Deploy postgres-operator v0.3.0-alpha.19 to keiailab EKS cluster

[eightynine01]

Context

postgres-operator v0.3.0-alpha.19 has been validated on a local kind cluster with successful e2e results. This issue tracks deployment to the actual keiailab EKS cluster (force-msa-dev) to begin the CNPG replacement process.

Pre-deployment Checklist

GitHub repo: https://github.com/keiailab/postgres-operator (source of truth)

Recent PRs merged (v0.3.0-alpha.19):

• PR fix(olm): OLM v1 compliance + dead code removal (3900 lines) #124: OLM v1 bundle compliance + dead code removal (3900 lines)
• PR docs: CNPG gap analysis — feature comparison + 8-sprint migration roadmap #125: CNPG gap analysis documentation (en/ko/ja/zh)
• PR feat(controller): config hot-reload via pg_reload_conf() [Sprint S2] #126: postgresql.conf hot-reload via pg_reload_conf()
• PR feat(backup): WAL archiving + version bump v0.3.0-alpha.19 [Sprint S3] #127: WAL archiving via pgbackrest archive-push + version bump
• PR fix(helm): add imagecatalogs RBAC — unblock reconciliation #128: RBAC fix — imagecatalogs ClusterRole permissions

Kind cluster e2e results (2026-05-25):

• Helm install: ✅ SUCCESS
• Operator Pod: 1/1 Running ✅
• CRDs registered: 10/10 ✅
• PostgresCluster CR: Phase=Ready ✅
• PostgreSQL 17.9: SELECT version() confirmed ✅
• RBAC bug found and fixed during testing (PR fix(helm): add imagecatalogs RBAC — unblock reconciliation #128) ✅

Deployment Steps

1. Build and push container image

# On a machine with Docker + ghcr.io push access
docker build -t ghcr.io/keiailab/postgres-operator:0.3.0-alpha.19 .
docker push ghcr.io/keiailab/postgres-operator:0.3.0-alpha.19

2. Deploy operator via Helm

# Switch to EKS context
kubectl config use-context arn:aws:eks:us-east-1:650557269383:cluster/force-msa-dev

# Create namespace
kubectl create namespace postgres-operator

# Install via Helm (from GitHub repo checkout)
helm upgrade --install postgres-operator charts/postgres-operator \
  --namespace postgres-operator \
  --set image.tag=0.3.0-alpha.19 \
  --wait --timeout 120s

3. Verify operator

kubectl get pods -n postgres-operator
kubectl logs -n postgres-operator deployment/postgres-operator-controller-manager --tail=20
# Should see: "Starting workers" for all 6 controllers, no RBAC errors

4. Create test PostgresCluster

cat <<EOF | kubectl apply -f -
apiVersion: postgres.keiailab.io/v1alpha1
kind: PostgresCluster
metadata:
  name: e2e-test
  namespace: default
spec:
  postgresVersion: "17"
  shardingMode: none
  shards:
    initialCount: 1
    replicas: 0
    storage:
      size: 10Gi
EOF

5. Verify PostgreSQL

# Wait for Ready phase
kubectl get postgrescluster e2e-test -w

# Connect and verify
kubectl exec e2e-test-shard-0-0 -- psql -U postgres -tAc "SELECT version();"

6. Begin CNPG comparison

# Check current CNPG clusters
kubectl get clusters.postgresql.cnpg.io --all-namespaces

# Plan migration: create equivalent PostgresCluster CRs
# for each CNPG Cluster, test side-by-side

Gap Analysis Reference

See docs/gap-analysis/GAP-ANALYSIS.md in the repo for full CNPG comparison.

Remaining P0 gaps before CNPG replacement:

• P0 chore(ci): GH Actions 폐기 + 로컬 4 계층 게이트 (RFC 0002, ADR 0009) #2: PITR from object store (Sprint S4)
• P0 feat(security): 데이터플레인 NetworkPolicy 3개 활성화 (P0-3, ADR 0006) #5: Backup retention cleanup (Sprint S5)

Already implemented:

• P0 feat: P0-1/P0-5 + 19 권장 거버넌스 등록 (Bitnami + PGO 교차검증) #1: WAL archiving (PR feat(backup): WAL archiving + version bump v0.3.0-alpha.19 [Sprint S3] #127) ✅
• P0 feat(controller): 데이터플레인 PodSecurityContext defaults (P0-2, ADR 0006) #3: TLS Phase 3 (already in codebase) ✅
• P0 test(controller): cascade delete 회귀 차단 envtest (P0-4, ADR 0008) #4: Config hot-reload (PR feat(controller): config hot-reload via pg_reload_conf() [Sprint S2] #126) ✅
• P1 docs(tasks): P0 6/6 완료 마킹 + 8 PR commit hash 매핑 #9: pg_hba.conf reload (PR feat(controller): config hot-reload via pg_reload_conf() [Sprint S2] #126) ✅

Post-deployment

After successful EKS deployment:

1. Run the full smoke test: bash hack/smoke.sh (adapted for EKS)
2. Create a ScheduledBackup CR to test backup scheduling
3. Create a Pooler CR to test PgBouncer integration
4. Monitor operator metrics via Prometheus
5. Begin Sprint S4 (PITR) implementation

Notes

• AWS access: Container image must be pushed from a CI/CD pipeline or authorized workstation (not from development machines)
• CNPG coexistence: postgres-operator uses postgres.keiailab.io API group — no conflict with CNPG's postgresql.cnpg.io
• Rollback: helm uninstall postgres-operator -n postgres-operator removes all operator resources; CRDs persist for safety

---

GitLab 이관 · 원본 keiailab/upstream/postgres-operator#1 · 작성 @phil · 2026-05-25 · 라벨: deployment, operator, priority::high · 이관 2026-06-03