Add support for running GCI on the GCE cloud provider #25425
Conversation
k8s-github-robot
added
size/XXL
roberthbailey
changed the title
roberthbailey
added
release-note
|
Update the PR by running "hack/verify-flags-underscore.py -e > hack/verify-flags/exceptions.txt". But I have no idea why there are some flags not related with this PR updated by the script. |
| @@ -1116,11 +1118,10 @@ function kube::release::gcs::copy_release_artifacts() { | |||
|
|
|||
| # Having the configure-vm.sh script and trusty code from the GCE cluster | |||
There was a problem hiding this comment.
should this comment now say "configure-vm.sh script and gci code from ..."
There was a problem hiding this comment.
Thanks for catching it. Will correct this.
|
@zmerlynn can you take a look as well since these scripts are closely related to configure-vm.sh? |
|
@roberthbailey An updated version is uploaded, which addressed your comments. But we still have a couple of comments waiting for more information from @zmerlynn and @mikedanese . I am reruning e2e for at least 5 times against this new version |
|
@zmerlynn @roberthbailey I just updated the PR by adding the /etc/motd support (see functions set-broken-motd and reset-motd) and licensing work. Since /usr/local/ is readonly in GCI, I push LICENSE file in /home/kubernetes, and display the info by /etc/motd. I tested the new version using e2e tests. |
k8s-github-robot
added
the
needs-rebase
| # Finds the master PD device; returns it in MASTER_PD_DEVICE | ||
| function find-master-pd { | ||
| MASTER_PD_DEVICE="" | ||
| # TODO(zmerlynn): GKE is still lagging in master-pd creation. |
There was a problem hiding this comment.
Since you forced into rebase anyways, can you remove this TODO from both sides? I've been meaning to remove it from one, but I'd rather not see it copied. :) (It's been done for a while.)
|
LGTM. I left a nit but you can fix it or not. |
|
@zmerlynn @roberthbailey The PR is rebased. I am running e2e on GCI again to make sure no new problem introduced. |
k8s-github-robot
removed
the
needs-rebase
| cp "${KUBE_HOME}/kube-manifests/kubernetes/kube-registry-proxy.yaml" /etc/kubernetes/manifests | ||
| fi | ||
| fi | ||
| reset-motd |
There was a problem hiding this comment.
This should be the last thing that we do. Why do you configure fluentd after putting up the message?
There was a problem hiding this comment.
Probably a paste error. Will update it.
|
Let me also backport #25504 into this PR. In anyway I will need to merge it from trusty code to gci. |
|
Please check the latest version. I already ran e2e on GCI to verify it. |
k8s-github-robot
added
the
needs-rebase
|
Rebase done |
k8s-github-robot
removed
the
needs-rebase
roberthbailey
added
lgtm
|
@lavalamp -- Adding the e2e-not-required label since the only part of this PR that would be tested by the submit queue in the 1 line change to |
|
GCE e2e build/test passed for commit a737e1e. |
|
GKE cluster start up broke... this is the only thing that changed the start up path or that looks remotely suspicious. Reverting, sorry. |
|
Looks like there is no plan to cherry pick this to I'd like to send a PR directly to the |
Not that I know of. I think @andyzheng0831's plan was to keep using trusty, since it exists on that release branch. |
The supports and configs for GCI added here will be used in 1.3 release and upwards. For 1.2.X releases, we will still use cluster/gce/trusty code. Therefore, we will keep maintaining the trusty code as long as 1.2-release branch is not deprecated. Unlike the trusty code, the GCI cluster initialization is through systemd.
This PR is fully tested using e2e tests. I wrote a script to repeatedly spin up a cluster, run CI and slow tests, and then tear down for 15 times for both pure GCI cluster and hybrid cluster (master on GCI, nodes on containervm). The entire tests lasted for more than 3 days, and no failure was shown. I also verified creating trusty cluster, as this PR revises some places affecting it.
@roberthbailey @dchen1107 please review it.
cc/ @fabioy @wonderfly @kubernetes/goog-image FYI.