Implement persistence with StatefulSets #201

asymmetric · 2018-03-06T14:36:00Z

This PR

Changes the type of object we create from Deployment to StatefulSet
Adds a Persistence key to the Habitat type
Starts splitting controller.go into separate modules
Removes docs/api.md and just points users to types.go
~~Adds tests for persistence~~

The main issue with using Deployments for persistence was that there was no automatic way to assign a different PVC to each Pod, so we ended up with Pods that shared the same Volume.

See #44 and #195 for more in-depth discussions.

NOTE: Tests are included but skipped, as they don't work in our current testing environment (it's not possible to provision PersistentVolumes when running minikube with --vm-driver=none).

Closes #44.
Closes #38.
Closes #181.

asymmetric · 2018-03-06T14:36:46Z

@lilic could you run the deepcopy script against this branch, and see if there's any change?

jeremymv2 · 2018-03-08T23:34:53Z

lilic

Left some initial comments, didn't have time to look deeper into this, as it is a large PR.

lilic · 2018-03-08T09:26:04Z

examples/rbac/rbac.yml

@@ -34,7 +34,7 @@ rules:
 - apiGroups:
  - apps
  resources:
-  - deployments


Should we not need Volume or something similar here as well?

No I don't think so, as it's not the operator that's creating Volumes, it's the StatefulSet.

You were right btw, we did have to: the PV/PVC ListWatchers running in the operator means we need to assign list and watch permissions.

So I am guessing as this is not needed anymore the above comment is not valid, right?

lilic · 2018-03-08T09:26:43Z

pkg/apis/habitat/v1beta1/zz_generated.deepcopy.go

@@ -142,6 +151,22 @@ func (in *HabitatStatus) DeepCopy() *HabitatStatus {
 	return out
 }

+// DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.


Cool you got it to work! 👍 How did you manage to do that?

I didn't :) This is your commit actually, from the Deployments PR.

lilic · 2018-03-09T11:33:37Z

examples/persisted/habitat.yml

+    mountPath: /tmp/foobar
+  env:
+    - name: HAB_REDIS
+      value: '{ "protected-mode": "no" }'


Please add a note of warning here.

lilic · 2018-03-09T11:39:20Z

pkg/apis/habitat/v1beta1/types.go

+	Persistence *Persistence `json:"persistence,omitempty"`
+}
+
+type Persistence struct {


Can we call this Storage instead? Things make a bit more sense, for example you then have Size, ClassName as part of Storage. WDYT?

Not sure that Storage by itself makes it clear that this is persistent rather than ephemeral.

PersistantStorage? :)

I'm afraid you're gonna have to re-run the script though :)

lilic · 2018-03-09T11:45:52Z

pkg/controller/controller.go

@@ -922,11 +679,20 @@ func habitatKeyFromLabeledResource(r metav1.Object) (string, error) {
 func newConfigMap(ip string, h *habv1beta1.Habitat) *apiv1.ConfigMap {
 	return &apiv1.ConfigMap{
 		ObjectMeta: metav1.ObjectMeta{
+			// TODO the name should also change to support multiple habitats


Can you explain this TODO a bit more?

Oops, that was just for me, I'll remove. Anyway, I think that making the CM name a constant has the drawback that we can only have one. In the future, this could pose problems if we have more than one ring in the same namespace.

The problem with is that is how would we handle binding if they are not in one ring. That's why we decided to go with one ring...

Yes, I didn't mean to say that I thought it through, that's why it was just there for me as a reminder.

lilic · 2018-03-09T11:55:03Z

pkg/controller/stateful_sets.go

+	h, err := hc.getHabitatFromLabeledResource(d)
+	if err != nil {
+		// Could not find Habitat, it must have already been removed.
+		level.Debug(hc.logger).Log("msg", "Could not find Habitat for StatefulSet", "name", d.Name)


Hmm this msg doesn't make much sense to me as a consumer, maybe adding Could not find Habitat resource for StatefulSet would be better.

Well, it just uses the same format as every other object. We don't say "StatefulSet resource" after all.

Yes understand, but it can be confusing for a user as the name conveys two things. :) Anyways, more of a nit for clarity.

lilic · 2018-03-09T12:00:50Z

test/e2e/operator_test.go

@@ -183,3 +187,67 @@ func TestHabitatDelete(t *testing.T) {
 		t.Fatal(err)
 	}
 }
+
+func TestHabitatPersistence(t *testing.T) {
+	// We run minikube in a VM on Travis. In that environment, we cannot create PersistentVolumes.


Can you explain this a bit, what is the reason behind this?

Also if it doesn't work why do we have this test? I am a bit confused.

The reason is that since we run the operator in containers on Travis, the dynamic volume provisioning that Minikube normally does (when it's on a VM) doesn't work.

I think leaving the test in makes sense because:

it took time to write it, so we would lose that by deleting it

we can reactivate it when we move away from testing with minikube (which I think we should)

it documents that there's a problem and what it is

Personally don't like dead code in master, maybe you could just cherry-pick this and keep it in a branch so the work is not lost.

I think that if it's marked as skip, then it's fine. It's common for tests to be temporarily marked as skippable IMO. It would be different if it was business logic, but I think it's fine for tests. I'd rather keep this in unless it's blocking for you.

lilic · 2018-03-13T09:58:31Z

@asymmetric I wanted to try this out but can't seem to compile it locally. But it seems to be running fine on travis. 😕 Does make build work for you?

lilic

First true pass over the PR. Left a few comments.

Also don't forget to update the docs/api.md file ;)

lilic · 2018-03-13T16:54:41Z

pkg/controller/stateful_sets.go

+
+		q, err := resource.ParseQuantity(ps.Size)
+		if err != nil {
+			return nil, err


Can we format this error with a prefix, something like "PersistantVolume resource size error: ...".

The reason for this is, until we have the validations in place, the errors this will generate sometimes seem to be very non descriptive.

lilic · 2018-03-13T16:57:56Z

pkg/controller/persistent_volumes.go

+	source := newListWatchFromClientWithLabels(
+		hc.config.KubernetesClientset.CoreV1().RESTClient(),
+		"persistentvolumes",
+		apiv1.NamespaceAll,


Should we be caching all the PV in all the namespaces?

I guess so, because users can create Habitats (and therefore PVs) in any namespace. Why would this case be different from other resources we watch?

lilic · 2018-03-13T16:59:34Z

pkg/controller/persistent_volumes.go

+	)
+
+	hc.pvInformer.AddEventHandler(cache.ResourceEventHandlerFuncs{
+		DeleteFunc: hc.handlePVDelete,


What about on update? Should we react on that as well?

I think that at least for now, we can avoid implementing storage resize: it's an optional feature that introduces quite a lot of complexity. I'd wait for the changes introduced in this PR to have stabilized a bit (and for someone to request it) before we figure out how to do it best.

But we should definitely make this clear and notify the user somehow, when they change the Size in the manifest, that it's not supported. How to do that though? This is probably one of the things we should add to #200 (and/or to #185?)

Then IMO I don't see the value of caching PV informer or rather reacting to the changes. Same for the PVC informer. Clean up of volumes is intentionally manual and the rest is already handled by the StatefulSet. The user is already informed, via monitoring, logs, etc., so I feel like we are just duplicating that here. Especially since we are caching ALL the PVC/PV, that can be a huge number in a normal cluster.

We are only caching ones with label habitat: true. Also, the point of caching is that they could be removed by anything/anybody else, and we need to be able to notify the user - and most importantly, in the future, update the Status of the Habitat object.

According to the Controller Bible, verse 7:

There are other actors in the system. Just because you haven't changed an object doesn't mean that somebody else hasn't.

and we need to be able to notify the user

But that is already done by its respective controller.

Agree that we should react, but we are not doing anything apart from logging. So I don't see the point of it, IMHO. :)

We will update the Status in #200 :)

lilic · 2018-03-13T17:01:57Z

pkg/controller/persistent_volumes.go

+	} else if !exists {
+		level.Debug(hc.logger).Log("msg", "No matching Habitat found for PVC", "name", pv.Name)
+		return
+	} else {


I think we can get rid of this else as we return in other cases anyways.

I think in this case it makes more sense to keep it in in this case, because it's a 3-way situation we're dealing with, and having them together helps (me) map the idea to the code better.

Usually with 2-branch conditionals that's not necessary, but I find that it helps here.

WDYT?

I ended up agreeing with you :)

lilic · 2018-03-13T17:09:08Z

pkg/controller/persistent_volume_claims.go

@@ -0,0 +1,88 @@
+// Copyright (c) 2017 Chef Software Inc. and/or applicable contributors


It's 2018 :)

lilic · 2018-03-13T17:09:32Z

pkg/controller/persistent_volumes.go

@@ -0,0 +1,69 @@
+// Copyright (c) 2017 Chef Software Inc. and/or applicable contributors


It's 2018 :)

lilic · 2018-03-13T17:09:43Z

pkg/controller/stateful_sets.go

@@ -0,0 +1,325 @@
+// Copyright (c) 2017 Chef Software Inc. and/or applicable contributors


It's 2018 :)

asymmetric · 2018-03-14T13:16:49Z

About the docs/api.md file: is it useful? I keep forgetting to keep it in sync, and I'd like to know if there's a good reason to do that.

Can't we just have a link to pkg/apis/habitat/v1beta1/types.go? Fields are commented there as well.

lilic · 2018-03-14T13:23:15Z

About the docs/api.md file: is it useful? I keep forgetting to keep it in sync, and I'd like to know if there's a good reason to do that.

I just find the documentation helpful when needing to reference something, but we could also implement a helper script for that? Or just get rid of it, up to you. :)

lilic · 2018-03-14T14:02:04Z

pkg/controller/controller.go

@@ -927,6 +684,14 @@ func newConfigMap(ip string, h *habv1beta1.Habitat) *apiv1.ConfigMap {
 			Labels: map[string]string{
 				habv1beta1.HabitatLabel: "true",
 			},
+			OwnerReferences: []metav1.OwnerReference{
+				metav1.OwnerReference{
+					APIVersion: fmt.Sprintf("%s/%s", habv1beta1.GroupName, habv1beta1.Version),


There is h.APIVersion you can use.

lilic · 2018-03-14T14:02:27Z

pkg/controller/controller.go

+			OwnerReferences: []metav1.OwnerReference{
+				metav1.OwnerReference{
+					APIVersion: fmt.Sprintf("%s/%s", habv1beta1.GroupName, habv1beta1.Version),
+					Kind:       "Habitat",


Same here there is h.Kind you can use.

lilic · 2018-03-14T14:06:49Z

pkg/controller/stateful_sets.go

+	base := &appsv1beta1.StatefulSet{
+		ObjectMeta: metav1.ObjectMeta{
+			Name: h.Name,
+		},


Add a OwnerReference to the StatefulSet as well, like above in the ConfigMap.

We currently do that here, but I agree we could simply delegate this to Kubernetes.

lilic · 2018-03-14T14:07:41Z

pkg/controller/controller.go

+					Kind:       "Habitat",
+					Name:       h.Name,
+					UID:        h.UID,
+				},


Also I noticed we do not say anything about BlockOwnerDeletion, I would add this here. It is very useful IMO.

if the owner has the "foregroundDeletion" finalizer

Is this our case?

It's not entirely clear for me if any defaults are defined for the owning resource (Habitat). Otherwise, it seems to me that propagationPolicy could be specified by a user in their DELETE request. Is this something we should be worried about?

Not sure if it's something to worry about, just thought it was nice to have.

krnowak

Nits mostly I think.

krnowak · 2018-03-19T12:57:31Z

pkg/controller/persistent_volume_claims.go

+	hc.pvcInformerSynced = hc.pvcInformer.HasSynced
+}
+
+// findHabForPVC looks for a matching Habitat object for a PVC.


I'd name it rather like checkHabForPVC. findHabForPVC makes me think that it will return a Habitat object.

krnowak · 2018-03-19T13:01:01Z

pkg/controller/persistent_volumes.go

+		return
+	}
+
+	_, exists, err := hc.habInformer.GetStore().GetByKey(key)


That part of code looks like a copy pasta of the findHabForPVC function. Maybe just call it here?

Actually, I think there are 2 problems with my approach here:

We are not telling users that they PersistentVolumes they create (if they're going with static provisioning) need to have the Habitat=true label

We find the Habitat by name, which seems like a weak approach. Maybe a label would be better? Considering that the PV might be created beforehand, it seems weird to force the user to know already what the name should be. A label, OTOH, can be added later, and seems a better approach than asking the user to rename the PV.

krnowak · 2018-03-19T13:08:43Z

pkg/controller/stateful_sets.go

+			ReadOnly:  false,
+		}
+
+		base.Spec.Template.Spec.Containers[0].VolumeMounts = append(base.Spec.Template.Spec.Containers[0].VolumeMounts, *secretVolumeMount)


You could probably do spec := &base.Spec.Template.Spec and then spec.Containers[0].VolumeMounts = … to drop a bit of a repeating the same thing.

Or even move this spec := … outside this block, so it could be also used for mounting the persistent volume or the ring secret volume.

krnowak · 2018-03-19T13:20:06Z

pkg/controller/controller.go

+	habInformerSynced cache.InformerSynced
+	stsInformerSynced cache.InformerSynced
+	cmInformerSynced  cache.InformerSynced
+	pvInformerSynced  cache.InformerSynced


What are pvInformerSynced and pvcInformerSynced for? They seem to be unused. Did you forget to use them in WaitForCacheSync call?

To highlight the "happy path".

Makes it clear what the gola of the method is.

"find" made it seem like the method returned a Habitat object, whereas we are only interested in the side effects.

The operator runs ListWatchers for PVs and PVCs.

It is a duplicate of cluster-role.yml

The `TypeMeta` field is not populated for the object passed to the handler, so we need to calculate these two fields ourselves.

The updates serve to clarify that both static and dynamic provisioning are supported, but that users should read the official documentation in order to find out how to set them up.

The implementation relied on PVs being labeled with Habitat=true, which does not happen automatically when PVs are dynamically provisioned.

This reverts commit 5910452. The changes here were related to PV and PVC watching, which have been removed in a previous patch.

We are not watching PVs and PVCs anymore in this PR. This reverts commit 453e05a.

lilic

Few questions. Otherwise think we should be done with this soon and you can iterate on this in the future.

lilic · 2018-03-20T13:45:27Z

.circleci/config.yml

@@ -0,0 +1,39 @@
+version: 2


Is now both travis and circleci used? 🤔

lilic · 2018-03-20T13:45:51Z

docs/api.md

@@ -1,39 +0,0 @@
-# API v1beta1


lilic · 2018-03-20T13:48:20Z

examples/rbac/rbac.yml

@@ -34,7 +34,7 @@ rules:
 - apiGroups:
  - apps
  resources:
-  - deployments


So I am guessing as this is not needed anymore the above comment is not valid, right?

lilic · 2018-03-20T14:07:10Z

pkg/controller/controller.go

@@ -347,15 +274,15 @@ func (hc *HabitatController) enqueueCM(obj interface{}) {
 }

 func (hc *HabitatController) handleCMAdd(obj interface{}) {
-	hc.enqueueCM(obj)
+	hc.handleCM(obj)


Not sure if this is in the scope of this PR or not, but since you are redoing the logic of resources being created or not, I open an issue about CM not being deleted when Habitat resource is deleted, so maybe it's something worth looking into. It can be outside of this PR as well, up to you...

👍. I'll tackle it in another PR.

lilic · 2018-03-20T14:08:51Z

pkg/controller/stateful_sets.go

+	"fmt"
+
+	"github.com/go-kit/kit/log/level"
+	habv1beta1 "github.com/habitat-sh/habitat-operator/pkg/apis/habitat/v1beta1"


My personal preference is to have the these imports separated with a line break from the outside imports.

Your wish is my command.

lilic · 2018-03-20T14:09:11Z

pkg/controller/stateful_sets.go

+	"k8s.io/client-go/tools/cache"
+)
+
+const persistentVolumeName = "persistent"


Why is this name hardcoded?

Because it never changes ;)

It's a bit like the name for the CM volume. We hardcode it, and since we are good programmers, we move it to a constant ;)

Basically, there's no reason to expose this name and make it configurable I think.

lilic · 2018-03-20T14:13:41Z

test/e2e/framework/helpers.go

@@ -223,7 +224,7 @@ func ConvertSecret(pathToYaml string) (*v1.Secret, error) {
 	return &s, nil
 }

-func convertToK8sResource(pathToYaml string, into interface{}) error {
+func convertToK8sResource(pathToYaml string, into runtime.Object) error {


lilic · 2018-03-20T14:19:12Z

BTW since this is a huge breaking change IMO we really need to bump up the version here of the API, so to v2beta1 or something like that. As that we don't get problems if this operator is deployed into a cluster where Habitat resources, e.g. Deployments already exist as that cannot be handled by this operator.

For clarity.

asymmetric · 2018-03-21T12:59:20Z

We agreed offline to keep the current Deployment-based behavior in v1beta1, alongside the new StatetfulSet-based one on v1beta2. This will be done in a folloging PR.

So please go ahead and do your final reviews :)

lilic

Mainly questions and suggestions. Nothing blocking.

lilic · 2018-03-21T13:17:56Z

pkg/apis/habitat/v1beta1/types.go

+	// MountPath is the path at which the PersistentVolume will be mounted.
+	MountPath string `json:"mountPath"`
+	// StorageClassName is the name of the StorageClass that the StatefulSet will request.
+	StorageClassName string `json:"storageClassName"`


Nit: Can we name this just ClassName as it's under the PersistantStorage anyways. WDYT?

I gave it the same name it has in the PersistentVolume resource, because it maps 1:1 with that field. My hope is that it will make it easier for people to understand how things tie together.

Cool, that makes sense!

lilic · 2018-03-21T13:18:32Z

pkg/apis/habitat/v1beta1/types.go

+type PersistentStorage struct {
+	// Size is the volume's size.
+	// It uses the same format as Kubernetes' size fields, e.g. 10Gi
+	Size string `json:"size"`


You are missing the omitempty, or are these are required? Then we should document that as we don't have the api docs.md anymore. :)

What do you mean? Right now we only explicitly document that a field is optional, otherwise it's mandatory.

Ok, just before in the docs.md file it was more clear at least to me, what is required. But if it makes sense to you 👍

AFAICT that's also how its' done in upstream.

lilic · 2018-03-21T13:25:02Z

pkg/controller/controller.go

+			OwnerReferences: []metav1.OwnerReference{
+				metav1.OwnerReference{
+					APIVersion: habv1beta1.SchemeGroupVersion.String(),
+					Kind:       habv1beta1.HabitatKind,


What do we use this OwnerReference for now?

In theory so that the CM is deleted with the Habitat, but apparently that's not working?

That doesn't link to anything for me? 🤔

lilic · 2018-03-21T13:33:54Z

examples/persisted/README.md

@@ -0,0 +1,37 @@
+# Persistent Storage example


So the whole point of persistence is that you can mount already existing volumes, can you maybe write up some steps of how a user can do that.

I explicitly avoided to do that because it's a big topic and I would only make a poorer job at explaining it than the official docs do, which I linked. I anyway added some commits after your comment which should hopefully make things clearer.

BTW, it's not necessary that the PV has been created beforehand, if the cluster supports dynamic provisioning.

TL;DR: IMHO explaining how persistence works in Kubernetes is beside the scope of this README.

lilic · 2018-03-21T13:34:38Z

examples/persisted/README.md

@@ -0,0 +1,37 @@
+# Persistent Storage example
+
+Habitat objects are translated by the operator into `StatefulSet` objects, which


Nit: Maybe add a note that the user needs to know about basic understanding of persistent volumes in K8s and link to it?

OK, I'll add a link to this

lilic · 2018-03-21T13:38:35Z

pkg/controller/stateful_sets.go

+
+const persistentVolumeName = "persistent"
+
+func (hc *HabitatController) newStatefulSet(h *habv1beta1.Habitat) (*appsv1beta1.StatefulSet, error) {


You know now that you extracted this you can actually write a unit test for this. WDYT?

lilic · 2018-03-21T13:51:17Z

pkg/controller/controller.go

@@ -802,15 +526,16 @@ func (hc *HabitatController) processNextItem() bool {

 // conform is where the reconciliation takes place.
 // It is invoked when any of the following resources get created, updated or deleted:
-// Habitat, Pod, Deployment, ConfigMap.
+// Habitat, Pod, StatefulSet, ConfigMap.
 func (hc *HabitatController) conform(key string) error {
 	obj, exists, err := hc.habInformer.GetStore().GetByKey(key)


Should we not handle conforming the PersistentVolumeClaim as well? Or is there nothing that needs to be done, when for example Habitat resource gets deleted or updated?

As mentioned in the example README, we do nothing to the PV/PVC when the Habitat gets deleted, because we want to prioritize data safety (which is also the default behavior with STS).

Do you think we should do something?

To avoid having to have specific instructions for each platform. On GKE or minikube, users could use the default SC (thereby leaving the field empty), or could specify them by their name (and those name differ between GKE and minikube). Instead, we just ask them to create a custom StorageClass for the example.

These rules are required for clusters running on minikube.

The file has been removed in a previous commit.

Should act as an entry point for users interested in learning how persistent storage works in Kubernetes.

lilic

LGTM if it works :)

lilic · 2018-03-21T14:44:22Z

examples/rbac/minikube.yml

+  kind: ClusterRole
+  name: cluster-writer
+  apiGroup: rbac.authorization.k8s.io
+


Line too much?

lilic · 2018-03-21T14:44:27Z

examples/rbac/minikube.yml

+    verbs: ["*"]
+  - nonResourceURLs: ["*"]
+    verbs: ["*"]
+


One line too much. :)

It's required/recommended to have a newline at the end of a file. My editor adds it automatically.

Git complains if you don't have it.

Except that's not the end 🤣

lilic · 2018-03-21T14:44:37Z

examples/rbac/minikube.yml

+  apiGroup: rbac.authorization.k8s.io
+
+---
+


Remove line? :)

I actually think that's not even needed, not sure why it's there in the first place. There's no sd-build ServiceAccount on my minikube cluster.

Wrong position 😂 My comment referred to the bottom ClusterRoleBinding.

lilic · 2018-03-21T14:45:13Z

examples/persisted/README.md

+* specify the `name` of the aforementioned `StorageClass` object under
+`spec.persistence.storageClassName` in the Habitat object's manifest
+
+An example `StorageClass` for clusters running on minikube is provided in


Maybe note it doesn't work if it's vm-driver=none?

I don't think anyone's gonna try it with that flag, considering it's not even documented ;)

Think it is now. :D

The `sd-build` `ServiceAccount` is not present anyway.

asymmetric requested review from zeenix, krnowak and lilic March 6, 2018 14:36

asymmetric force-pushed the asymmetric/sts branch from a2142bf to 9028fad Compare March 6, 2018 15:48

asymmetric changed the title ~~Implement persistence with StatefulSets~~ [WIP] Implement persistence with StatefulSets Mar 6, 2018

asymmetric force-pushed the asymmetric/sts branch 10 times, most recently from b9d4d1c to 0986653 Compare March 7, 2018 15:49

asymmetric changed the title ~~[WIP] Implement persistence with StatefulSets~~ Implement persistence with StatefulSets Mar 7, 2018

lilic suggested changes Mar 9, 2018

View reviewed changes

asymmetric force-pushed the asymmetric/sts branch 3 times, most recently from 88654b5 to 66bd8f9 Compare March 12, 2018 16:17

lilic suggested changes Mar 13, 2018

View reviewed changes

lilic suggested changes Mar 14, 2018

View reviewed changes

asymmetric force-pushed the asymmetric/sts branch 2 times, most recently from d58e58f to 3a1151d Compare March 14, 2018 14:42

krnowak suggested changes Mar 19, 2018

View reviewed changes

asymmetric added 12 commits March 20, 2018 14:55

Extract else branch out of conditional

6dfc118

To highlight the "happy path".

Improve documentation of findHabForPVC method

89e4eea

Makes it clear what the gola of the method is.

Clarify TODO in findHabForPVC

0573853

Rename findHabForPVC to checkHabForPVC

78468bb

"find" made it seem like the method returned a Habitat object, whereas we are only interested in the side effects.

Add cluster-level permissions for operator

f4b9093

The operator runs ListWatchers for PVs and PVCs.

Remove rbac.yml

8d7720c

It is a duplicate of cluster-role.yml

Calculate APIVersion and Kind

d17efa3

The `TypeMeta` field is not populated for the object passed to the handler, so we need to calculate these two fields ourselves.

Update PersistentStorage README

96d7aa1

The updates serve to clarify that both static and dynamic provisioning are supported, but that users should read the official documentation in order to find out how to set them up.

Stop caching PVs and PVCs

8e9e0e2

The implementation relied on PVs being labeled with Habitat=true, which does not happen automatically when PVs are dynamically provisioned.

Revert "Update PersistentStorage README"

5d16382

This reverts commit 5910452. The changes here were related to PV and PVC watching, which have been removed in a previous patch.

Fix key name in README

1ed628c

Revert "Add cluster-level permissions for operator"

4a6f295

We are not watching PVs and PVCs anymore in this PR. This reverts commit 453e05a.

asymmetric force-pushed the asymmetric/sts branch from 349a57e to 4a6f295 Compare March 20, 2018 13:56

lilic suggested changes Mar 20, 2018

View reviewed changes

Visually separate internal imports

64e7f10

For clarity.

lilic reviewed Mar 21, 2018

View reviewed changes

asymmetric added 3 commits March 21, 2018 15:20

Add RBAC rules for minikube

4f327c9

These rules are required for clusters running on minikube.

Remove stale link ot docs/api.md

efe0e8f

The file has been removed in a previous commit.

asymmetric force-pushed the asymmetric/sts branch from 906a68f to efe0e8f Compare March 21, 2018 14:22

Add link to PersistentVolume in example

f3af7cd

Should act as an entry point for users interested in learning how persistent storage works in Kubernetes.

lilic approved these changes Mar 21, 2018

View reviewed changes

asymmetric added 2 commits March 21, 2018 15:55

Remove unneeded ClusterRoleBinding

b8876df

The `sd-build` `ServiceAccount` is not present anyway.

Add warning about minikube --vm-driver=none

49da9a4

krnowak approved these changes Mar 21, 2018

View reviewed changes

asymmetric merged commit 5e2c533 into master Mar 21, 2018

asymmetric deleted the asymmetric/sts branch March 21, 2018 15:23

asymmetric mentioned this pull request Mar 30, 2018

Old operator image in examples #231

Open

		@@ -0,0 +1,88 @@
		// Copyright (c) 2017 Chef Software Inc. and/or applicable contributors

		@@ -0,0 +1,69 @@
		// Copyright (c) 2017 Chef Software Inc. and/or applicable contributors

		@@ -0,0 +1,325 @@
		// Copyright (c) 2017 Chef Software Inc. and/or applicable contributors

		@@ -0,0 +1,37 @@
		# Persistent Storage example

		Habitat objects are translated by the operator into `StatefulSet` objects, which


		const persistentVolumeName = "persistent"

		func (hc HabitatController) newStatefulSet(h habv1beta1.Habitat) (*appsv1beta1.StatefulSet, error) {

Implement persistence with StatefulSets #201

Implement persistence with StatefulSets #201

Conversation

asymmetric commented Mar 6, 2018 • edited Loading

asymmetric commented Mar 6, 2018

jeremymv2 commented Mar 8, 2018

lilic left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

asymmetric Mar 19, 2018 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

asymmetric Mar 12, 2018 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

asymmetric Mar 12, 2018 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

lilic commented Mar 13, 2018 • edited Loading

lilic left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

asymmetric Mar 14, 2018 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

asymmetric Mar 14, 2018 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

asymmetric Mar 14, 2018 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

asymmetric commented Mar 14, 2018 • edited Loading

lilic commented Mar 14, 2018

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

krnowak left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

asymmetric Mar 19, 2018 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

lilic left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

asymmetric commented Mar 6, 2018 •

edited

Loading

asymmetric Mar 19, 2018 •

edited

Loading

asymmetric Mar 12, 2018 •

edited

Loading

asymmetric Mar 12, 2018 •

edited

Loading

lilic commented Mar 13, 2018 •

edited

Loading

asymmetric Mar 14, 2018 •

edited

Loading

asymmetric Mar 14, 2018 •

edited

Loading

asymmetric Mar 14, 2018 •

edited

Loading

asymmetric commented Mar 14, 2018 •

edited

Loading

asymmetric Mar 19, 2018 •

edited

Loading