^^ These are the logs https://logs.openlabtesting.org/logs/periodic-6/18/github.com/kubernetes-sigs/kind/master/kind-integration-test-arm64/230250b/ |
@aojea Hi, sorry for the delay, we also had the similar problem in cloud-provider-openstack job, and my colleague checked yesterday, it turns out it is a wrong use of |
@aojea Hi, seems the job successed once in 8.20 and I can see both results from testgrid and openlab: But then the job starts to fail again. |
@aojea Hmm, seems there is something wrong setting up the env again and the tests did not run and thus nothing can be uploaded. |
@ZhengZhenyu the e2e.sh script changed at that time but don't know if one of those changes broke the openlab CI. Seems that the containers for the kubernetes components are not able to spawn, i.e the kubelet fails and there are no logs for the kubeapi-server , .... 97b044c#diff-d9fa0450190d60ba133fb92282a94725 I've sent a PR to try to align the CI job with the new changed on the e2e.sh, and we can iterate from m there. |
@fejta-bot: Closing this issue. In response to this:
Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes/test-infra repository. |
KIND supports arm64 out of the box now, but getting CI for every possible configuration is not super maintainable for us. We rely on upstream Kubernetes working on ARM through whatever ARM supporters want to do, and then kind avoids doing things architecture specific, sticking to portable languages and tools. |
hack: README - add steps for single E2E test
Fix: storageclass parameter
per discussion in #kind slack, we should setup some CI with openlab to get
kind
on arm64 xref #166@dims was able to get arm64 working, but we'll need some set this up to keep it working once that goes in, as the maintainers do not have access to suitable arm machines to test on otherwise.
/assign
/kind feature
/priority important-longterm
The text was updated successfully, but these errors were encountered: