Skip to content
Hands-on IT IT Drills

Appendix A: Cheat Sheet

The most important commands and patterns from Day 0 to Day 20, ordered by topic. The days are in brackets. You replace placeholders in angle brackets. Values such as lab-key-123 and lab-zot-123 are lab values. All commands run on prod; the registry is zot at localhost:30500 (HTTP, hence --plain-http). Inline comments in the code blocks are kept as in the labs (German); the note below each block translates them.

Terrain and environment (Day 0)

# on prod
export KUBECONFIG=$HOME/.kube/config
helm version --short
helm env | grep -E 'NAMESPACE|KUBECONTEXT'
helm list --short --kube-context helm-lab -n default
  • Without -n, the context’s namespace applies – for Helm too.
  • In scripts, always set --kube-context and -n.

Lifecycle (Day 1, 13)

# on prod
helm repo add zot https://zotregistry.dev/helm-charts && helm repo update
helm search repo zot/zot --versions | head -3
helm show values zot/zot --version 0.1.134 | grep -A3 '^service:'
helm pull zot/zot --version 0.1.134 --untar
helm install <rel> <chart> -n <ns> --create-namespace -f werte.yaml --wait
helm upgrade --install <rel> <chart> -n <ns> -f werte.yaml --wait --timeout 5m \
  --rollback-on-failure --history-max 5 --description "Ticket LAB-1"
helm rollback <rel> [<rev>] -n <ns> --wait --description "zurück, Ticket LAB-2"
helm uninstall <rel> -n <ns> [--keep-history]

werte.yaml is your values file; zurück means “back”.

Inspection – the first minute (Day 1, 13, 19)

# on prod
helm list -A                                   # Helm 4: jeder Status
helm list -A --failed --pending                # nach aktueller Revision
helm status <rel> -n <ns>                      # inkl. Ressourcen und Pods
helm history <rel> -n <ns> --show-rollback-revision
helm get values <rel> -n <ns> [--all] [--revision N]
helm get manifest <rel> -n <ns> [--revision N]
helm get hooks <rel> -n <ns>
helm get notes <rel> -n <ns>
helm get metadata <rel> -n <ns>                # APPLY_METHOD, Version
kubectl -n <ns> get events --field-selector reason=Failed \
  -o custom-columns=MELDUNG:.message | tail -1

Comments: Helm 4 lists every status; --failed/--pending filter by the current revision; status includes resources and pods. MELDUNG means “message”.

Read a release Secret (Day 13):

# on prod
kubectl -n <ns> get secret -l owner=helm,name=<rel> \
  --sort-by=.metadata.creationTimestamp
kubectl -n <ns> get secret sh.helm.release.v1.<rel>.v<N> -o jsonpath='{.data.release}' \
  | base64 -d | base64 -d | gzip -d | jq -c '.config.global, .info.status'

Values and precedence (Day 3)

values.yaml < -f a.yaml < -f b.yaml < --set (rightmost wins).

# on prod
helm install x ./notes --dry-run=client -o json --set a=1 | jq .config
--set-string env.NOTES_PAGE_SIZE=25      # String, Schema verlangt ihn
--set 'hosts={a.lab,b.lab}'              # Liste
--set-json 'resources={"limits":{"memory":"64Mi"}}'
--set-file motd=motd.txt
--set key=null                           # Standard löschen

Comments: --set-string forces a string because the schema requires one; braces build a list; key=null deletes a default.

Upgrade with Default values Your own values
nothing new those of the last deployed revision
-f/--set new only the ones given
--reuse-values old (computed) old + new
--reset-then-reuse-values new old + new
--reset-values new only the ones given

Lists are replaced, maps are merged – define environment variables as a map.

Template patterns (Day 4–8)

{{/* rendern: helm template <rel> ./notes -s templates/<datei>.yaml */}}
metadata:
  labels:
    {{- include "notes.labels" . | nindent 4 }}       # nindent, nie indent
spec:
  replicas: {{ hasKey .Values "replicaCount" | ternary .Values.replicaCount 2 }}
  template:
    metadata:
      annotations:
        checksum/config: {{ include (print $.Template.BasePath "/config.yaml") .
          | sha256sum }}
    spec:
      {{- with .Values.nodeSelector }}
      nodeSelector:
        {{- toYaml . | nindent 8 }}
      {{- end }}
      containers:
        - image: {{ printf "%s:%s" .Values.image.repository
            (.Values.image.tag | default .Chart.AppVersion) | quote }}
          env:
            - name: NOTES_API_KEY
              valueFrom:
                secretKeyRef:
                  name: {{ required "apiKey.existingSecret fehlt"
                    .Values.apiKey.existingSecret }}
                  key: api-key

Comments: the first line shows how to render a single file (rendern = render, <datei> = file); always nindent, never indent. fehlt means “is missing”.

{{- range .Values.ingress.hosts }}
- host: {{ tpl . $ | quote }}                          # $ = Wurzel im range
  service: {{ include "notes.fullname" $ }}
{{- end }}
{{- if .Capabilities.APIVersions.Has "gateway.networking.k8s.io/v1" }} … {{- end }}
uid: {{ dig "metadata" "uid" "leer" (lookup "v1" "Namespace" "" "kube-system") }}
sekunden: {{ durationSeconds .Values.shutdownTimeout }}  # neu in Helm 4.3
data:
  {{- (.Files.Glob "config/*.conf").AsConfig | nindent 2 }}

Comments: $ is the root inside range; durationSeconds is new in Helm 4.3. leer means “empty”, sekunden “seconds”.

Trap Remedy Day
default swallows 0/false hasKey + ternary, kindIs 5
{{- … -}} glues lines together trim on one side only, render 5
. inside range/with $ for the root 5
template … \| nindent include 6
quote does not rescue 1.20 quote it in the values file 4
lookup empty with template, Argo CD template must handle the empty case 7, 18
randAlphaNum in a Secret lookup + reuse, or existingSecret 7
.Files returns "" path, .helmignore, not in templates/ 8

Schema (Day 8, 9)

{"$schema": "http://json-schema.org/draft-07/schema#", "type": "object",
 "additionalProperties": false, "required": ["apiKey"],
 "properties": {"replicaCount": {"type": ["integer", "null"], "minimum": 0},
                "postgres": {"type": "object"}, "global": {"type": "object"}}}

Applies to install, upgrade, lint, template. Allow subchart keys and global, otherwise you get “additional properties … not allowed”.

Subcharts and library charts (Day 9, 10)

# Chart.yaml
dependencies:
  - name: postgres
    version: "~0.1.0"
    repository: oci://localhost:30500/charts
    condition: postgres.enabled
  - name: plattform
    version: "~0.1.0"
    repository: oci://localhost:30500/charts
# on prod
helm dependency update notes --plain-http    # löst Bereich auf, schreibt Chart.lock
helm dependency build notes --plain-http     # baut genau nach Chart.lock
helm dependency list notes

Comments: update resolves the range and writes Chart.lock; build builds exactly from Chart.lock.

Subchart values go under its name (postgres.enabled), shared values under global. type: library cannot be installed.

Hooks and tests (Day 11)

metadata:
  annotations:
    "helm.sh/hook": post-install,pre-upgrade
    "helm.sh/hook-weight": "0"
    "helm.sh/hook-delete-policy": before-hook-creation
# on prod
helm test <rel> -n <ns> --logs
helm upgrade <rel> <chart> -n <ns> --no-hooks

Hooks are not part of the release (they survive uninstall); pre-install runs before all objects – even before Secrets and subcharts. Migrations: post-install,pre-upgrade.

CRDs (Day 12)

# on prod
helm template x ./chart --include-crds
helm install x ./chart --skip-crds
helm show crds ./chart | kubectl apply --server-side --field-manager=helm -f -

crds/: only at installation, never updated, never deleted. A separate CRD chart with "helm.sh/resource-policy": keep.

Caution: Deleting a CRD deletes all objects of that type in the cluster.

Rolling out safely (Day 13)

Flag Effect
without --wait waits only for hooks (hookOnly)
--wait / --wait=watcher waits for readiness (kstatus)
--wait=legacy Helm 3 behaviour
--wait-for-jobs waits for Jobs (only with --wait)
--rollback-on-failure rollback or uninstall on failure, sets --wait
--dry-run=server renders with the cluster and lookup, stores nothing, does not send the objects to the API for validation (Day 4)
--history-max N revisions to keep (default 10)

Server-Side Apply and Helm 4 (Day 14)

# on prod
helm get metadata <rel> -n <ns> | grep APPLY_METHOD
helm upgrade <rel> <chart> -n <ns> --server-side=true      # true|false|auto
helm upgrade <rel> <chart> -n <ns> --force-conflicts
kubectl -n <ns> get deploy <name> --show-managed-fields -o json \
  | jq -r '.metadata.managedFields[]
    | [.manager, .operation, .subresource // "-"] | @tsv'
helm diff upgrade <rel> <chart> -n <ns> --three-way-merge --detailed-exitcode
Helm 3 Helm 4
--atomic --rollback-on-failure
--force --force-replace (not with SSA)
helm list -a helm list (every status), filters --deployed etc.
--post-renderer ./skript --post-renderer <plugin> (type postrenderer/v1)
three-way merge, drift silently overwritten SSA, conflict naming the field manager

OCI registry (Day 15)

# on prod
echo lab-zot-123 | helm registry login localhost:30500 -u lab-zot-user \
  --password-stdin --plain-http                                   # nur der Host
helm package notes
helm push notes-0.12.1.tgz oci://localhost:30500/charts --plain-http  # stdout!
helm show chart oci://localhost:30500/charts/notes --version 0.12.1 --plain-http
helm install <rel> oci://localhost:30500/charts/notes@sha256:<digest> --plain-http
curl -s -u lab-zot-user:lab-zot-123 localhost:30500/v2/charts/notes/tags/list

Comments: registry login takes only the host; helm push writes its output (including the digest) to stdout.

+ becomes _ in the tag; tags can be overwritten, the digest cannot.

Signing and secrets (Day 16)

# on prod
gpg --export-secret-keys '<Name>' > secring.gpg
helm package notes --sign --key '<Name>' --keyring secring.gpg
helm verify notes-0.12.1.tgz --keyring pubring.gpg
helm install <rel> ./notes-0.12.1.tgz --verify --keyring pubring.gpg
cosign signing-config create --out offline.json          # ohne Rekor
cosign sign --key cosign.key -y --signing-config offline.json --allow-http-registry \
  localhost:30500/charts/notes@<digest>
cosign verify --key cosign.pub --allow-http-registry --insecure-ignore-tlog \
  localhost:30500/charts/notes:0.12.1
sops encrypt -i secrets.dev.yaml
helm upgrade <rel> ./notes -f secrets://secrets.dev.yaml

Comment: ohne Rekor means “without Rekor” (no transparency log).

Anything that ends up in Helm as a value is stored in plain text in the release Secret. Better: existingSecret and a Secret from outside (ESO, Sealed Secrets, CSI).

Checking without a cluster (Day 17)

# on prod
set -o pipefail
helm lint notes --strict --with-subcharts -f values/prod.yaml
for env in dev staging prod; do
  helm template notes ./notes -f values/$env.yaml \
    | kubeconform -strict -summary -kubernetes-version 1.37.0
done
helm template notes ./notes -f values/staging.yaml --skip-tests --no-hooks \
  | kubectl apply --dry-run=server --server-side -n <staging> -f -
trivy config --severity HIGH,CRITICAL --exit-code 1 rendering.yaml
pluto detect rendering.yaml --target-versions k8s=<version>
pluto detect-helm
helm unittest notes

CI and GitOps (Day 18)

# on prod
ct lint                       # liest ct.yaml, verlangt neue Version bei Änderung
ct install --namespace <ns>   # installiert, helm test, deinstalliert
flux get sources oci -n <ns>
flux get helmreleases -n <ns>
kubectl kustomize --enable-helm .

Comments: ct lint reads ct.yaml and demands a new version when a chart changes; ct install installs, runs helm test and uninstalls.

Flux Argo CD Kustomize helmCharts
helm list yes empty empty
lookup cluster empty empty
Hooks Helm hooks sync hooks normal objects

Troubleshooting (Day 19)

# on prod
helm template <rel> ./chart -f werte.yaml --debug        # Schicht Rendern
helm template <rel> ./chart | kubectl apply --dry-run=server -f -   # Schicht API
helm rollback <rel> <letzte-deployed> -n <ns> --wait     # pending-upgrade lösen
kubectl delete sts <name> --cascade=orphan               # volumeClaimTemplates ändern
helm upgrade <rel> <chart> --take-ownership --force-conflicts

Comments: line 1 checks the rendering layer, line 2 the API layer; rolling back to the last deployed revision (<letzte-deployed>) resolves pending-upgrade; orphan-deleting the StatefulSet lets you change volumeClaimTemplates.

Never without thinking

  • kubectl delete secret -l owner=helm – deletes Helm’s memory.
  • --force-replace, --cascade=orphan, kubectl delete crd – back up first.
  • --reuse-values in pipelines – swallows new defaults.
  • Secrets via --set – stored in plain text in every revision.
  • Upgrades without --wait – broken revisions become the “last good” one.
  • --force-conflicts across the board in pipelines – hides drift.
  • Installing by tag instead of digest, without signature verification.

Error patterns

Message Cause Day
execution error at (notes/templates/deployment.yaml:20:20) required/fail in the template 7, 19
YAML parse error … mapping values are not allowed indentation, indent/dash 5, 20
can't evaluate field Values in type string . instead of $ in range 5, 20
additional properties '…' not allowed schema, typo, subchart key 8, 9, 20
library charts are not installable type: library 10
no matches for kind CRD missing, or in templates/ of the same chart 12
another operation … is in progress orphaned pending-* revision 19, 20
release has no 0 version rollback without a good revision 19
field is immutable selector, volumeClaimTemplates, Job 6, 19, 20
invalid ownership metadata object belongs to no release or another one 12, 19
conflict with "…" using … another field manager 14, 19
cannot use server-side apply and force replace together --force-replace with SSA 19
unable to build kubernetes objects from current release manifest removed API in the release 19
context deadline exceeded image, probe, hook, PVC 13, 19, 20
chart version not ok. Needs a version bump! ct lint without a new version 18
no signatures found chart not signed 16, 20
sha256 sum does not match package changed after signing 16
invalid registry registry login with a path or oci:// 15
improper constraint _ instead of + in --version 15