81b024edd6
Chicken-and-egg encontrado en smoke E2E DO (cluster 91c93c9c): - ns 'apps' tiene default-deny NetworkPolicy entregada por stack-deployer. - Job Alembic (pre-install hook) ejecuta InitContainer wait-for-secret que hace REST API call al kube-apiserver para chequear Secret. - La NetworkPolicy del chart (que permite DNS + apiserver egress) estaba como RESOURCE NORMAL, asi que solo se aplicaba en la FASE PRINCIPAL del helm install, DESPUES de los hooks. - Resultado: el InitContainer del hook quedaba bloqueado DNS por default-deny → curl timeout 240s → helm install fail → NetworkPolicy nunca se aplica. Fix: - Annotations 'helm.sh/hook: pre-install,pre-upgrade' + 'hook-weight: -8' + 'hook-delete-policy: before-hook-creation'. - Weight ordering: -10 AddonClaim → -8 NetworkPolicy → -7 Role/RoleBinding → -5 SA → 0 Job Alembic. - delete-policy: before-hook-creation (no hook-succeeded): el Deployment de la fase principal necesita la NP viva en runtime. - Comentario en la regla egress :443 aclarando que cubre kube-apiserver (10.43.0.1) + Harbor (sin necesidad de regla adicional). Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
89 lines
3.2 KiB
YAML
89 lines
3.2 KiB
YAML
{{- if .Values.networkPolicy.enabled }}
|
|
# NetworkPolicy mínima del shape: bajo deny-all-default del namespace, permite:
|
|
# - Ingress: rke2-ingress-nginx (web-backend público) + Prometheus (scrape de /metrics).
|
|
# - Egress: DNS, Harbor (pull en startup), ns-addons del tenant (BD).
|
|
# Nota: el cluster tenant usa rke2-ingress-nginx en kube-system, NO Traefik.
|
|
# Materializada como Helm hook (pre-install/pre-upgrade, weight "-8") para que
|
|
# este disponible ANTES de que el Job Alembic (hook weight 0) ejecute su
|
|
# InitContainer wait-for-secret — el InitContainer hace REST API call al
|
|
# kube-apiserver y necesita egress DNS + API permitidos. Sin esto, el chicken-
|
|
# and-egg con default-deny del namespace `apps` (entregado por stack-deployer)
|
|
# bloquea el hook → curl falla DNS → Job timeout → helm install fail → la
|
|
# NetworkPolicy nunca se aplica en la fase principal.
|
|
#
|
|
# Weight ordering (mas chico = antes):
|
|
# -10 AddonClaim -> -8 NetworkPolicy -> -7 Role/RoleBinding -> -5 SA -> 0 Job Alembic
|
|
#
|
|
# delete-policy: before-hook-creation (NO hook-succeeded): la fase principal del
|
|
# Deployment necesita la NetworkPolicy viva en runtime; solo se borra/recrea al
|
|
# inicio del proximo hook (idempotente entre upgrades).
|
|
apiVersion: networking.k8s.io/v1
|
|
kind: NetworkPolicy
|
|
metadata:
|
|
name: {{ .Release.Name }}
|
|
labels:
|
|
{{- include "app.labels" . | nindent 4 }}
|
|
annotations:
|
|
"helm.sh/hook": pre-install,pre-upgrade
|
|
"helm.sh/hook-weight": "-8"
|
|
"helm.sh/hook-delete-policy": before-hook-creation
|
|
spec:
|
|
podSelector:
|
|
matchLabels:
|
|
{{- include "app.selectorLabels" . | nindent 6 }}
|
|
policyTypes: [Ingress, Egress]
|
|
|
|
ingress:
|
|
# Tráfico HTTP desde rke2-ingress-nginx (Ingress estándar).
|
|
# El controller corre en kube-system por default en RKE2.
|
|
- from:
|
|
- namespaceSelector:
|
|
matchLabels:
|
|
kubernetes.io/metadata.name: {{ .Values.networkPolicy.ingressControllerNamespace | default "kube-system" }}
|
|
ports:
|
|
- protocol: TCP
|
|
port: 8000
|
|
# Scrape de métricas Prometheus
|
|
- from:
|
|
- namespaceSelector:
|
|
matchLabels:
|
|
kubernetes.io/metadata.name: cattle-monitoring-system
|
|
podSelector:
|
|
matchLabels:
|
|
app.kubernetes.io/name: prometheus
|
|
ports:
|
|
- protocol: TCP
|
|
port: 8000
|
|
|
|
egress:
|
|
# DNS
|
|
- to:
|
|
- namespaceSelector:
|
|
matchLabels:
|
|
kubernetes.io/metadata.name: kube-system
|
|
ports:
|
|
- protocol: UDP
|
|
port: 53
|
|
- protocol: TCP
|
|
port: 53
|
|
# ns-addons del tenant: MariaDB primary + pool
|
|
- to:
|
|
- namespaceSelector:
|
|
matchLabels:
|
|
kubernetes.io/metadata.name: {{ .Values.networkPolicy.addonsNamespace }}
|
|
ports:
|
|
- protocol: TCP
|
|
port: 3306
|
|
- protocol: TCP
|
|
port: 6033
|
|
# kube-apiserver (10.43.0.1:443) y Harbor pull (registro de imágenes).
|
|
# 0.0.0.0/0:443 cubre ambos. El InitContainer wait-for-secret del Job
|
|
# Alembic necesita egress al apiserver para hacer GET secret via REST API.
|
|
- to:
|
|
- ipBlock:
|
|
cidr: 0.0.0.0/0
|
|
ports:
|
|
- protocol: TCP
|
|
port: 443
|
|
{{- end }}
|