Kubernetes 1.10 #258

asymmetric · 2018-04-25T13:14:41Z

See #219 for things this changes.

Closes #219.
Closes #171.

krnowak

Looks mostly fine I suppose, but thought that we are going to make v1beta3 custom version for the v1beta2 statefulsets. But maybe it will not be required, as supposedly you should be able to create a v1beta1 statefulset and handle it as v1beta2 statefulset.

krnowak · 2018-04-27T09:08:17Z

.circleci/config.yml

      - run:
          name: code-gen script
          environment:
            CODEGEN_PKG: ../../../../src/k8s.io/code-generator
-            CODEGEN_VERSION: release-1.9
+            CODEGEN_VERSION: kubernetes-1.10.0


Shouldn't it be release-1.10?

Whatever outcome is (either a tag or a branch), we should keep docs in the update codegen script in sync.

The release- are branches, i.e. they get updated, so we cannot guarantee that it will always be the same commit.

What docs are you referring to?

These ones:

habitat-operator/hack/update-codegen.sh

Lines 20 to 28 in 3adc004

# First of all, the code-generator needs to be fetched:

# $ go get -u k8s.io/code-generator

#

# Now checkout the appropriate branch for the Kubernetes version that

# you're building for. For example if you're building for Kubernetes

# 1.10.0 you would checkout the `release-1.10` branch.

#

# Then run this from the repo's root with:

# $ CODEGEN_PKG=../../../k8s.io/code-generator hack/update-codegen.sh

krnowak · 2018-04-27T09:09:15Z

.circleci/config.yml

-          name: print minikube logs
-          command: minikube logs
-          when: on_fail
+      # NOTE: this relies on journalctl, which is not present on the version of


So how did it work before? Old minikube/k8s didn't require journalctl?

Old minikube didn't use journalctl.

krnowak · 2018-04-27T09:09:27Z

.circleci/config.yml

-          command: minikube logs
-          when: on_fail
+      # NOTE: this relies on journalctl, which is not present on the version of
+      # Ubuntu currenlty run by CircleCI


krnowak · 2018-04-27T09:09:59Z

.circleci/config.yml

+      # NOTE: this relies on journalctl, which is not present on the version of
+      # Ubuntu currenlty run by CircleCI
+      # - run:
+          # - runame: print minikube logs


runame rucommand and ruwhen?

krnowak · 2018-04-27T09:10:31Z

.travis.yml

-    - CODEGEN_VERSION=release-1.9
+    - K8S_VERSION=v1.10.0
+    - MINIKUBE_VERSION=v0.26.1
+    - CODEGEN_VERSION=kubernetes-1.10.0


release-1.10?

Same as above

krnowak · 2018-04-27T09:16:46Z

pkg/controller/v1beta1/controller.go

@@ -167,11 +167,11 @@ func (hc *HabitatController) cacheHabitats() {
 }

 func (hc *HabitatController) cacheDeployments() {
-	source := newListWatchFromClientWithLabels(
+	source := cache.NewFilteredListWatchFromClient(


Should we touch this code at all? v1beta1 custom version is frozen.

That's true, OTOH it should be a drop-in replacement. Not sure, should I remove?

Yeah, I'd just leave it alone.

krnowak · 2018-04-27T09:18:12Z

pkg/controller/v1beta2/controller.go

@@ -548,11 +548,11 @@ func (hc *HabitatController) conform(key string) error {
 	}

 	// Create StatefulSet, if it doesn't already exist.
-	if _, err := hc.config.KubernetesClientset.AppsV1beta1().StatefulSets(h.Namespace).Create(sts); err != nil {
+	if _, err := hc.config.KubernetesClientset.AppsV1beta2().StatefulSets(h.Namespace).Create(sts); err != nil {


This probably should be a part of v1beta3 custom version, no? But maybe releasing v1beta3 will not be required, as supposedly you should be able to create a v1beta1 statefulset and handle it as v1beta2 statefulset. Maybe same goes for caching and so on. Could you check it?

So I've:

Started a 1.10 cluster

Run the operator (out of cluster) on master (meaning, sts v1beta1 controller)

Created the standalone example

Killed the operator

Checked out this branch

Started the operator

Edited the Habitat manifest

The updates were picked up by the operator.

So the v1beta1 resource is delivered to the v1beta2 watcher. So we don't need to upgrade to v1beta3 for this change.

asymmetric · 2018-05-02T10:59:25Z

PTALA.

krnowak

LFAD in general, but we probably want sign-offs in all the commits, no? Some initial commits seem to lack it.