Update deploy-lab.sh

2026-07-18 03:39:35 +00:00
parent b2c6e28068
commit fc9dfb9374
1 changed files with 196 additions and 46 deletions
@@ -1,5 +1,5 @@
 #!/usr/bin/env bash
-# Laboratorio estable v4.4.3: ZimaOS + Docker + k3d + Argo CD + Multi-Repo + Governance + Monitoring Recovery + Runner
+# Laboratorio estable v4.4.6.1: ZimaOS + Docker + k3d + Argo CD + Multi-Repo + Governance + Monitoring + CI/CD
 #
 # Modos:
 #   bootstrap          Crea el clúster si no existe; si existe, lo inicia y reconcilia.
@@ -10,6 +10,7 @@
 #   gitops             Actualiza el repo y aplica root-apps-registry.
 #   monitoring-diagnose Diagnostica kube-prometheus-stack/Grafana.
 #   monitoring-recover  Finaliza una operación obsoleta y resincroniza Grafana.
+#   postdeploy-secrets  Regenera y valida credenciales de Gitea Actions.
 #   runner             Adopta, crea o recupera el Gitea Actions runner.
 #   install-autostart  Instala servicio/timer systemd, orden de montajes y autorreparación.
 #   reset              Reconstrucción limpia; requiere confirmación explícita.
@@ -52,6 +53,10 @@ EXPECTED_HOST_IP="${EXPECTED_HOST_IP:-192.168.68.61}"
 ALLOW_IP_CHANGE="${ALLOW_IP_CHANGE:-false}"
 INSTALL_HEADLAMP_RBAC="${INSTALL_HEADLAMP_RBAC:-true}"
 APPLY_POSTDEPLOY_RBAC="${APPLY_POSTDEPLOY_RBAC:-true}"
+POSTDEPLOY_SERVICE_ACCOUNT="${POSTDEPLOY_SERVICE_ACCOUNT:-gitea-postdeploy-validator}"
+POSTDEPLOY_TOKEN_SECRET="${POSTDEPLOY_TOKEN_SECRET:-gitea-postdeploy-validator-token}"
+POSTDEPLOY_NAMESPACE="${POSTDEPLOY_NAMESPACE:-ecommerce}"
+POSTDEPLOY_TOKEN_WAIT_SECONDS="${POSTDEPLOY_TOKEN_WAIT_SECONDS:-90}"
 WAIT_SECONDS="${WAIT_SECONDS:-240}"
 ARGOCD_ROLLOUT_TIMEOUT="${ARGOCD_ROLLOUT_TIMEOUT:-420}"
 AGENT_RECOVERY_WAIT="${AGENT_RECOVERY_WAIT:-75}"
@@ -869,20 +874,24 @@ apply_postdeploy_rbac() {

  log "Aplicando RBAC de solo lectura para validación postdeploy"

-  kubectl create namespace ecommerce --dry-run=client -o yaml | kubectl apply -f -
+  kubectl create namespace "$POSTDEPLOY_NAMESPACE" \
+    --dry-run=client \
+    -o yaml |
+    kubectl apply -f -

-  cat <<'YAML' | kubectl apply -f -
+  cat <<YAML | kubectl apply -f -
 apiVersion: v1
 kind: ServiceAccount
 metadata:
-  name: gitea-postdeploy-validator
-  namespace: ecommerce
+  name: ${POSTDEPLOY_SERVICE_ACCOUNT}
+  namespace: ${POSTDEPLOY_NAMESPACE}
+automountServiceAccountToken: false
 ---
 apiVersion: rbac.authorization.k8s.io/v1
 kind: Role
 metadata:
-  name: gitea-postdeploy-validator
-  namespace: ecommerce
+  name: ${POSTDEPLOY_SERVICE_ACCOUNT}
+  namespace: ${POSTDEPLOY_NAMESPACE}
 rules:
  - apiGroups: ["apps"]
    resources: ["deployments", "replicasets"]
@@ -894,51 +903,114 @@ rules:
 apiVersion: rbac.authorization.k8s.io/v1
 kind: RoleBinding
 metadata:
-  name: gitea-postdeploy-validator
-  namespace: ecommerce
+  name: ${POSTDEPLOY_SERVICE_ACCOUNT}
+  namespace: ${POSTDEPLOY_NAMESPACE}
 subjects:
  - kind: ServiceAccount
-    name: gitea-postdeploy-validator
-    namespace: ecommerce
+    name: ${POSTDEPLOY_SERVICE_ACCOUNT}
+    namespace: ${POSTDEPLOY_NAMESPACE}
 roleRef:
  apiGroup: rbac.authorization.k8s.io
  kind: Role
-  name: gitea-postdeploy-validator
+  name: ${POSTDEPLOY_SERVICE_ACCOUNT}
 YAML
+
+  ensure_postdeploy_token_secret
+}
+
+ensure_postdeploy_token_secret() {
+  local sa_uid secret_sa_uid=""
+
+  sa_uid="$(
+    kubectl get serviceaccount "$POSTDEPLOY_SERVICE_ACCOUNT" \
+      -n "$POSTDEPLOY_NAMESPACE" \
+      -o jsonpath='{.metadata.uid}'
+  )"
+
+  [[ -n "$sa_uid" ]] ||
+    die "No fue posible obtener el UID del ServiceAccount postdeploy."
+
+  if kubectl get secret "$POSTDEPLOY_TOKEN_SECRET" \
+      -n "$POSTDEPLOY_NAMESPACE" >/dev/null 2>&1; then
+    secret_sa_uid="$(
+      kubectl get secret "$POSTDEPLOY_TOKEN_SECRET" \
+        -n "$POSTDEPLOY_NAMESPACE" \
+        -o jsonpath='{.metadata.annotations.kubernetes\.io/service-account\.uid}' \
+        2>/dev/null || true
+    )"
+
+    if [[ -n "$secret_sa_uid" && "$secret_sa_uid" != "$sa_uid" ]]; then
+      warn "El token postdeploy pertenece a un ServiceAccount anterior; recreándolo."
+      kubectl delete secret "$POSTDEPLOY_TOKEN_SECRET" \
+        -n "$POSTDEPLOY_NAMESPACE"
+    fi
+  fi
+
+  if ! kubectl get secret "$POSTDEPLOY_TOKEN_SECRET" \
+      -n "$POSTDEPLOY_NAMESPACE" >/dev/null 2>&1; then
+    cat <<YAML | kubectl apply -f -
+apiVersion: v1
+kind: Secret
+metadata:
+  name: ${POSTDEPLOY_TOKEN_SECRET}
+  namespace: ${POSTDEPLOY_NAMESPACE}
+  annotations:
+    kubernetes.io/service-account.name: ${POSTDEPLOY_SERVICE_ACCOUNT}
+type: kubernetes.io/service-account-token
+YAML
+  fi
 }

 generate_postdeploy_credentials() {
  [[ "$APPLY_POSTDEPLOY_RBAC" == "true" ]] || return 0

  local secret_dir="${APP_DATA_PATH}/gitea-actions-secrets"
-  local owner tmp_dir token ca_b64 server kubeconfig_file output_file
+  local owner tmp_dir token_b64="" token="" ca_b64="" server
+  local kubeconfig_file output_file deadline

  owner="$(id -un)"
  server="https://${LAB_HOST_IP}:${API_PORT}"
  tmp_dir="$(mktemp -d)"
  kubeconfig_file="${tmp_dir}/postdeploy-validator.kubeconfig"

-  # No aplicamos chmod a un glob con archivos históricos que podrían pertenecer
-  # a root. Generamos todo en un directorio temporal y luego instalamos cada
-  # archivo con propietario y permisos explícitos.
  sudo_cmd install -d -m 0700 -o "$owner" "$secret_dir"

-  token="$(
-    kubectl create token gitea-postdeploy-validator \
-      -n ecommerce \
-      --duration=8760h
-  )"
+  ensure_postdeploy_token_secret

+  deadline=$((SECONDS + POSTDEPLOY_TOKEN_WAIT_SECONDS))
+
+  while (( SECONDS < deadline )); do
+    token_b64="$(
+      kubectl get secret "$POSTDEPLOY_TOKEN_SECRET" \
+        -n "$POSTDEPLOY_NAMESPACE" \
+        -o jsonpath='{.data.token}' 2>/dev/null || true
+    )"
    ca_b64="$(
-    kubectl config view \
-      --raw \
-      --minify \
-      --flatten \
-      -o jsonpath='{.clusters[0].cluster.certificate-authority-data}'
+      kubectl get secret "$POSTDEPLOY_TOKEN_SECRET" \
+        -n "$POSTDEPLOY_NAMESPACE" \
+        -o jsonpath='{.data.ca\.crt}' 2>/dev/null || true
    )"

-  [[ -n "$token" ]] || die "No fue posible generar el token postdeploy."
-  [[ -n "$ca_b64" ]] || die "No fue posible leer la CA del clúster."
+    if [[ -n "$token_b64" && -n "$ca_b64" ]]; then
+      break
+    fi
+
+    sleep 2
+  done
+
+  [[ -n "$token_b64" ]] ||
+    die "El controlador no generó el token en Secret/${POSTDEPLOY_TOKEN_SECRET}."
+
+  [[ -n "$ca_b64" ]] ||
+    die "El controlador no generó ca.crt en Secret/${POSTDEPLOY_TOKEN_SECRET}."
+
+  token="$(
+    printf '%s' "$token_b64" |
+      base64 --decode
+  )"
+
+  [[ -n "$token" ]] ||
+    die "El token postdeploy quedó vacío."

  printf '%s\n' "$token" > "${tmp_dir}/K8S_TOKEN.txt"
  printf '%s\n' "$ca_b64" > "${tmp_dir}/K8S_CA_B64.txt"
@@ -955,20 +1027,18 @@ clusters:
      server: ${server}
      tls-server-name: ${LAB_HOST_IP}
 contexts:
-  - name: gitea-postdeploy-validator@${CLUSTER_NAME}
+  - name: ${POSTDEPLOY_SERVICE_ACCOUNT}@${CLUSTER_NAME}
    context:
      cluster: ${CLUSTER_NAME}
-      namespace: ecommerce
-      user: gitea-postdeploy-validator
-current-context: gitea-postdeploy-validator@${CLUSTER_NAME}
+      namespace: ${POSTDEPLOY_NAMESPACE}
+      user: ${POSTDEPLOY_SERVICE_ACCOUNT}
+current-context: ${POSTDEPLOY_SERVICE_ACCOUNT}@${CLUSTER_NAME}
 users:
-  - name: gitea-postdeploy-validator
+  - name: ${POSTDEPLOY_SERVICE_ACCOUNT}
    user:
      token: ${token}
 EOF

-  # Compatibilidad con el workflow histórico que usa un único secret
-  # KUBE_CONFIG_DATA en Base64, pero ahora con privilegios mínimos.
  base64 "$kubeconfig_file" |
    tr -d '\n' \
    > "${tmp_dir}/KUBE_CONFIG_DATA.txt"
@@ -988,19 +1058,85 @@ EOF
      "${secret_dir}/${output_file}"
  done

-  # Los kubeconfig históricos también contienen credenciales. Ajustamos
-  # únicamente propietario y permisos; no asumimos ningún grupo llamado devops.
-  for output_file in kubeconfig-gitea.yaml lab-cluster.kubeconfig; do
-    if [[ -f "${secret_dir}/${output_file}" ]]; then
-      sudo_cmd chown "$owner" "${secret_dir}/${output_file}"
-      sudo_cmd chmod 0600 "${secret_dir}/${output_file}"
-    fi
-  done
+  sudo_cmd install \
+    -m 0600 \
+    -o "$owner" \
+    "$kubeconfig_file" \
+    "${secret_dir}/postdeploy-validator.kubeconfig"

  rm -rf "$tmp_dir"

  warn "Credenciales postdeploy regeneradas en ${secret_dir}."
-  warn "Actualiza en Gitea K8S_TOKEN/K8S_CA_B64/K8S_SERVER/K8S_TLS_SERVER_NAME o el secret histórico KUBE_CONFIG_DATA."
+  warn "Actualiza en Gitea el secret KUBE_CONFIG_DATA con KUBE_CONFIG_DATA.txt."
+  ok "Token persistente asociado a Secret/${POSTDEPLOY_TOKEN_SECRET}."
+}
+
+
+validate_postdeploy_credentials() {
+  local secret_dir="${APP_DATA_PATH}/gitea-actions-secrets"
+  local expected_server="https://${LAB_HOST_IP}:${API_PORT}"
+  local server_file="${secret_dir}/K8S_SERVER.txt"
+  local tls_file="${secret_dir}/K8S_TLS_SERVER_NAME.txt"
+  local kubeconfig_file="${secret_dir}/postdeploy-validator.kubeconfig"
+  local actual_server actual_tls kube_server kube_tls can_read
+
+  [[ -s "$server_file" ]] ||
+    die "No existe ${server_file}."
+
+  [[ -s "$tls_file" ]] ||
+    die "No existe ${tls_file}."
+
+  [[ -s "$kubeconfig_file" ]] ||
+    die "No existe ${kubeconfig_file}."
+
+  actual_server="$(tr -d '\r\n\t ' < "$server_file")"
+  actual_tls="$(tr -d '\r\n\t ' < "$tls_file")"
+
+  [[ "$actual_server" == "$expected_server" ]] ||
+    die "K8S_SERVER incorrecto. Actual=${actual_server}; esperado=${expected_server}."
+
+  [[ "$actual_tls" == "$LAB_HOST_IP" ]] ||
+    die "K8S_TLS_SERVER_NAME incorrecto. Actual=${actual_tls}; esperado=${LAB_HOST_IP}."
+
+  kube_server="$(
+    KUBECONFIG="$kubeconfig_file" \
+      kubectl config view \
+        --minify \
+        -o jsonpath='{.clusters[0].cluster.server}' 2>/dev/null || true
+  )"
+
+  kube_tls="$(
+    KUBECONFIG="$kubeconfig_file" \
+      kubectl config view \
+        --minify \
+        -o jsonpath='{.clusters[0].cluster.tls-server-name}' 2>/dev/null || true
+  )"
+
+  [[ "$kube_server" == "$expected_server" ]] ||
+    die "El kubeconfig postdeploy apunta a ${kube_server:-vacío}; esperado=${expected_server}."
+
+  [[ "$kube_tls" == "$LAB_HOST_IP" ]] ||
+    die "El kubeconfig postdeploy usa tls-server-name=${kube_tls:-vacío}; esperado=${LAB_HOST_IP}."
+
+  can_read="$(
+    KUBECONFIG="$kubeconfig_file" \
+      kubectl auth can-i get deployments \
+        -n "$POSTDEPLOY_NAMESPACE" 2>/dev/null || true
+  )"
+
+  [[ "$can_read" == "yes" ]] ||
+    die "El ServiceAccount postdeploy no puede leer Deployments en ecommerce."
+
+  KUBECONFIG="$kubeconfig_file" \
+    kubectl get deployments \
+      -n "$POSTDEPLOY_NAMESPACE" \
+      --request-timeout=15s \
+      >/dev/null
+
+  ok "Credenciales postdeploy verificadas contra ${expected_server}."
+  printf 'K8S_SERVER esperado          : %s\n' "$expected_server"
+  printf 'K8S_TLS_SERVER_NAME esperado : %s\n' "$LAB_HOST_IP"
+  printf 'Archivos para Gitea          : %s\n' "$secret_dir"
 }


@@ -2880,6 +3016,20 @@ main() {
      MONITORING_FORCE_RECOVERY=true recover_monitoring_operation true
      return
      ;;
+    postdeploy-secrets)
+      require_commands
+      ensure_docker
+      validate_data_mount
+      validate_host_ip
+      prepare_directories
+      cluster_exists || die "No existe ${CLUSTER_NAME}."
+      write_kubeconfig
+      wait_for_api
+      apply_postdeploy_rbac
+      generate_postdeploy_credentials
+      validate_postdeploy_credentials
+      return
+      ;;
  esac

  require_commands
@@ -2906,7 +3056,7 @@ main() {
      reset_cluster
      ;;
    *)
-      die "Modo no válido: ${MODE}. Usa bootstrap, recover, ensure, status, repo-sync, gitops, runner, monitoring-diagnose, monitoring-recover, install-autostart o reset."
+      die "Modo no válido: ${MODE}. Usa bootstrap, recover, ensure, status, repo-sync, gitops, runner, monitoring-diagnose, monitoring-recover, postdeploy-secrets, install-autostart o reset."
      ;;
  esac