If you are running kops as part of an e2e test, the following tips may be useful.
Set the KubernetesVersion to a http://
or https://
base url, such as https://storage.googleapis.com/kubernetes-release-dev/ci/v1.4.0-alpha.2.677+ea69570f61af8e/
We expect the base url to have bin/linux/amd64
directory containing:
- kubelet
- kubelet.sha1
- kubectl
- kubectl.sha1
- kube-apiserver.docker_tag
- kube-apiserver.tar
- kube-apiserver.tar.sha1
- kube-controller-manager.docker_tag
- kube-controller-manager.tar
- kube-controller-manager.tar.sha1
- kube-proxy.docker_tag
- kube-proxy.tar
- kube-proxy.tar.sha1
- kube-scheduler.docker_tag
- kube-scheduler.tar
- kube-scheduler.tar.sha1
Do this with kops edit cluster <clustername>
. The spec should look like
...
spec:
kubernetesVersion: "https://storage.googleapis.com/kubernetes-release-dev/ci/v1.4.0-alpha.2.677+ea69570f61af8e/"
cloudProvider: aws
etcdClusters:
- etcdMembers:
- name: us-east-1c
zone: us-east-1c
name: main
...
The simple way:
# cd wherever you tend to put git repos
git clone https://github.com/kubernetes/test-infra.git
export KOPS_E2E_STATE_STORE=s3://your-kops-state-store # Change to your state store path
export KOPS_E2E_CLUSTER_NAME=e2e.cluster.name # Change to an FQDN for your e2e cluster name
test-infra/jobs/ci-kubernetes-e2e-kops-aws.sh |& tee /tmp/testlog
This:
- Brings up a cluster using the latest
kops
build frommaster
(see below for how to use your current build) - Runs the default series of tests (which the Kubernetes team is also running here) (see below for how to override the test list)
- Tears down the cluster
- Pipes all output to
/tmp/testlog
(Note: By default this script assumes that your AWS credentials are in
~/.aws/credentials
, and the SSH keypair you want to use is
~/.ssh/kube_aws_rsa
. You can override JENKINS_AWS_CREDENTIALS_FILE
,
JENKINS_AWS_SSH_PRIVATE_KEY_FILE
and JENKINS_AWS_SSH_PUBLIC_KEY_FILE
if you
want to change this.)
This isn't yet terribly useful, though - it just shows how to replicate the
existing job, but not with your custom code. To test a custom kops
build, you
can do the following:
To use S3:
# cd to your kops repo
export S3_BUCKET_NAME=<yourbucketname>
make kops-install dev-upload UPLOAD_DEST=s3://${S3_BUCKET_NAME}
KOPS_VERSION=`bazel run //cmd/kops version -- --short`
export KOPS_BASE_URL=https://${S3_BUCKET_NAME}.s3.amazonaws.com/kops/${KOPS_VERSION}/
To use GCS:
export GCS_BUCKET_NAME=kops-dev-${USER}
make kops-install dev-upload UPLOAD_DEST=gs://${GCS_BUCKET_NAME}
KOPS_VERSION=`bazel run //cmd/kops version -- --short`
export KOPS_BASE_URL=https://${GCS_BUCKET_NAME}.storage.googleapis.com/kops/${KOPS_VERSION}/
You can create a cluster using kops create cluster <clustername> --zones us-east-1b
Then follow the test directions above.
To override the test list for the job, you need to familiar with the
ginkgo.focus
and ginkgo.skip
flags. Using these flags, you can do:
export GINKGO_TEST_ARGS="--ginkgo.focus=\[Feature:Performance\]"
and follow the instructions above. Here are some other examples from the e2e.go
documentation..
If you want to test against an existing cluster, you can do:
export E2E_UP=false; export E2E_DOWN=false
and follow the instructions above. This is particularly useful for testing the
myriad of kops
configuration/topology options without having to modify the
testing infrastructure. Note: This is also the only way currently to test a
custom Kubernetes build
(see
kubernetes/test-infra#1454).
If you want to upload a custom Kubernetes build, here is a simple way (note:
this assumes you've run make quick-release
in the Kubernetes repo first):
# cd wherever you tend to put git repos
git clone https://github.com/kubernetes/release.git
# cd back to your kubernetes repo
/path/to/release/push-build.sh # Fix /path/to/release with wherever you cloned the release repo
That will upload the release to a GCS bucket and make it public. You can then
use the outputted URL in kops
with --kubernetes-version
.
If you need it private in S3, here's a manual way:
make quick-release
cd ./_output/release-tars/
# ??? rm -rf kubernetes/
tar zxf kubernetes-server-linux-amd64.tar.gz
rm kubernetes/server/bin/federation*
rm kubernetes/server/bin/hyperkube
rm kubernetes/server/bin/kubeadm
rm kubernetes/server/bin/kube-apiserver
rm kubernetes/server/bin/kube-controller-manager
rm kubernetes/server/bin/kube-discovery
rm kubernetes/server/bin/kube-dns
rm kubernetes/server/bin/kubemark
rm kubernetes/server/bin/kube-proxy
rm kubernetes/server/bin/kube-scheduler
rm kubernetes/kubernetes-src.tar.gz
find kubernetes/server/bin -type f -name "*.tar" | xargs -I {} /bin/bash -c "sha1sum {} | cut -f1 -d ' ' > {}.sha1"
find kubernetes/server/bin -type f -name "kube???" | xargs -I {} /bin/bash -c "sha1sum {} | cut -f1 -d ' ' > {}.sha1"
aws s3 sync --acl public-read kubernetes/server/bin/ s3://${S3_BUCKET_NAME}/kubernetes/dev/v1.6.0-dev/bin/linux/amd64/
go run hack/e2e.go -v -up -down -kops `which kops` -kops-cluster test.test-aws.k8s.io -kops-state s3://k8s-kops-state-store/ -kops-nodes=
4 -deployment kops --kops-kubernetes-version https://storage.googleapis.com/kubernetes-release-dev/ci/$(curl -SsL https://storage.googleapis.com/kubernetes-release-dev/ci/latest-green.txt)
(note the v1.6.0-dev
: we insert a kubernetes version so that kops can
automatically detect which k8s version is in use, which it uses to control
flags that are not compatible between versions)
Then:
-
kops create cluster ... --kubernetes-version https://${S3_BUCKET_NAME}.s3.amazonaws.com/kubernetes/dev/v1.6.0-dev/
-
for an existing cluster:
kops edit cluster
and setKubernetesVersion
tohttps://${S3_BUCKET_NAME}.s3.amazonaws.com/kubernetes/dev/v1.6.0-dev/