Day 7: Helm Provider – Postgres and notes as a Chart
Goal: Install Helm charts with OpenTofu: the Postgres operator CloudNativePG, a database as a cluster, and notes as its own chart – connected through the Secret the operator creates. · Duration: approx. 90 min · Prerequisite: Day 6
Briefing
- The Helm provider installs charts like
helm install, but with plan and state:helm_releasewith repository, chart, version, and values. - It’s best to pass values as
values = [yamlencode({ … })]: HCL maps become YAML, and variables and references go straight in. - CloudNativePG is an operator for Postgres in
Kubernetes. A
Clusterresource describes the database; the operator creates pods, volumes, services, and a Secret with the ready-made connection URL. - The operator brings its own resource types (CRDs). Anything that
uses such a type depends on the operator –
depends_onmakes that visible.
helm_release.cnpg (operator, namespace cnpg-system)
└─► helm_release.db (chart "cluster" → Cluster notes-db-cluster, Secret …-app)
└─► helm_release.notes (own chart, NOTES_DB_URL from Secret …-app/uri)
└─► http://notes-helm.<VM_IP>.sslip.io
Drill 1: The Operator
mkdir -p ~/tf-lab/tag07 && cd ~/tf-lab/tag07
cat > versions.tf <<'EOF'
terraform {
required_version = ">= 1.13.0"
required_providers {
helm = {
source = "hashicorp/helm"
version = "~> 3.3"
}
}
}
provider "helm" {
kubernetes = {
config_path = "/etc/rancher/k3s/k3s.yaml"
}
}
EOF
cat > operator.tf <<'EOF'
resource "helm_release" "cnpg" {
name = "cnpg"
repository = "https://cloudnative-pg.github.io/charts"
chart = "cloudnative-pg"
version = "0.29.1"
namespace = "cnpg-system"
create_namespace = true
wait = true
}
EOF
tofu init -no-color | grep -E 'Installed|initialized'
tofu apply -auto-approve -no-color | grep -E 'Apply complete'
kubectl -n cnpg-system get deployments
kubectl get crd -o name | grep -c 'postgresql.cnpg.io'Expected (age differs):
- Installed hashicorp/helm v3.3.0 (signed, key ID 0C0AF313E5FD9F80)
OpenTofu has been successfully initialized!
Apply complete! Resources: 1 added, 0 changed, 0 destroyed.
NAME READY UP-TO-DATE AVAILABLE AGE
cnpg-cloudnative-pg 1/1 1 1 40s
11
- In version 3 of the Helm provider,
kubernetesis an attribute with=, not a block – older examples on the web still showkubernetes { … }. wait = truewaits until all pods of the chart are ready. After that, the eleven new resource types exist, among themclusters.postgresql.cnpg.io.
Save your progress? Sign up for free
Drill 2: A Database
cat > db.tf <<'EOF'
resource "helm_release" "db" {
name = "notes-db"
repository = "https://cloudnative-pg.github.io/charts"
chart = "cluster"
version = "0.9.0"
namespace = "notes-helm"
create_namespace = true
values = [yamlencode({
type = "postgresql"
cluster = {
instances = 1
storage = { size = "1Gi" }
}
})]
depends_on = [helm_release.cnpg]
}
locals {
db_secret = "${helm_release.db.name}-cluster-app"
}
EOF
tofu apply -auto-approve -no-color | grep -E 'Apply complete'
kubectl -n notes-helm wait cluster/notes-db-cluster --for=condition=Ready --timeout=300s
kubectl -n notes-helm get secret notes-db-cluster-app -o jsonpath='{.data.uri}' \
| base64 -d; echoExpected (after one to two minutes; the password is random and longer):
Apply complete! Resources: 1 added, 0 changed, 0 destroyed.
cluster.postgresql.cnpg.io/notes-db-cluster condition met
postgresql://app:kbdPex8sZQv5…@notes-db-cluster-rw.notes-helm:5432/app
- The
clusterchart only creates theClusterresource and is done immediately; the actual work is then done by the operator.kubectl waitwaits for it. - The Secret
notes-db-cluster-appcontains user, password, and underurithe ready-made connection URL. The operator generated the password – it appears nowhere in the code and not in OpenTofu’s state. local.db_secretderives the name from the release. If someone changes the release’s name, the Secret follows.
Save your progress? Sign up for free
Drill 3: notes as Its Own Chart
The companion code contains a small chart for notes:
ls ~/tf-lab/code/charts/notes ~/tf-lab/code/charts/notes/templates
grep -A3 'secretKeyRef' ~/tf-lab/code/charts/notes/templates/deployment.yaml | head -4Expected:
/home/ubuntu/tf-lab/code/charts/notes:
Chart.yaml
templates
values.yaml
/home/ubuntu/tf-lab/code/charts/notes/templates:
deployment.yaml
ingress.yaml
secret.yaml
service.yaml
secretKeyRef:
name: {{ .Values.db.secretName }}
key: uri
{{- end }}
An init container runs notes-api migrate before every
start; both read the URL from the Secret whose name is passed in as a
value.
cat > notes.tf <<'EOF'
variable "vm_ip" {
type = string
}
resource "helm_release" "notes" {
name = "notes"
chart = "${pathexpand("~")}/tf-lab/code/charts/notes"
namespace = "notes-helm"
values = [yamlencode({
replicas = 1
db = { secretName = local.db_secret }
host = "notes-helm.${var.vm_ip}.sslip.io"
})]
}
output "url" {
value = "http://notes-helm.${var.vm_ip}.sslip.io"
}
EOF
export TF_VAR_vm_ip=$VM_IP
tofu apply -auto-approve -no-color | grep -E 'Apply complete'
U=$(tofu output -raw url)
curl -s -X POST -H "X-API-Key: $NOTES_API_KEY" -H 'Content-Type: application/json' \
-d '{"text":"in Postgres auf k3s"}' "$U/notes" | jq -r .text
kubectl -n notes-helm get podsExpected (pod names and age differ):
Apply complete! Resources: 1 added, 0 changed, 0 destroyed.
in Postgres auf k3s
NAME READY STATUS RESTARTS AGE
notes-5fbc5f7cf9-65kvp 1/1 Running 0 20s
notes-5fbc5f7cf9-fhs9z 1/1 Running 0 20s
notes-db-cluster-1 1/1 Running 0 2m
A chart from a local folder doesn’t need a repository.
pathexpand("~") turns ~ into the home
directory – OpenTofu itself doesn’t understand ~.
Save your progress? Sign up for free
Drill 4: Changing Values
sed -i 's/^ replicas = 1$/ replicas = 2/' notes.tf
tofu plan -no-color | grep -E '[-+] +"replicas"|Plan:'
tofu apply -auto-approve -no-color | grep -E 'Apply complete'
kubectl -n notes-helm get deployment notes -o jsonpath='{.status.readyReplicas}'; echo
kubectl -n notes-helm get secrets -l owner=helm,name=notes --no-headers | wc -lExpected:
- "replicas": 1
+ "replicas": 2
Plan: 0 to add, 1 to change, 0 to destroy.
Apply complete! Resources: 0 added, 1 changed, 0 destroyed.
2
2
The plan shows the difference in the YAML that
yamlencode generates – with quotes around every key. Helm
stores every version of a release in a Secret – after the upgrade there
are two. A rollback with helm rollback would bypass
OpenTofu; the right way back is the old value in the code and a new
apply.
Save your progress? Sign up for free
Drill 5: New Chart Version
The chart version in the code changes – OpenTofu sees it from the folder’s contents:
sed -i 's/^version: 1.0.0$/version: 1.1.0/' ~/tf-lab/code/charts/notes/Chart.yaml
tofu plan -no-color | grep -E '^ ~ version|Plan:'
tofu apply -auto-approve -no-color | grep -E 'Apply complete'
tofu state show helm_release.notes | grep -E '^ +(version|status) +='Expected:
~ version = "1.0.0" -> "1.1.0"
Plan: 0 to add, 1 to change, 0 to destroy.
Apply complete! Resources: 0 added, 1 changed, 0 destroyed.
status = "deployed"
version = "1.1.0"
For a chart from a repository, the version is in the code as
version = "…"; a new chart then only arrives when you
change it.
Save your progress? Sign up for free
Checkpoint
curl -s "$(tofu output -raw url)/notes" | jq -r '.[].text'Expected:
in Postgres auf k3s
tofu state listExpected:
helm_release.cnpg
helm_release.db
helm_release.notes
kubectl -n notes-helm get cluster notes-db-cluster -o jsonpath='{.status.phase}'; echoExpected:
Cluster in healthy state
Challenge
The values for notes should come from a file
notes-values.yaml.tftpl so that someone without HCL
knowledge can read them too. Afterwards, the plan must not show any
change.
Hint:
values = [templatefile("…", { … })].
Solution
cat > notes-values.yaml.tftpl <<'EOF'
replicas: 2
db:
secretName: ${db_secret}
host: notes-helm.${vm_ip}.sslip.io
EOF
python3 - <<'PY'
import re
s = open("notes.tf").read()
s = re.sub(r"values = \[yamlencode\(\{.*?\}\)\]",
'values = [templatefile("${path.module}/notes-values.yaml.tftpl", {\n'
' db_secret = local.db_secret\n vm_ip = var.vm_ip\n })]',
s, flags=re.S)
open("notes.tf", "w").write(s)
PY
tofu fmt >/dev/null
gen=$(kubectl -n notes-helm get deployment notes -o jsonpath='{.metadata.generation}')
tofu plan -no-color | grep -E 'Plan:|No changes'
tofu apply -auto-approve -no-color | grep -E 'Apply complete'
neu=$(kubectl -n notes-helm get deployment notes -o jsonpath='{.metadata.generation}')
[[ $gen == "$neu" ]] && echo "Deployment unverändert"Expected:
Plan: 0 to add, 1 to change, 0 to destroy.
Apply complete! Resources: 0 added, 1 changed, 0 destroyed.
Deployment unverändert
OpenTofu compares the text of the values:
yamlencode writes "replicas": 2, the template
replicas: 2. That’s why it plans a change. Helm renders the
same manifests from both – the Deployment stays untouched; only a new
release version is created. Watch the indentation in templates: YAML
doesn’t forgive wrong spacing.
Save your progress? Sign up for free
Enterprise Note
- Pin versions: Charts without
versioninstall whatever is newest – anapplyafter a few weeks can then bring a major upgrade. - CRDs and operators belong in a separate
configuration that runs before the applications. Resources via
kubernetes_manifestneed the CRDs already at plan time; in the same configuration as the operator, the first run fails. - OpenTofu or GitOps? Many teams install platform components (operators, ingress, monitoring) with OpenTofu and applications with Argo CD or Flux.