prole/scripts/k3d-knoe-up.sh

#!/usr/bin/env bash
# scripts/k3d-knoe-up.sh
# Bring up the k3d-knoe local dev cluster: CNPG operator + single-replica DB + KDC + schema.
#
# Usage:
#   ./scripts/k3d-knoe-up.sh
#
# Prerequisites:
#   k3d, kubectl, docker (running)
#
# What it does:
#   1. Creates k3d cluster "k3d-knoe" (if not present)
#   2. Installs CNPG operator (v1.29.0)
#   3. Applies knoe-db CNPG Cluster (single-replica, local-path storage)
#   4. Waits for CNPG cluster to be healthy
#   5. Applies KDC manifests (configmap, pvc, deployment, service)
#   6. Runs etc/init_knoe_auth.sh schema --mode k3d to seed knoe.* tables

set -euo pipefail

SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
REPO_ROOT="$(cd "$SCRIPT_DIR/.." && pwd)"
K8S_KNOE="$REPO_ROOT/k8s/knoe"

CLUSTER_NAME="${K3D_CLUSTER_NAME:-k3d-knoe}"
CTX="k3d-${CLUSTER_NAME}"
DB_NS="knoe-db-0"
SYS_NS="knoe-system"
CNPG_VERSION="${CNPG_OPERATOR_VERSION:-1.29.0}"
CNPG_NS="cnpg-system"

log()  { echo "[k3d-knoe-up] $*"; }
die()  { echo "[k3d-knoe-up] ERROR: $*" >&2; exit 1; }

kube() { kubectl --context="$CTX" "$@"; }

# ── 1. Cluster ────────────────────────────────────────────────────────────────

if k3d cluster list 2>/dev/null | awk '{print $1}' | grep -qx "${CLUSTER_NAME#k3d-}"; then
  log "k3d cluster '${CLUSTER_NAME#k3d-}' already exists — skipping create."
else
  log "Creating k3d cluster '${CLUSTER_NAME#k3d-}'..."
  k3d cluster create "${CLUSTER_NAME#k3d-}" \
    --agents 0 \
    --k3s-arg "--disable=traefik@server:0" \
    --wait
  log "Cluster created."
fi

# Ensure kubeconfig is merged
k3d kubeconfig merge "${CLUSTER_NAME#k3d-}" --kubeconfig-merge-default >/dev/null 2>&1 || true

# ── 2. CNPG operator ──────────────────────────────────────────────────────────

log "Installing CNPG operator v${CNPG_VERSION}..."
kube apply --server-side -f \
  "https://raw.githubusercontent.com/cloudnative-pg/cloudnative-pg/release-${CNPG_VERSION%.*}/releases/cnpg-${CNPG_VERSION}.yaml" \
  2>/dev/null || \
kube apply -f \
  "https://raw.githubusercontent.com/cloudnative-pg/cloudnative-pg/release-${CNPG_VERSION%.*}/releases/cnpg-${CNPG_VERSION}.yaml"

log "Waiting for CNPG operator to be ready..."
kube -n "$CNPG_NS" rollout status deployment/cnpg-controller-manager --timeout=120s

# ── 3. knoe-db namespace + secret + cluster ───────────────────────────────────

kube get namespace "$DB_NS" >/dev/null 2>&1 || kube create namespace "$DB_NS"

# Create the knoe-db-user secret (CNPG bootstrap owner secret)
if ! kube -n "$DB_NS" get secret knoe-db-user >/dev/null 2>&1; then
  log "Creating knoe-db-user secret..."
  kube -n "$DB_NS" create secret generic knoe-db-user \
    --from-literal=username=knoe \
    --from-literal=password=knoe-dev-password
fi

# Render the knoe-db manifest for k3d:
#   - instances: 1  (single-replica)
#   - remove synology-iscsi storageClassName + selector blocks
#   - remove affinity with kubernetes.io/hostname topology
#   - remove plugins block (no barman in k3d)
#   - remove certificates block (no pre-issued TLS secrets in k3d)
#   - remove env block referencing AWS/Garage (not needed locally)
#   - substitute image registry placeholder with a public postgres image
log "Rendering knoe-db manifest for k3d..."
RENDERED=$(python3 - <<'PYEOF'
import re, sys

with open("k8s/knoe/knoe-db.yaml") as f:
    text = f.read()

# instances: 3 -> 1
text = re.sub(r'(\binstances:\s*)3\b', r'\g<1>1', text)

# Remove plugins block (barman)
text = re.sub(r'\n  plugins:\n(  - .*\n(    .*\n)*)+', '\n', text)

# Remove certificates block
text = re.sub(r'\n  certificates:\n(    .*\n)+', '\n', text)

# Remove env block (AWS/Garage refs)
text = re.sub(r'\n  env:\n(    - .*\n(      .*\n)*)+', '\n', text)

# Remove walStorage block (not needed for single-replica dev)
text = re.sub(r'\n  walStorage:\n(    .*\n)+', '\n', text)

# Replace synology-iscsi storageClassName
text = re.sub(r'storageClassName: synology-iscsi', '', text)

# Remove selector block with synology.storage labels
text = re.sub(r'\n      selector:\n        matchLabels:\n          synology\.storage/role: \w+\n', '\n', text)

# Remove affinity block (contains kubernetes.io/hostname)
text = re.sub(r'\n  affinity:\n(    .*\n)+', '\n', text)

# Replace image placeholder with standard postgres (CNPG uses its own image)
text = re.sub(
    r'imageName: \$\{KNOE_IMAGE_REGISTRY\}/knoe-db:\$\{KNOE_DB_IMAGE_TAG\}',
    'imageName: ghcr.io/cloudnative-pg/postgresql:17',
    text
)

# Remove maxSyncReplicas (irrelevant for single instance)
text = re.sub(r'\n  maxSyncReplicas: \d+\n', '\n', text)

print(text)
PYEOF
)

echo "$RENDERED" | kube apply -f -

log "Waiting for CNPG cluster 'knoe-db' to be healthy (up to 5 min)..."
for i in $(seq 1 60); do
  phase=$(kube -n "$DB_NS" get cluster knoe-db \
    -o jsonpath='{.status.phase}' 2>/dev/null || echo "")
  if [[ "$phase" == "Cluster in healthy state" ]]; then
    log "CNPG cluster is healthy."
    break
  fi
  if [[ $i -eq 60 ]]; then
    die "CNPG cluster did not become healthy in time. Phase: $phase"
  fi
  echo "  [$i/60] phase='$phase' — waiting 5s..."
  sleep 5
done

# ── 4. KDC ────────────────────────────────────────────────────────────────────

kube get namespace "$SYS_NS" >/dev/null 2>&1 || kube create namespace "$SYS_NS"

log "Provisioning KDC (realm KNOE.LOCAL)..."
KNOE_MODE=k3d APP_CLUSTER_KUBECONTEXT="$CTX" \
  "$REPO_ROOT/etc/init_knoe_auth.sh" initialize --mode k3d --context "$CTX"

log ""
log "=== k3d-knoe is up ==="
log "  CNPG cluster: kubectl --context=$CTX -n $DB_NS get cluster knoe-db"
log "  KDC:          kubectl --context=$CTX -n $SYS_NS get deploy knoe-kdc"
log ""
log "Next: make k3d-knoe-pf   (opens port-forwards)"
log "Then: export KRB5_CONFIG=\$PWD/etc/krb5.local.conf"
log "      mvn -pl authority spring-boot:run"