Helm — Advanced Patterns

Beyond basics — patterns that trip up experienced engineers and aren't well-captured in training data.

Docs: https://helm.sh/docs/ | Sprig: https://masterminds.github.io/sprig/

Library Charts

Shared template logic across multiple charts. Not deployable on their own.

# Chart.yaml
apiVersion: v2
name: mylib
type: library   # Not "application"
version: 0.1.0

Merge Utility Pattern

The canonical way to let consumers override library-defined resources:

# mylib/templates/_util.tpl
{{- define "mylib.util.merge" -}}
{{- $top := first . -}}
{{- $overrides := fromYaml (include (index . 1) $top) | default (dict ) -}}
{{- $base := fromYaml (include (index . 2) $top) | default (dict ) -}}
{{- toYaml (merge $overrides $base) -}}
{{- end -}}

Consumer chart uses it to override specific fields while inheriting defaults:

# mychart/templates/deployment.yaml
{{- include "mylib.util.merge" (list . "mychart.deployment.overrides" "mylib.deployment") }}

# mychart/templates/_overrides.tpl
{{- define "mychart.deployment.overrides" -}}
spec:
  replicas: {{ .Values.replicaCount }}
{{- end -}}

Shared Label/Annotation Templates

# mylib/templates/_labels.tpl
{{- define "mylib.labels" -}}
app.kubernetes.io/name: {{ .Chart.Name }}
app.kubernetes.io/instance: {{ .Release.Name }}
app.kubernetes.io/version: {{ .Chart.AppVersion | quote }}
app.kubernetes.io/managed-by: {{ .Release.Service }}
helm.sh/chart: {{ printf "%s-%s" .Chart.Name .Chart.Version | replace "+" "_" | trunc 63 }}
{{- end -}}

{{- define "mylib.selectorLabels" -}}
app.kubernetes.io/name: {{ .Chart.Name }}
app.kubernetes.io/instance: {{ .Release.Name }}
{{- end -}}

Values Schema Validation

values.schema.json

{
  "$schema": "https://json-schema.org/draft-07/schema#",
  "type": "object",
  "required": ["image", "service"],
  "properties": {
    "replicaCount": {
      "type": "integer",
      "minimum": 1,
      "maximum": 50
    },
    "image": {
      "type": "object",
      "required": ["repository"],
      "properties": {
        "repository": { "type": "string", "minLength": 1 },
        "tag": { "type": "string" },
        "pullPolicy": { "type": "string", "enum": ["Always", "IfNotPresent", "Never"] }
      },
      "additionalProperties": false
    },
    "resources": {
      "type": "object",
      "properties": {
        "requests": { "$ref": "#/definitions/resourceBlock" },
        "limits": { "$ref": "#/definitions/resourceBlock" }
      }
    }
  },
  "definitions": {
    "resourceBlock": {
      "type": "object",
      "properties": {
        "cpu": { "type": "string", "pattern": "^[0-9]+m?$" },
        "memory": { "type": "string", "pattern": "^[0-9]+(Mi|Gi)$" }
      }
    }
  }
}

Generate from values files:

helm plugin install https://github.com/losisin/helm-values-schema-json && helm schema -input values.yaml

The

lookup

Function

Query live cluster state during rendering. Returns empty dict in

--dry-run=client

--dry-run=server

# Only create Secret if it doesn't already exist (preserve generated passwords across upgrades)
{{- if not (lookup "v1" "Secret" .Release.Namespace "myapp-credentials") }}
apiVersion: v1
kind: Secret
metadata:
  name: myapp-credentials
type: Opaque
data:
  password: {{ randAlphaNum 32 | b64enc | quote }}
{{- end }}

# Iterate over existing namespaces to deploy NetworkPolicies
{{- range (lookup "v1" "Namespace" "" "").items }}
{{- if hasKey .metadata.labels "app.kubernetes.io/managed-by" }}
# ... generate NetworkPolicy for {{ .metadata.name }}
{{- end }}
{{- end }}

Gotcha:

lookup

{{- if .Release.IsInstall }}

--dry-run=server

Template Pitfalls & Solutions

Nil Map Panics

# WRONG — panics if .Values.ingress is nil
{{- if .Values.ingress.enabled }}

# CORRECT — safe navigation
{{- if ((.Values.ingress).enabled) }}

# Also works with dig (Sprig)
{{- if (dig "ingress" "enabled" false .Values) }}

Whitespace Control

# Produces clean YAML without extra blank lines
{{- if .Values.podAnnotations }}
      annotations:
        {{- toYaml .Values.podAnnotations | nindent 8 }}
{{- end }}

-

{{-

-}}

Type Coercion

# Values from --set are always strings. "true" != true
# Use quote + explicit type handling
{{- if eq (toString .Values.debug) "true" }}

# Or in schema, coerce with: {"type": ["boolean", "string"]}

Range Scope

# WRONG — $ is the global scope, . changes inside range
{{- range .Values.hosts }}
  host: {{ . }}
  service: {{ $.Values.serviceName }}   # Use $ to access root
{{- end }}

CRD Management

Two approaches, both with tradeoffs:

crds/

Directory (Helm 3)

• CRDs in

  crds/

  are installed once on first install
• Never updated or deleted by Helm (by design)
• Simple but no lifecycle management

Hook-Based CRDs

# templates/crds.yaml
{{- if .Values.installCRDs }}
apiVersion: apiextensions.k8s.io/v1
kind: CustomResourceDefinition
metadata:
  annotations:
    "helm.sh/hook": pre-install,pre-upgrade
    "helm.sh/hook-weight": "-5"
  name: myresources.example.com
spec:
  # ...
{{- end }}

Gives upgrade control but risks deleting CRDs (and all CRs) on uninstall. Add

"helm.sh/resource-policy": keep

Post-Renderers

Mutate rendered manifests before apply. Useful for injecting sidecars, labels, or policy annotations without modifying chart templates.

⚠️ Helm 4 breaking change: Post-renderers must be installed Helm plugins — passing an executable path (

--post-renderer ./script.sh

--post-renderer <plugin-name>

Helm 3 (executable path — still works in Helm 3.x):

helm upgrade --install myapp ./mychart --post-renderer ./inject-labels.sh

#!/bin/bash
# inject-labels.sh — add labels via kustomize
cat > /tmp/kustomization.yaml <<EOF
apiVersion: kustomize.config.k8s.io/v1beta1
kind: Kustomization
resources:
  - rendered.yaml
commonLabels:
  company.io/cost-center: platform
EOF
cat <&0 > /tmp/rendered.yaml
kustomize build /tmp

Helm 4 (plugin name — required in Helm 4, also works in Helm 3.14+):

# Install a post-renderer plugin, then reference it by name
helm plugin install https://github.com/example/helm-post-renderer-kustomize
helm upgrade --install myapp ./mychart --post-renderer kustomize

See Helm 4 plugin system (HIP-0026) and example plugins for WebAssembly-based post-renderer plugins.

Multi-Tenant / Subchart Patterns

Global Values

# Parent values.yaml
global:
  imageRegistry: 123456789.dkr.ecr.us-east-1.amazonaws.com
  storageClass: longhorn

# Accessible in subcharts as .Values.global.imageRegistry

Conditional Subcharts

# Chart.yaml
dependencies:
  - name: postgresql
    version: "15.x"
    repository: "oci://registry-1.docker.io/bitnamicharts"
    condition: postgresql.enabled
    import-values:
      - child: primary.service
        parent: database

import-values

postgresql.primary.service.port

Per-Environment Overrides

# Layered values: base → environment → secrets
helm upgrade --install myapp ./mychart \
  -f values.yaml \
  -f values-production.yaml \
  -f secrets-production.yaml.dec \
  --set image.tag=${GIT_SHA}

Secrets Handling

Charts should never contain secrets. Integration patterns:

# External Secrets Operator — reference in templates
apiVersion: external-secrets.io/v1beta1
kind: ExternalSecret
metadata:
  name: {{ include "myapp.fullname" . }}-db
spec:
  refreshInterval: 1h
  secretStoreRef:
    name: aws-secrets-manager
    kind: ClusterSecretStore
  target:
    name: {{ include "myapp.fullname" . }}-db
  data:
    - secretKey: password
      remoteRef:
        key: {{ .Values.externalSecrets.dbPasswordKey }}

# SOPS-encrypted values (decrypt before helm)
# values-secrets.yaml.enc → decrypt with sops
helm secrets upgrade myapp ./mychart -f values.yaml -f values-secrets.yaml

Testing with helm-unittest

# tests/deployment_test.yaml
suite: deployment
templates:
  - templates/deployment.yaml
tests:
  - it: should not render when disabled
    set:
      enabled: false
    asserts:
      - hasDocuments:
          count: 0

  - it: should set resource limits from values
    set:
      resources:
        limits:
          memory: 1Gi
    asserts:
      - equal:
          path: spec.template.spec.containers[0].resources.limits.memory
          value: 1Gi

  - it: should fail without required values
    set:
      image.repository: ""
    asserts:
      - failedTemplate: {}

helm plugin install https://github.com/helm-unittest/helm-unittest
helm unittest ./mychart --color

Helm 4 CLI Flag Renames

Helm 4 (released November 2025) renamed two common flags. The old names still work but emit deprecation warnings — update scripts and CI pipelines:

--atomic

--rollback-on-failure

--force

--force-replace

# Helm 4 — use new flag names
helm upgrade --install myapp ./mychart --rollback-on-failure --timeout 10m
helm upgrade myapp ./mychart --force-replace

Debugging

# Render one template with debug output
helm template myapp ./mychart -s templates/deployment.yaml --debug 2>&1

# Diff what would change (requires helm-diff plugin)
helm diff upgrade myapp ./mychart -f values-production.yaml

# Validate against cluster without applying
helm upgrade --install myapp ./mychart --dry-run=server --debug

# Check computed values after all overlays
helm get values myapp -n production -a  # -a = all (computed)

Common Advanced Errors

nil pointer evaluating interface {}

dig

((.Values.x).y)

schema validation failed

helm lint

crds/

lookup

--dry-run=server

{{- if not .Release.IsInstall }}

postgresql.auth.password

References

• troubleshooting.md

  — Install and upgrade failures, template rendering, release management, and repository issues

Cross-References

• argocd — GitOps deployment of Helm releases via ArgoCD Applications
• crossplane — Helm provider for deploying charts as Crossplane managed resources
• external-secrets — Inject secrets into Helm releases without embedding them in values
• cert-manager — TLS certificate management commonly deployed via Helm
• kueue — Job queueing — Helm charts for ML workloads often need Kueue integration
• operator-sdk — Building operators that use Helm charts as the reconciliation engine
• github-actions — CI/CD for Helm chart linting, packaging, and publishing
• docker-buildx — Build container images referenced by Helm charts
• prometheus-grafana — kube-prometheus-stack deployed via Helm
• gpu-operator — GPU Operator deployed and configured via Helm