-
Notifications
You must be signed in to change notification settings - Fork 196
improve the k-a versioning in the test-infra kubeadm runner #555
Comments
All we need is some way to note the stable version to use for CI. I'm open to anything... I think if we do keep the version specified at the test-config we should remove this checkout from the runner and instead specify it with --repo on the kubeadm jobs. This will update when the PR merges instead of after upgrading the images, and it potentially lets us freeze the version for older kubernetes releases. |
i see a but it points to the main repo. there is this check: if we remove it and always checkout a fresh copy of a stable k-a branch wouldn't this image work when the periodics trigger without updating the images everytime? i must admit that i don't understand how it all works. the idea of the
|
Yes, you can supply multiple repos, the first one will be $PWD for testing, but the others will be checked out along side it as well, and you can specify a hash for those. We can remove the checkout from the runner and instead specify which release to checkout on each job as an extra repo, which also makes it configurable per-job / per-kubernetes-release |
if we hardcode the hash or release (is it a k-a release?) in the job config, wouldn't that mean that if we bump k-a by one patch/fix/commit we still need to update the job config too and possibly make a new k-a release? is it possible to specify a branch (tip) instead? |
We can do that though, and selectively for the tests that actually need it (probably just recent k/k branches)
does k-a have releases?
yep you can just specify a branch without a sha. independently of what way we specify the stable version (sha, branch, ...) we should probably move specifying it to be on the jobs so updates to this are quick (and likely only involve cluster-lifecycle to review & merge), versus needing a new image to be published first. |
ok, given right now we always use the tip of k-a, all tests are conformant with the tip of k-a. ideally the tests by default should use the latest (presumably patched branch tip), but if a test complains we can specify an exact commit.
it has one... and eventually in time deprecating it for e2e.
would that be a valid diff and would it replace the checkout in the runner?
|
Hmm I'll defer on how we should manage this repo, but:
move the repo flag up by the other one (but AFTER it!), somewhere before the bare |
@detiber @chuckha - 1st use case for aws implementation of clusterapi. Need to chat with @countspongebob on this. |
@BenTheElder ack! thanks for the explanation. |
can we please get some super-powers for the decision on this? summary: @errordeveloper agreed to this, but he wanted confirmation from others too. also /cc @roberthbailey as he might have write access and an opinion on this. |
I'm a +1 to just getting this done, but I have now powers here. |
@errordeveloper |
@neolit123 I've create the |
@errordeveloper |
we eventually want to deprecate kubernetes-anywhere in the e2e suite.
but this isn't that easy, so we have to stick to it for a while.
right now we have a fixed k-a (kuberenetes-anywhere) commit in the image for the kubeadm runner.
this imposes the difficulty that we need to PR two changes:
runner
.https://github.com/neolit123/test-infra/blob/8f0bd23ac99e7c8809ccc6aefb42273dc96ca066/images/kubeadm/runner#L29
my proposal:
kubeadm-e2e
in this repokubeadm-e2e
in the runner!!kubeadm-e2e
branch to the k-a master branch if needed.future changes in kubeadm are going to make k-a very hard to maintain for both users and e2e tests.
/assign @timothysc
/assign @BenTheElder
/priority critical-urgent
/kind feature
/cc @errordeveloper @luxas @mikedanese
for super powers on creating a k-a branch called
kubeadm-e2e
if this is approved by @BenTheElder and @timothyscThe text was updated successfully, but these errors were encountered: