Day 1: Installing kro, First RGD
Goal: kro runs in the cluster. You have built your first API
Greeting, created an instance and seen how kro guards the generated objects. · Duration: approx. 45 min · Prerequisite: Day 0
Briefing
- kro comes as a Helm chart from
registry.k8s.io(the image registry of the Kubernetes project). kro is a subproject of Kubernetes SIG Cloud Provider. - The chart brings three CRDs:
ResourceGraphDefinition(shortrgd),Graph(alpha, Day 14) andGraphRevision(shortgr, Day 11). - When you create an RGD, kro does four things: it validates it, generates a CRD for your new kind, starts a controller for it and creates the objects as soon as someone creates an instance.
RGD greeting ──► CRD greetings.kro.run ──► Instance Greeting/hallo
│ kro applies
▼
ConfigMap hallo
Drill 1: Installing kro
helm install kro oci://registry.k8s.io/kro/charts/kro \
--namespace kro-system --create-namespace --version 0.10.0Expected (shortened):
Pulled: registry.k8s.io/kro/charts/kro:0.10.0
Digest: sha256:56561368ed4c574520e8bd6a677a2e6c66333fb90e759ff4f2f068341fe9956c
NAME: kro
NAMESPACE: kro-system
STATUS: deployed
REVISION: 1
oci://: The chart lives as an OCI artifact in a registry; nohelm repo addis needed.--version 0.10.0pins the version. Without it, Helm takes the latest.
kubectl -n kro-system get podsExpected:
NAME READY STATUS RESTARTS AGE
kro-69965dc9cc-cgc4f 1/1 Running 0 31s
A single pod; that is the whole controller.
Save your progress? Sign up for free
Drill 2: What kro Brings Along
kubectl get crd | grep kroExpected:
graphrevisions.internal.kro.run 2026-11-11T17:28:00Z
graphs.kro.run 2026-11-11T17:28:00Z
resourcegraphdefinitions.kro.run 2026-11-11T17:28:00Z
kubectl api-resources --api-group=kro.runExpected:
NAME SHORTNAMES APIVERSION NAMESPACED KIND
graphs kro.run/v1alpha1 true Graph
resourcegraphdefinitions rgd kro.run/v1alpha1 false ResourceGraphDefinition
An RGD is cluster-scoped
(NAMESPACED false). It defines an API for the whole
cluster, just like a CRD.
Helm shows you the chart’s settings:
helm show values oci://registry.k8s.io/kro/charts/kro --version 0.10.0 \
| grep -A3 '^rbac:'Expected:
rbac:
# Specifies how to manage access for kro in the cluster
# Two modes are currently supported:
# - unrestricted: Grants cluster-wide full access to all resources
The default is unrestricted: kro may do
everything in the cluster. That’s convenient for the
lab. On Day 12 you switch to aggregation.
Save your progress? Sign up for free
Drill 3: The First RGD
You build an API Greeting. Each instance gets a
ConfigMap with a message.
mkdir -p ~/kro-lab/tag01
cd ~/kro-lab/tag01
cat > greeting-rgd.yaml <<'EOF'
apiVersion: kro.run/v1alpha1
kind: ResourceGraphDefinition
metadata:
name: greeting
spec:
schema:
apiVersion: v1alpha1
kind: Greeting
spec:
message: string
resources:
- id: config
template:
apiVersion: v1
kind: ConfigMap
metadata:
name: ${schema.metadata.name}
data:
message: ${schema.spec.message}
EOFschemadescribes the new API: kindGreeting, versionv1alpha1, one fieldmessageof typestring. The group is missing, so kro useskro.run.resourcesis the list of objects. Each has anid(hereconfig) and atemplate, a completely normal Kubernetes manifest.${schema.metadata.name}is a CEL expression. It reads the name of the instance.${schema.spec.message}reads the field the team sets.'EOF'in quotes is important: that way the shell leaves the${…}alone.
kubectl apply -f greeting-rgd.yaml
kubectl get rgdExpected:
resourcegraphdefinition.kro.run/greeting created
NAME APIVERSION KIND STATE READY AGE
greeting v1alpha1 Greeting Active True 6s
Active and True: kro has validated and
accepted the RGD. What it made of it is shown by
-o wide:
kubectl get rgd greeting -o wide
kubectl get crd greetings.kro.runExpected:
NAME APIVERSION KIND STATE READY TOPOLOGICALORDER AGE
greeting v1alpha1 Greeting Active True ["config"] 6s
NAME SCOPE VERSIONS CREATED AT
greetings.kro.run Namespaced v1alpha1(storage) 2026-11-11T17:29:59Z
TOPOLOGICALORDERis the order in which kro creates the objects. With a single object it’s boring; from Day 4 on, it won’t be.- kro generated the CRD
greetings.kro.runitself. From now on, the API server knows the kindGreeting:
kubectl explain greeting.specExpected:
GROUP: kro.run
KIND: Greeting
VERSION: v1alpha1
FIELD: spec <Object>
DESCRIPTION:
<empty>
FIELDS:
message <string>
<no description>
Save your progress? Sign up for free
Drill 4: The First Instance
kubectl create namespace app
cat > hallo.yaml <<'EOF'
apiVersion: kro.run/v1alpha1
kind: Greeting
metadata:
name: hallo
namespace: app
spec:
message: Hallo kro
EOF
kubectl apply -f hallo.yaml
kubectl -n app get greetingExpected:
namespace/app created
greeting.kro.run/hallo created
NAME STATE READY AGE
hallo ACTIVE True 4s
To the team, Greeting looks like any other Kubernetes
kind. What was created behind it:
kubectl -n app get configmap hallo -o yamlExpected (shortened):
apiVersion: v1
data:
message: Hallo kro
kind: ConfigMap
metadata:
annotations:
internal.kro.run/apply-order: "1"
internal.kro.run/node-path: config
labels:
app.kubernetes.io/managed-by: kro
applyset.kubernetes.io/part-of: applyset-Iiz_0lx5c0WuukoyYjCPBF1CcU-I56PoX...
kro.run/instance-kind: Greeting
kro.run/instance-name: hallo
kro.run/instance-namespace: app
kro.run/kro-version: v0.10.0
kro.run/node-id: config
kro.run/owned: "true"
name: hallo
namespace: app
- The
kro.run/…labels say which instance and which node (node-id) the object belongs to. With them you’ll later find everything that belongs to an instance. applyset.kubernetes.io/part-of: kro keeps track of everything it has created via an ApplySet. If an object drops out of the RGD, kro knows what it has to clean up.
Save your progress? Sign up for free
Drill 5: kro Is Watching
Change the ConfigMap by hand:
kubectl -n app patch configmap hallo --type merge -p '{"data":{"message":"von Hand"}}'
kubectl -n app get configmap hallo -o jsonpath='{.data.message}{"\n"}'Expected:
configmap/hallo patched
Hallo kro
The change was gone again before you could look. kro watches the objects it has created and immediately resets any deviation. This is called drift correction.
Delete the ConfigMap:
kubectl -n app delete configmap hallo
kubectl -n app get configmap halloExpected:
configmap "hallo" deleted from app namespace
NAME DATA AGE
hallo 1 5s
It is back immediately. What the instance itself says about its state:
kubectl -n app get greeting hallo \
-o jsonpath='{range .status.conditions[*]}{.type}={.status}{"\n"}{end}'Expected:
InstanceManaged=True
GraphResolved=True
ResourcesReady=True
Ready=True
| Condition | Meaning |
|---|---|
InstanceManaged |
kro has set finalizer and labels |
GraphResolved |
All expressions could be evaluated |
ResourcesReady |
All objects are created and ready |
Ready |
All three above are True |
Save your progress? Sign up for free
Drill 6: What the Controller Says
kubectl -n kro-system logs deploy/kro | grep greetingExpected (shortened):
... INFO rgd-controller Creating CRD {"controller": "ResourceGraphDefinition",
..., "name": "greeting", ..., "name": "greetings.kro.run"}
... INFO rgd-controller Waiting for CRD to become ready {..., "name": "greeting",
...}
The logs are your first port of call when an RGD doesn’t become
Active.
Save your progress? Sign up for free
Checkpoint
kubectl get rgd greetingExpected: STATE Active, READY
True.
kubectl -n app get greeting halloExpected: STATE ACTIVE, READY
True.
kubectl -n app get configmap -l kro.run/instance-name=halloExpected:
NAME DATA AGE
hallo 1 3m
The label finds all objects of an instance, regardless of kind.
Challenge
Change the message of the instance hallo to “Servus kro”
without touching the ConfigMap. Then check the ConfigMap and finally
delete the instance. What happens to the ConfigMap?
Note: The instance is an object like any other;
kubectl patch or kubectl edit work.
Solution
kubectl -n app patch greeting hallo --type merge -p '{"spec":{"message":"Servus kro"}}'
kubectl -n app get configmap hallo -o jsonpath='{.data.message}{"\n"}'
kubectl -n app delete greeting hallo
kubectl -n app get configmap halloExpected:
greeting.kro.run/hallo patched
Servus kro
greeting.kro.run "hallo" deleted from app namespace
Error from server (NotFound): configmaps "hallo" not found
Together with the instance, kro deletes everything it created for it.
It removes the finalizer kro.run/finalizer on the instance
only once the ConfigMap is gone.
Save your progress? Sign up for free
Enterprise Note
- Pin the version: kro is still pre-1.0; the API is
called
v1alpha1. Pin the chart version and read the upgrade notes before every upgrade (Day 14). - One controller per cluster: kro runs as one Deployment with leader election. More replicas don’t make kro faster; they only ensure a quick failover when a pod fails.
unrestrictedis for the lab: In this mode kro may do everything, and anyone allowed to create an RGD can use it to create any object in the cluster. Day 12 shows the way out.