Day 9: Helm – Using Charts
Goal: Install a third-party chart from an OCI registry, customize it with your own values, upgrade it, roll it back and remove it again. · Duration: approx. 75 min · Prerequisite: Day 8
Briefing
- A chart is a package of manifest templates (Go
templates) and default values (
values.yaml). Charts live in repositories or, as today, as an OCI artifact in a container registry. helm installrenders the templates with your values and creates a release. Every change creates a new revision; you can go back to any of them.- Helm 4 (since late 2025) works without a server component like Helm
3, but uses server-side apply by default, waits via kstatus (“watcher”)
and has renamed some flags (
--atomicis now called--rollback-on-failure). - Demo chart: podinfo 6.17.0, a small Go web server, in its own
namespace
podinfo.
chart (oci://…/podinfo:6.17.0) ─┐
values.yaml in the chart ───────┼─► helm template ─► manifests ─► API server
-f podinfo-values.yaml, --set ──┘
release podinfo: revision 1, 2, 3 … as Secret sh.helm.release.v1.podinfo.vN
Drill 1: Helm and a Chart from the Registry
source ~/k8s-lab/setup/env.sh
mkdir -p ~/k8s-lab/tag09 && cd ~/k8s-lab/tag09
cp -R "$LAB/code/notes/." . && cp -R "$LAB/code/tag08/notes/." .
cp "$LAB/code/tag09/podinfo-values.yaml" .
helm version
helm show chart oci://ghcr.io/stefanprodan/charts/podinfo --version 6.17.0Expected (abridged):
version.BuildInfo{Version:"v4.3.1", GitCommit:"c0d08d0…", GitTreeState:"clean", …}
Pulled: ghcr.io/stefanprodan/charts/podinfo:6.17.0
Digest: sha256:258d0da5…
apiVersion: v1
appVersion: 6.17.0
description: Podinfo Helm chart for Kubernetes
home: https://github.com/stefanprodan/podinfo
kubeVersion: '>=1.23.0-0'
…
name: podinfo
…
version: 6.17.0
versionis the version of the chart,appVersionthat of the application in it. For podinfo both are the same; for most charts they aren’t.- OCI charts don’t need
helm repo add; the address starts withoci://.--versionpins – without it, Helm takes the newest version, and at some point your lab changes underneath you.
The state of Day 8 stays untouched: notes keeps running, and today you work next to it.
Save your progress? Sign up for free
Drill 2: Inspect and Render Values
C=oci://ghcr.io/stefanprodan/charts/podinfo
helm show values $C --version 6.17.0 2>/dev/null | wc -l
helm show values $C --version 6.17.0 2>/dev/null | grep -E -A3 '^(replicaCount|ui):'
cat podinfo-values.yamlExpected:
327
replicaCount: 1
logLevel: info
host: #0.0.0.0
backend: #http://backend-podinfo:9898/echo
--
ui:
color: "#34577c"
message: ""
logo: ""
# podinfo-values.yaml – eigene Werte für das Demo-Chart (Tag 9)
replicaCount: 2
ui:
message: "Hallo aus dem Trainingslager"
color: "#0f766e"
resources:
requests:
cpu: 20m
memory: 32Mi
limits:
memory: 64Mi
The 327 lines are the chart’s interface. You only override what you
need. helm template shows what comes out of it – without
cluster access:
helm template podinfo $C --version 6.17.0 -n podinfo 2>/dev/null \
| grep -E '^kind:|^ name:'
helm template podinfo $C --version 6.17.0 -n podinfo -f podinfo-values.yaml \
--set ui.message="Hallo von --set" 2>/dev/null \
| grep -E -A1 'replicas:|PODINFO_UI_(MESSAGE|COLOR)' | head -12Expected (abridged, suffixes will differ on your machine):
kind: Service
name: podinfo
kind: Deployment
name: podinfo
kind: Pod
name: podinfo-grpc-test-jkmji
kind: Pod
name: podinfo-jwt-test-bczoc
kind: Pod
name: podinfo-service-test-a3fjm
replicas: 2
strategy:
--
- name: PODINFO_UI_MESSAGE
value: "Hallo von --set"
- name: PODINFO_UI_COLOR
value: "#0f766e"
- Order of the sources:
values.yamlin the chart, then every-ffile, then every--set. Whatever comes later wins. - The three Pods are tests (
helm test), not workloads; Helm doesn’t create them oninstall. On Day 10 you write one yourself.
Save your progress? Sign up for free
Drill 3: Install
helm install podinfo $C --version 6.17.0 -n podinfo --create-namespace \
-f podinfo-values.yaml --wait
helm list
helm list -n podinfo
kubectl -n podinfo get deploy,podsExpected (abridged, names, IPs and times will differ on your machine):
Pulled: ghcr.io/stefanprodan/charts/podinfo:6.17.0
Digest: sha256:258d0da5…
NAME: podinfo
LAST DEPLOYED: Fri Oct 30 14:47:43 2026
NAMESPACE: podinfo
STATUS: deployed
REVISION: 1
DESCRIPTION: Install complete
NOTES:
1. Get the application URL by running these commands:
echo "Visit http://127.0.0.1:8080 to use your application"
kubectl -n podinfo port-forward deploy/podinfo 8080:9898
NAME NAMESPACE REVISION UPDATED STATUS CHART APP VERSION
NAME NAMESPACE REVISION UPDATED STATUS CHART APP VERSION
podinfo podinfo 1 2026-10-30 … deployed podinfo-6.17.0 6.17.0
NAME READY UP-TO-DATE AVAILABLE AGE
deployment.apps/podinfo 2/2 2 2 24s
NAME READY STATUS RESTARTS AGE
pod/podinfo-6d9f6b786d-ngl55 1/1 Running 0 24s
pod/podinfo-6d9f6b786d-pss8p 1/1 Running 0 24s
- Like
kubectl, Helm takes the namespace of the context. That is set tonotes– which is why the firsthelm listis empty. With Helm,-nbelongs to every command. --waitwaits until all objects are ready (default timeout 5 min). Without it,installreturns as soon as the objects have been created.NOTES:is rendered by the chart fromtemplates/NOTES.txt– the user manual.
The Service is ClusterIP. For a quick look, a port
forward is enough:
kubectl -n podinfo port-forward svc/podinfo 9898:9898 >/dev/null &
sleep 2
curl -s localhost:9898 | jq -c '{hostname, message, color, version}'
kill %1Expected (abridged):
{"hostname":"podinfo-6d9f6b786d-pss8p","message":"Hallo aus dem Trainingslager",
"color":"#0f766e","version":"6.17.0"}
port-forward to a Service picks one Pod and sticks with
it – it is no good for load balancing, but fine for diagnosis.
Save your progress? Sign up for free
Drill 4: What Helm Remembers
helm get values podinfo -n podinfo
helm get values podinfo -n podinfo --all | wc -l
helm get manifest podinfo -n podinfo | grep -E '^(# Source|kind)'
kubectl -n podinfo get secret -l owner=helm --show-labelsExpected (abridged):
USER-SUPPLIED VALUES:
replicaCount: 2
resources:
limits:
memory: 64Mi
requests:
cpu: 20m
memory: 32Mi
ui:
color: '#0f766e'
message: Hallo aus dem Trainingslager
187
# Source: podinfo/templates/service.yaml
kind: Service
# Source: podinfo/templates/deployment.yaml
kind: Deployment
NAME TYPE DATA AGE LABELS
sh.helm.release.v1.podinfo.v1 helm.sh/release.v1 1 34s modifiedAt=…,
name=podinfo,owner=helm,status=deployed,version=1
Every revision is stored as a Secret in the release’s namespace: chart, values and rendered manifests, compressed. That is why Helm needs no server – and why the namespace is part of the release.
Save your progress? Sign up for free
Drill 5: Upgrade and the Values Trap
You only want to change the message:
helm upgrade podinfo $C --version 6.17.0 -n podinfo \
--set ui.message="Version 2" --wait 2>/dev/null | grep -E '^(STATUS|REVISION)'
helm get values podinfo -n podinfo
kubectl -n podinfo get deploy podinfoExpected (abridged):
STATUS: deployed
REVISION: 2
USER-SUPPLIED VALUES:
ui:
message: Version 2
NAME READY UP-TO-DATE AVAILABLE AGE
podinfo 1/1 1 1 39s
One replica, no color, no resources: upgrade with
--set or -f starts from the chart defaults.
Your file from revision 1 is gone. Without any
--set/-f, on the other hand, Helm takes over
the old values. Rely on neither, and instead specify everything on every
upgrade:
helm upgrade podinfo $C --version 6.17.0 -n podinfo -f podinfo-values.yaml \
--set ui.message="Version 3" --wait 2>/dev/null | grep -E '^(STATUS|REVISION)'
kubectl -n podinfo get deploy podinfo
helm history podinfo -n podinfoExpected (abridged):
STATUS: deployed
REVISION: 3
NAME READY UP-TO-DATE AVAILABLE AGE
podinfo 2/2 2 2 55s
REVISION UPDATED STATUS CHART APP VERSION DESCRIPTION
1 Fri Oct 30 14:47:43 2026 superseded podinfo-6.17.0 6.17.0 Install complete
2 Fri Oct 30 14:48:07 2026 superseded podinfo-6.17.0 6.17.0 Upgrade complete
3 Fri Oct 30 14:48:24 2026 deployed podinfo-6.17.0 6.17.0 Upgrade complete
--reuse-values (old values plus new ones) and
--reset-then-reuse-values exist too. Still, a values file
in Git is the only source you can rely on.
Save your progress? Sign up for free
Drill 6: Rollback
helm get values podinfo -n podinfo --revision 1 | tail -2
helm rollback podinfo 1 -n podinfo --wait
helm history podinfo -n podinfo | tail -2
kubectl -n podinfo get rsExpected (abridged):
color: '#0f766e'
message: Hallo aus dem Trainingslager
Rollback was a success! Happy Helming!
3 Fri Oct 30 14:48:24 2026 superseded podinfo-6.17.0 6.17.0 Upgrade complete
4 Fri Oct 30 14:48:40 2026 deployed podinfo-6.17.0 6.17.0 Rollback to 1
NAME DESIRED CURRENT READY AGE
podinfo-6d9f6b786d 2 2 2 71s
podinfo-86f77b6996 0 0 0 30s
podinfo-bfdb7fbb7 0 0 0 47s
A rollback writes revision 1 as the new revision 4 – history only
moves forward. The Deployment is back on the Pod template of revision 1,
so the old ReplicaSet 6d9f6b786d scales up again.
helm rollback without a number goes back one revision. By
default, Helm keeps ten revisions (--history-max).
Save your progress? Sign up for free
Drill 7: Automatic Rollback with
--rollback-on-failure
An upgrade with an image tag that doesn’t exist:
helm upgrade podinfo $C --version 6.17.0 -n podinfo -f podinfo-values.yaml \
--set image.tag=0.0.0 --rollback-on-failure --timeout 60s
echo "Exit: $?"
helm history podinfo -n podinfo | tail -3
kubectl -n podinfo get events --field-selector reason=Failed \
-o custom-columns=MSG:.message | sort -u | head -3 | cut -c1-80Expected (abridged):
Pulled: ghcr.io/stefanprodan/charts/podinfo:6.17.0
Digest: sha256:258d0da5…
level=WARN msg="upgrade failed" name=podinfo error="resource Deployment/podinfo/podinfo
not ready. status: InProgress, message: Pending termination: 1\ncontext deadline
exceeded"
Error: UPGRADE FAILED: release podinfo failed, and has been rolled back due to
rollback-on-failure being set: resource Deployment/podinfo/podinfo not ready. …
context deadline exceeded
Exit: 1
4 Fri Oct 30 14:48:40 2026 superseded podinfo-6.17.0 6.17.0 Rollback to 1
5 Fri Oct 30 14:49:03 2026 failed podinfo-6.17.0 6.17.0 Upgrade "podi…
6 Fri Oct 30 14:50:05 2026 deployed podinfo-6.17.0 6.17.0 Rollback to 4
Error: ErrImagePull
Error: ImagePullBackOff
Failed to pull image "ghcr.io/stefanprodan/podinfo:0.0.0": rpc error: code = NotFoun
--rollback-on-failureturns on--wait, waits until--timeoutand, on an error, rolls back to the last successful revision. Oninstallit uninstalls instead.- Without the flag, revision 5 would stay
failed, and the Deployment would hang with one Pod inImagePullBackOff– and the nextupgradewould have to sort it out. - Helm 4 still knows
--atomic, but reportsFlag --atomic has been deprecated, use --rollback-on-failure instead(verified live).
In a pipeline, this is the standard combination:
--rollback-on-failure --timeout with a value that matches
the probes.
Save your progress? Sign up for free
Drill 8: Third-Party Charts and Pod Security
The namespace podinfo has no PSA label. kind sets no
cluster default, so privileged applies there – everything
is allowed. Would podinfo run under the rules of notes?
kubectl get ns podinfo --show-labels
kubectl label --dry-run=server --overwrite ns podinfo \
pod-security.kubernetes.io/enforce=baseline
kubectl label --dry-run=server --overwrite ns podinfo \
pod-security.kubernetes.io/enforce=restricted
kubectl -n podinfo exec deploy/podinfo -- idExpected (abridged):
NAME STATUS AGE LABELS
podinfo Active 24s kubernetes.io/metadata.name=podinfo,name=podinfo
namespace/podinfo labeled (server dry run)
namespace/podinfo labeled (server dry run)
Warning: existing pods in namespace "podinfo" violate the new PodSecurity enforce
level "restricted:latest"
Warning: podinfo-6d9f6b786d-24rcl (and 1 other pod): allowPrivilegeEscalation !=
false, unrestricted capabilities, runAsNonRoot != true, seccompProfile
uid=100(app) gid=101(app) groups=101(app)
baseline yes, restricted no – the same four
points as notes before Day 8, even though podinfo doesn’t run as root at
all. Many public charts are like this: they set no
securityContext, but offer values for it
(helm show values … | grep -i securitycontext). That is the
challenge.
Save your progress? Sign up for free
Checkpoint
helm list -n podinfoExpected: release podinfo, deployed,
revision 6.
NAME NAMESPACE REVISION UPDATED STATUS CHART APP VERSION
podinfo podinfo 6 2026-… deployed podinfo-6.17.0 6.17.0
helm history podinfo -n podinfoExpected: six revisions, including the failed upgrade from Drill 7.
REVISION UPDATED STATUS CHART APP VERSION DESCRIPTION
1 … superseded podinfo-6.17.0 6.17.0 Install complete
2 … superseded podinfo-6.17.0 6.17.0 Upgrade complete
3 … superseded podinfo-6.17.0 6.17.0 Upgrade complete
4 … superseded podinfo-6.17.0 6.17.0 Rollback to 1
5 … failed podinfo-6.17.0 6.17.0 Upgrade "podinfo" fa…
6 … deployed podinfo-6.17.0 6.17.0 Rollback to 4
helm get values podinfo -n podinfoExpected: your values from podinfo-values.yaml,
replicaCount: 2.
USER-SUPPLIED VALUES:
replicaCount: 2
resources:
limits:
memory: 64Mi
requests:
cpu: 20m
memory: 32Mi
ui:
color: '#0f766e'
message: Hallo aus dem Trainingslager
kubectl -n podinfo get deploy podinfoExpected: two podinfo Pods ready.
NAME READY UP-TO-DATE AVAILABLE AGE
podinfo 2/2 2 2 113s
helm list -n notesExpected: only the header line. notes keeps running without Helm.
NAME NAMESPACE REVISION UPDATED STATUS CHART APP VERSION
Challenge
In the namespace podinfo, set enforce and
warn to restricted and get podinfo running
again only through an additional values file
podinfo-restricted.yaml – without changing the chart. Tear
everything down afterwards.
Hint: The chart has the values
podSecurityContext and securityContext;
remember the trap from Day 8 (Drill 4).
Solution
cat > podinfo-restricted.yaml <<'EOF'
# podinfo-restricted.yaml – Challenge Day 9
podSecurityContext:
runAsNonRoot: true
runAsUser: 100 # user app in the image – name instead of UID, see Day 8
runAsGroup: 101
seccompProfile:
type: RuntimeDefault
securityContext:
allowPrivilegeEscalation: false
readOnlyRootFilesystem: true
capabilities:
drop: [ALL]
EOF
kubectl label ns podinfo pod-security.kubernetes.io/enforce=restricted \
pod-security.kubernetes.io/warn=restricted
helm upgrade podinfo $C --version 6.17.0 -n podinfo -f podinfo-values.yaml \
-f podinfo-restricted.yaml --wait 2>&1 | grep -E '^(STATUS|REVISION|Error)'
kubectl -n podinfo exec deploy/podinfo -- id
kubectl -n podinfo rollout restart deploy/podinfoExpected (abridged):
Warning: existing pods in namespace "podinfo" violate the new PodSecurity enforce
level "restricted:latest"
Warning: podinfo-6d9f6b786d-24rcl (and 1 other pod): allowPrivilegeEscalation != …
namespace/podinfo labeled
STATUS: deployed
REVISION: 7
uid=100(app) gid=101(app) groups=101(app)
deployment.apps/podinfo restarted
The rollout restart without a Warning:
proves it: new Pods satisfy restricted. Without
runAsUser, the upgrade hangs until the timeout and the Pods
show CreateContainerConfigError with
image has non-numeric user (app) (verified live) – the
release revision is then failed until a corrected upgrade
follows.
To wrap up the day:
helm uninstall podinfo -n podinfo
kubectl delete ns podinfo
helm list -AExpected:
release "podinfo" uninstalled
namespace "podinfo" deleted
NAME NAMESPACE REVISION UPDATED STATUS CHART APP VERSION
helm uninstall deletes all objects of the release and
its history, but never the namespace – not even if
--create-namespace created it.
Save your progress? Sign up for free
Enterprise Note
- Charts from the internet are third-party code with cluster
permissions: pin the version, read
helm templatebefore first use, mirror charts into your own registry. Signatures and provenance are covered in “Secure supply chains hands-on”. - Values per environment as files in Git
(
values-prod.yaml), never just--setin a shell history. GitOps tools like Argo CD or Flux render charts declaratively and detect drift. helm diff(plugin) shows what changes before every upgrade – mandatory in reviews.- Registry credentials (
helm registry login) belong in the pipeline’s secret store, not in the admins’ home directories. - Release Secrets contain all values in plain text (only compressed). Anyone who may read Secrets can also read passwords from values – tailor RBAC accordingly.