Day 7: Conditions with includeWhen
Goal: Optional objects: a NetworkPolicy protects Postgres on request, and a PodDisruptionBudget appears on its own as soon as there is more than one replica. · Duration: approx. 40 min · Prerequisite: Day 6
Briefing
includeWhenis a list of conditions on an object. kro creates the object only if all of them aretrue. For OR, you write||within one condition.- kro re-evaluates the conditions on every reconcile. If one becomes
false, kro removes the object. - If an object drops out, all objects that depend on it drop out with it.
- Conditions on
schema.specare stable. Conditions on the status of other objects can flip back and forth and then keep creating and deleting objects (“flip-flop”).readyWhenis the better tool for that.
| Object | Condition |
|---|---|
dbPolicy (NetworkPolicy) |
${schema.spec.networkPolicy} |
budget (PodDisruptionBudget) |
${schema.spec.replicas > 1} |
Drill 1: Switches in the Schema
mkdir -p ~/kro-lab/tag07
cd ~/kro-lab/tag07
cp ../tag06/notesapp-rgd.yaml .Add a switch below database::
database:
storage: string | default="1Gi"
networkPolicy: boolean | default=false
Save your progress? Sign up for free
Drill 2: Two Optional Objects
Append to the end of the file (after the object
service):
- id: dbPolicy
includeWhen:
- ${schema.spec.networkPolicy}
template:
apiVersion: networking.k8s.io/v1
kind: NetworkPolicy
metadata:
name: ${database.metadata.name}
spec:
podSelector:
matchLabels: ${database.spec.selector.matchLabels}
policyTypes:
- Ingress
ingress:
- from:
- podSelector:
matchLabels: ${deployment.spec.selector.matchLabels}
ports:
- port: 5432
- id: budget
includeWhen:
- ${schema.spec.replicas > 1}
template:
apiVersion: policy/v1
kind: PodDisruptionBudget
metadata:
name: ${schema.metadata.name}
spec:
minAvailable: 1
selector:
matchLabels: ${deployment.spec.selector.matchLabels}
- The NetworkPolicy applies to the Postgres pods and only lets notes-api pods through on port 5432. The labels come from the selectors, not as text, so they always match.
- A PodDisruptionBudget with
minAvailable: 1and only one replica would block everykubectl drain. That’s why it only appears from two replicas on.
kubectl apply -f notesapp-rgd.yaml
kubectl get rgd notesapp
kubectl get rgd notesapp -o jsonpath='{.status.topologicalOrder}{"\n"}'Expected:
resourcegraphdefinition.kro.run/notesapp configured
NAME APIVERSION KIND STATE READY AGE
notesapp v1alpha1 NotesApp Active True 18m
["secret","dbSecret","dbData","database","dbService","deployment","service",
"dbPolicy","budget"]
Both new objects are in the order, but kro hasn’t created either of them yet:
kubectl -n app get networkpolicy,pdbExpected:
No resources found in app namespace.
Save your progress? Sign up for free
Drill 3: Open
Without a NetworkPolicy, every pod in the cluster can reach the
database. A test pod with nc (netcat) shows it:
kubectl -n app run probe --rm -i --restart=Never --image=busybox:1.38 \
-- nc -zv -w 2 notes-db 5432Expected (shortened):
notes-db (10.43.143.246:5432) open
pod "probe" deleted from app namespace
-zonly checks whether the port answers, without sending data.-w 2waits at most two seconds.--rm -istarts the pod, shows its output and deletes it afterwards.
Save your progress? Sign up for free
Drill 4: Closed
Turn on the NetworkPolicy and give the app two replicas:
kubectl -n app patch notesapp notes --type merge \
-p '{"spec":{"networkPolicy":true,"replicas":2}}'
kubectl -n app get networkpolicy,pdbExpected:
notesapp.lab.example.com/notes patched
NAME POD-SELECTOR AGE
networkpolicy.networking.k8s.io/notes-db app=notes-db 12s
NAME MIN AVAILABLE MAX UNAVAILABLE ALLOWED DISRUPTIONS AGE
poddisruptionbudget.policy/notes 1 N/A 1 12s
kubectl -n app run probe --rm -i --restart=Never --image=busybox:1.38 \
-- nc -zv -w 2 notes-db 5432Expected (shortened):
pod "probe" deleted from app namespace
pod app/probe terminated (Error)
No more open: after two seconds, nc gives
up. k3s enforces NetworkPolicies with its built-in policy controller.
notes-api is still allowed through:
curl -s localhost:30080/notesExpected:
[{"id":"5953a797-216a-45be-a2f9-fe1e5782a1f4","text":"bleibt erhalten"}]
Save your progress? Sign up for free
Drill 5: Removing
kubectl -n app patch notesapp notes --type merge \
-p '{"spec":{"networkPolicy":false,"replicas":1}}'
kubectl -n app get networkpolicy,pdbExpected:
notesapp.lab.example.com/notes patched
No resources found in app namespace.
Both conditions are false, and kro has deleted both
objects. The ApplySet from Day 1 tells kro which objects it has created.
Whatever no longer belongs in the graph gets thrown out.
Save your progress? Sign up for free
Checkpoint
kubectl get rgd notesappExpected: STATE Active.
kubectl -n app patch notesapp notes --type merge -p '{"spec":{"networkPolicy":true}}'
kubectl -n app get networkpolicyExpected:
notesapp.lab.example.com/notes patched
NAME POD-SELECTOR AGE
notes-db app=notes-db 3s
The NetworkPolicy stays turned on from now on.
diff notesapp-rgd.yaml ~/kro-lab/code/loesung/tag07/notesapp-rgd.yamlExpected: no output.
Challenge
In namespaces whose name starts with prod, the
NetworkPolicy should always apply, no matter what the
team sets. What is the condition?
Note: schema.metadata.namespace and the
string function startsWith.
Solution
includeWhen:
- ${schema.spec.networkPolicy || schema.metadata.namespace.startsWith("prod")}
One condition with || instead of two entries, because
multiple entries are combined with AND. You don’t need to apply this;
the fleet on Day 10 uses the switch.
Save your progress? Sign up for free
Enterprise Note
- Secure by default: In a real platform,
networkPolicywould default totrue, or there would be no switch at all. Anyone who needs an exception asks for it. - Policies don’t belong only in the RGD: Kyverno or ValidatingAdmissionPolicies enforce rules for all objects, including those not created via kro. See “Kyverno & NetworkPolicies hands-on”.
- Careful when removing: If an object drops out via
includeWhen, kro really deletes it. For volumes or databases, the deletion policy from Day 6 therefore belongs with it.