feat: AddonClaim + InitContainer wait-for-secret en chart helm
Some checks failed
__APP_NAME__ — build & deploy / Calidad + build + push (push) Failing after 1m19s
__APP_NAME__ — build & deploy / Helm upgrade --install (push) Has been skipped

Cierra GOLDEN-PATH-ADDONCLAIM-MISSING (P0 hands-off pipeline).

Cambios:
- helm/APP_NAME/templates/addonclaim.yaml (NEW): hook pre-install/pre-upgrade
  weight -10, SIN delete-policy (sobrevive upgrades, evita disparar finalizer
  cleanup-logical). Gated por .Values.database.enabled (default true).
- helm/APP_NAME/templates/role-db-secret-reader.yaml (NEW): Role + RoleBinding
  weight -7 que permite al SA leer el Secret <release>-db-creds. Lo consume
  el InitContainer wait-for-secret del Job Alembic.
- helm/APP_NAME/templates/job-alembic-upgrade.yaml: gated por
  database.enabled, anade InitContainer wait-for-secret (48x5s=240s) que
  espera a que addon-provisioner publique el Secret tras reconciliar el
  AddonClaim. Absorbe la latencia del @kopf.timer (RECONCILE_INTERVAL 300s
  + initial_delay 60s) sin tocar el operator.
- helm/APP_NAME/values.yaml: bloque database completo (enabled, addonType,
  databaseName, user.permissions, deletionPolicy Retain). Opt-out con
  database.enabled=false + alembic.enabled=false.
- .gitea/workflows/APP_NAME-build.yaml: helm timeout 5m -> 7m (margen para
  reconcile addon-provisioner + StatefulSet ready + DDL Job + Alembic).
- claim-mariadb.yaml (DELETED): obsoleto - estaba en la raiz del template
  y nunca se renderizaba por helm. Reemplazado por addonclaim.yaml dentro
  del chart helm.
- README.md: documenta nueva tabla de hook weights, opt-out database, y
  trade-off del orphan AddonClaim post-uninstall (cleanup explicito).

Decision arquitectural: Opcion D modificada (Opus 4.6 consult).

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
This commit is contained in:
ilopez
2026-05-25 07:56:42 -05:00
parent 1f2eb2f908
commit 5a215ed293
7 changed files with 208 additions and 43 deletions

View File

@@ -0,0 +1,39 @@
{{- if .Values.database.enabled }}
# AddonClaim — declara la intencion de la app de tener una BD (default MariaDB).
# Lo materializa addon-provisioner (CRD idp.coralware.cloud/v1alpha1):
# - reconcilia este claim cada RECONCILE_INTERVAL (default 300s) por TenantProfile.
# - crea BD logica + user + grant via Job DDL efimero en el cluster tenant.
# - publica el Secret `{{ .Values.database.secretName }}` en este ns (`apps`)
# con DATABASE_URL / DATABASE_HOST / DATABASE_PORT / DATABASE_USER /
# DATABASE_PASSWORD / DATABASE_NAME (contrato §2.3 del OpenSpec).
#
# Ordering con los hooks del chart (clave para que Alembic encuentre el Secret):
# weight -10 (este AddonClaim)
# -> weight -5 (ServiceAccount)
# -> weight 0 (Job Alembic — espera al Secret via InitContainer wait-for-secret)
#
# delete-policy: NO se declara intencionalmente.
# - Asi el AddonClaim sobrevive entre `helm upgrade` (re-apply idempotente).
# - Evita disparar el finalizer `cleanup-logical` en cada upgrade (que dropearia
# el user y regeneraria el password, causando ventana de downtime).
# - Trade-off documentado: `helm uninstall` deja el AddonClaim orfano y MariaDB
# sigue aprovisionada. Es coherente con `deletionPolicy: Retain` (default):
# borrar BD productiva debe ser decision explicita, no side-effect del uninstall.
# Cleanup explicito: `kubectl delete addonclaim {{ .Release.Name }}-db -n apps`.
apiVersion: idp.coralware.cloud/v1alpha1
kind: AddonClaim
metadata:
name: {{ .Release.Name }}-db
labels:
{{- include "app.labels" . | nindent 4 }}
annotations:
"helm.sh/hook": pre-install,pre-upgrade
"helm.sh/hook-weight": "-10"
spec:
addonType: {{ .Values.database.addonType | default "mariadb" | quote }}
databaseName: {{ .Values.database.databaseName | default (printf "%s_db" (.Release.Name | replace "-" "_")) | quote }}
user:
permissions: {{ .Values.database.user.permissions | default "rw" | quote }}
secretName: {{ .Values.database.secretName | quote }}
deletionPolicy: {{ .Values.database.deletionPolicy | default "Retain" | quote }}
{{- end }}

View File

@@ -1,8 +1,17 @@
{{- if .Values.alembic.enabled }}
{{- if and .Values.alembic.enabled .Values.database.enabled }}
# Job pre-upgrade que corre Alembic antes de rolar el Deployment.
# - backoffLimit: 0 si falla, NO reintenta; Helm aborta el upgrade.
# - hook-delete-policy hook-succeeded solo se borra al éxito; en fallo queda
# visible para `kubectl logs job/...` y diagnóstico.
# - backoffLimit: 0 -> si falla, NO reintenta; Helm aborta el upgrade.
# - hook-delete-policy hook-succeeded -> solo se borra al exito; en fallo queda
# visible para `kubectl logs job/...` y diagnostico.
#
# InitContainer `wait-for-secret`:
# - El AddonClaim (hook weight -10) ya fue aplicado, pero addon-provisioner es
# timer-based (RECONCILE_INTERVAL 300s, initial_delay 60s) y materializa el
# Secret `{{ .Values.database.secretName }}` de forma asincrona.
# - El InitContainer hace busy-wait hasta que el Secret existe — absorbe la
# latencia del reconcile dentro del pod, sin tocar el operator ni
# bajar RECONCILE_INTERVAL global.
# - Loop: 48 iteraciones x 5s = 240s — encaja dentro del helm timeout 7m.
apiVersion: batch/v1
kind: Job
metadata:
@@ -28,6 +37,42 @@ spec:
securityContext:
runAsNonRoot: true
runAsUser: 1001
initContainers:
- name: wait-for-secret
image: bitnami/kubectl:1.30
command:
- sh
- -c
- |
set -eu
# Espera hasta que addon-provisioner publique el Secret tras
# reconciliar el AddonClaim (hook weight -10). 48 x 5s = 240s.
SECRET="{{ .Values.database.secretName }}"
i=1
while [ "$i" -le 48 ]; do
if kubectl get secret "$SECRET" -n "$NAMESPACE" >/dev/null 2>&1; then
echo "Secret $SECRET encontrado en intento $i"
exit 0
fi
echo "[$i/48] Secret $SECRET aun no publicado, esperando 5s..."
sleep 5
i=$((i + 1))
done
echo "ERROR: Secret $SECRET no aparecio tras 240s — addon-provisioner stuck?"
exit 1
env:
- name: NAMESPACE
valueFrom:
fieldRef:
fieldPath: metadata.namespace
resources:
requests: {cpu: 10m, memory: 32Mi}
limits: {cpu: 100m, memory: 64Mi}
securityContext:
allowPrivilegeEscalation: false
readOnlyRootFilesystem: true
capabilities:
drop: ["ALL"]
containers:
- name: alembic
image: "{{ .Values.image.repository }}:{{ .Values.image.tag }}"
@@ -37,7 +82,7 @@ spec:
name: {{ .Values.database.secretName }}
env:
# Si el Secret expone MIGRATION_DATABASE_URL, lo usa; si no, env.py
# cae a DATABASE_URL automáticamente (ver settings.migration_url).
# cae a DATABASE_URL automaticamente (ver settings.migration_url).
- name: DATABASE_URL
valueFrom:
secretKeyRef:

View File

@@ -0,0 +1,50 @@
{{- if .Values.database.enabled }}
# Role + RoleBinding que permiten al SA de la app leer el Secret `<release>-db-creds`
# en el ns `apps`. Lo consume el InitContainer `wait-for-secret` del Job Alembic
# (hook pre-install/pre-upgrade weight 0) que hace busy-wait hasta que
# addon-provisioner publica el Secret tras reconciliar el AddonClaim.
#
# Hook weight "-7" — entre AddonClaim (-10) y SA (-5):
# - Garantiza que RBAC exista antes de que el SA del hook -5 se materialice
# y antes de que el Job -0 intente leer el Secret.
# - delete-policy: before-hook-creation para que sea idempotente entre upgrades
# sin dejar Roles huerfanos.
#
# Scope minimo: solo lectura (get/watch) del Secret `<release>-db-creds`, no
# secrets cluster-wide ni del ns `addons`. El root Secret de MariaDB vive en
# `addons` y nunca debe ser accesible desde `apps` (ADR-052 §6.1).
apiVersion: rbac.authorization.k8s.io/v1
kind: Role
metadata:
name: {{ .Release.Name }}-db-secret-reader
labels:
{{- include "app.labels" . | nindent 4 }}
annotations:
"helm.sh/hook": pre-install,pre-upgrade
"helm.sh/hook-weight": "-7"
"helm.sh/hook-delete-policy": before-hook-creation
rules:
- apiGroups: [""]
resources: ["secrets"]
resourceNames: [{{ .Values.database.secretName | quote }}]
verbs: ["get", "watch", "list"]
---
apiVersion: rbac.authorization.k8s.io/v1
kind: RoleBinding
metadata:
name: {{ .Release.Name }}-db-secret-reader
labels:
{{- include "app.labels" . | nindent 4 }}
annotations:
"helm.sh/hook": pre-install,pre-upgrade
"helm.sh/hook-weight": "-7"
"helm.sh/hook-delete-policy": before-hook-creation
subjects:
- kind: ServiceAccount
name: {{ .Release.Name }}
namespace: {{ .Release.Namespace }}
roleRef:
apiGroup: rbac.authorization.k8s.io
kind: Role
name: {{ .Release.Name }}-db-secret-reader
{{- end }}