Google Workload Identity Federation, OIDC, Dex y TypeScript: Conecte su B2B a Gemini y Vertex

Esta guía será exhaustiva y escabrosa para que pueda acceder a las piezas que desee directamente. habrá un repo GitHub acompañante (Vino pronto) que puede usar para pasar por esto. El Índice: About Errata What This Guide Covers and Does Not. What about LLMs and tools? (Gemini, Claude, Claude Code, Cursor, etc.) Definitions Requirements What is Google Cloud Workload Identity Flotation? Architecture Overview Configure a custom OIDP (Open ID Provider) using Dex. How to configure a custom one using DEX, which is an identity provider. How to connect a Docker application to pull it. There's also going to be some code, some JavaScript code that we're going to use to connect directly to it and issue some tokens. Follow for more. Sobre La configuración de una cuenta de Google Cloud Platform (GCP) empresarial presenta algunos retos (como acceder a Gemini a través de Vertex AI en un servicio externo como Vercel/Digital Ocean/Hetzner). Anteriormente, crearía una cuenta de servicio que mantendría los permisos para los servicios (por ejemplo, Vertex AI) y generaría una clave de larga duración (sin fecha de caducidad) que usaría para conectarse a GCP. Esto no se recomienda. La clave de cuenta de servicio puede estar comprometida, lo que dificulta la rotación. Google recomienda utilizar tokens de corta duración a través de Workload Identity Federation y Open ID Connect para comunicarse con los servicios. Pero usted genera en su lugar. __still use __a service account Un token de corta duración Un token de corta duración La advertencia es que esta configuración no es sencilla. La documentación es escasa, y es una confusión. Me tomó 1,5 meses a tiempo parcial para hacerlo bien. No te preocupes, Esta guía te enseñará cómo hacer esto paso a paso. Configuraremos un Proveedor de Identificación Abierta (Dex) que se utilizará para emitir tokens, y conectará Docker y nuestra aplicación para que podamos extraer imágenes de Artifact Registry y conectarnos a Vertex AI, todo a través de un VPS (Hetzner). erróneamente He hecho esto al mejor de mis capacidades. ¡Siéntate libre de llegar y sugerir mejoras y o corregir errores! Qué cubre esta guía y qué no. y cubiertas: Configurar un proveedor de identidad externo utilizando Dex que autentique contra Google Cloud Poner una imagen a través del registro de artefactos utilizando Docker y OpenID Connect/Workload Identity Federation. Emitir un token OIDC a través de una aplicación NodeJS usando google-auth-library. No se cubre: Cómo inicializar servidores en un VPS externo o Google Cloud. Cómo subir archivos a tu VPS Cómo utilizar la Infraestructura como Código (IAC) para configurar los Servicios de nube de Google. Cómo configurar acciones de GitHub para comunicarse con GCP. Esta guía cubrirá cómo configurar un proveedor de identidad utilizando Dex (Open Source), y generar tokens OIDC contra él, para que pueda conectarse a través de cualquier otro proveedor en el mundo. What about LLMs and tools? (Gemini, Claude, Claude Code, Cursor, etc.) Ellos ayudaron No sabían cómo mezclar la lógica requerida para la Federación de Identidad de Carga de Trabajo de Google con Dex. a lot Póngase el LLM a esta página para ayudarle a configurarlo. Definiciones Plataforma en la nube de Google (GCP) La solución en la nube que usaremos para autenticar con OIDC. abreviado como GCP para este artículo. Federación de Identidad de Trabajo: Es una forma de ejecutar cargas de trabajo (aplicaciones, servicios, tuberías CI/CD, VMs, contenedores, etc.) de un proveedor de nube para autenticarse de forma segura a las APIs de ese proveedor . Fuera without using long-lived service account keys En lugar de darle a su aplicación un archivo de clave estática (que es un riesgo de seguridad significativo si se levanta), WIF permite al proveedor de la nube confiar en una como GitHub Actions, GitLab, Kubernetes o cualquier proveedor compatible con OIDC/SAML. external identity provider (IdP) En otras palabras, genera tokens con permisos específicos que puede enviar de forma segura a través de su carga útil (por ejemplo: almacenamiento en la nube, lista, carga) para acceder a los servicios. Proveedores de identidad Es un hub central que un usuario o máquina (nuestro código) puede usar para “login” o autenticar contra uno o más servicios.En nuestro caso, es el sistema que se utilizará para generar un token válido que se intercambiará con GCP para dar acceso a nuestras aplicaciones. Verás más adelante cómo funciona en detalle. Ejemplo de proveedores de identidad Plataformas de desarrolladores (nativas de OIDC): Estas tienen puntos finales específicos que puede pasar a GCP y generar los tokens necesarios automáticamente. GitHub (Tokens de acciones de GitHub OIDC) GitLab (CI/CD pipelines con OIDC) Bitbucket (Pipelines OIDC) Otros proveedores : octa Autódromo Desarrollo de Azure Active Directory Plataforma de identidad de Google Ping de identidad Centro de Identidad de AWS IAM Keycloak Dex (El que vamos a usar) Dex https://dexidp.io/ Lo configuramos con un nombre de usuario y contraseña, y a través de un punto final REST, lo llamamos usando nuestra herramienta de elección (cURL, wget, JavaScript’s fetch/xhr/axios, etc.) y luego devolverá un JWT que podemos usar para comunicarnos con GCP. Why? Es de código abierto y se puede alojar libremente. Sólo se requiere un archivo YAML simple. You don’t need Kubernetes to run (Although it is popular in that environment). Cubriremos cómo implementar una conexión entre Google Cloud y una fuente externa (como un VPS) utilizando Google Cloud Workload Identity Federation, Dex (un proveedor de identidad), y más. Héctor https://www.hetzner.com/ It’s the Cloud provider chosen for its impressive price/performance ratio. Plus it has a very simple UI and IaC (Infrastructure as Code) plugins that make it a breeze to setup. Doppler https://www.doppler.com/ Es un gestor de secretos. Algo similar a HashiCorp Vault o AWS Secrets Manager. Doppler nos permite almacenar nuestras variables ambientales en diferentes proyectos y ramas, lo que permite a todos los miembros de nuestro equipo compartirlas.Podemos proteger, forcar, aislar, copiarlas sin compartir archivos .env entre los equipos. Incluso para trabajar solo, tiene un nivel gratuito muy generoso que cubrirá todas sus necesidades. lo he estado usando desde el año pasado para gestionar todas mis contraseñas para todos mis servicios. Aunque totalmente opcional, este post asume que estás usando Doppler para cargar los entornos (puedes cargar un .env junto a los comandos y estarás bien de ir) Los requisitos . A Google Cloud Platform (GCP) account Have a server you can connect to (SSH): Virtual Machine - VPS, Bare Metal, etc. A domain name (.com, .ai, .app, .io etc.) pointing to your server. Basic Knowledge of: Docker Google CLI Bash I assume: You have a working GCP project. You have Artifact Registry enabled with a docker image Have installed Docker in your VPS. Instale Google CLI en su computadora Descargar el enlace de instalación: https://cloud.google.com/sdk/docs/install Descargar el enlace de instalación: https://cloud.google.com/sdk/docs/install Nuestra primera tarea es configurar los permisos necesarios y habilitar los servicios. Estas son las ubicaciones desde las que se puede instalar: (Tenga en cuenta que los pasos adicionales se omiten en este artículo.) Para Windows: Descarga el instalador aquí MacOS: Instalar Brew (en el caso de que no lo hayas hecho): /bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)" Instalación de Google CLI: brew install --cask gcloud-cli Para Linux: (Descarga el archivo) curl -O https://dl.google.com/dl/cloudsdk/channels/rapid/downloads/google-cloud-cli-linux-x86_64.tar.gz Lo abrimos: tar -xf google-cloud-cli-linux-x86_64.tar.gz Create a Project in Google Cloud or use an existing one Vaya a : https://console.cloud.google.com/ Seleccione su cuenta y proceda a crear un nuevo proyecto si ya no lo ha hecho. Inicio Google Cloud CLI Después de instalarlo, abra un terminal y ejecute: gcloud init Esto iniciará el proceso de conexión. El CLI funciona mediante la vinculación de su cuenta de Google. Esta cuenta debe tener los permisos necesarios para realizar operaciones administrativas. El CLI funciona mediante la vinculación de su cuenta de Google. Esta cuenta debe tener los permisos necesarios para realizar operaciones administrativas. Choose the project you’ve created. Después de la autenticación, el CLI le pedirá que elija el proyecto en el que desee trabajar. This can be changed later on with: gcloud config set project PROJECT_ID Siempre puedes pasar por el Bandera de los comandos. --project PROJECT_ID all the gcloud Enable the Services: Usted tendrá que permitir esto. En tu terminal, corre: gcloud services enable \ iamcredentials.googleapis.com \ sts.googleapis.com \ iam.googleapis.com \ cloudresourcemanager.googleapis.com Variables Ambientales Usado en todo el post. Sustituye estos por los suyos. export PROJECT_ID="spiritual-slate-445211-i1" export PROJECT_NUMBER="1016670781645" export POOL_ID="hetzner-pool" export PROVIDER_ID="hetzner-provider" export SERVICE_ACCOUNT_ID="hetzner" export SERVICE_ACCOUNT="$SERVICE_ACCOUNT_ID@${PROJECT_ID}.iam.gserviceaccount.com" export DEX_ISSUER="https://auth.yourdomain.com" export ROLE_NAME="hetzner_role" Puede obtener el ID del proyecto y el Número del proyecto desde la página principal. Seleccione su proyecto desde el botón de la esquina superior izquierda. Puede obtener el ID del proyecto y el Número del proyecto desde la página principal. Seleccione su proyecto desde el botón de la esquina superior izquierda. Configurar la Federación de Identidad de Carga de Trabajo Cómo funciona en GCP: Crea un grupo de identidades que tendrá una serie de proveedores o clientes conectados a él. Estos proveedores son aquellos contra los que se utilizará el token de Dex para autenticarse. Tratamos de restringirlos usando un “mapeo de atributos”, analizando el JWT de Dex, comparando los campos del mapeo de atributos con los del token. Piense en el pool como un grupo de posibles métodos de autenticación que usted deja saber a GCP que otras terceras partes pueden autenticarse contra. Piense en el pool como un grupo de posibles métodos de autenticación que usted deja saber a GCP que otras terceras partes pueden autenticarse contra. Crea la piscina Run this command: gcloud iam workload-identity-pools create "$POOL_ID" \ --project="$PROJECT_ID" \ --location="global" \ --display-name="Hetzner Workload Identity Pool" \ --description="Pool for Hetzner VPS to access GCP services via OIDC" Example: gcloud iam workload-identity-pools create "hetzner-pool" \ --project="spiritual-slate-445211-i1" \ --location="global" \ --display-name="Hetzner Workload Identity Pool" \ --description="Pool for Hetzner VPS to access GCP services via OIDC" Ejemplo de PowerShell: gcloud iam workload-identity-pools create "hetzner-pool" ` --project="spiritual-slate-445211-i1" ` --location="global" ` --display-name="Hetzner Workload Identity Pool" ` --description="Pool for Hetzner VPS to access GCP services via OIDC" Creación del proveedor OIDC Este será el punto central de conexión con Google Cloud. Este será el punto de entrada que valida el JWT de Dex. Be careful with the --attribute-mapping flag. This is the one that tells GCP to compare the fields with JWT for OIDC. I recommend you (while testing) to start broad and then narrow down the permission scopes as you become more knowledgeable. Be careful with the --attribute-mapping flag. This is the one that tells GCP to compare the fields with JWT for OIDC. I recommend you (while testing) to start broad and then narrow down the permission scopes as you become more knowledgeable. # Create OIDC provider pointing to your DEX instance gcloud iam workload-identity-pools providers create-oidc "$PROVIDER_ID" \ --project="$PROJECT_ID" \ --location="global" \ --workload-identity-pool="$POOL_ID" \ --display-name="Hetzner Dex OIDC Provider" \ --description="OIDC provider using Dex for Hetzner VPS authentication" \ --attribute-mapping="google.subject=assertion.sub,attribute.email=assertion.email,attribute.groups=assertion.groups" \ --issuer-uri="$DEX_ISSUER" El ejemplo: # Create OIDC provider pointing to your DEX instance gcloud iam workload-identity-pools providers create-oidc "hetzner-provider" \ --project="spiritual-slate-445211-i1" \ --location="global" \ --workload-identity-pool="hetzner-pool" \ --display-name="Hetzner Dex OIDC Provider" \ --description="OIDC provider using Dex for Hetzner VPS authentication" \ --attribute-mapping="google.subject=assertion.sub,attribute.email=assertion.email,attribute.groups=assertion.groups" \ --issuer-uri="https://auth.yourdomain.com" This will inform GCP that the JWT’s subject and email must match those from the pool to be considered valid. Configuración de la cuenta de servicio Sí, todavía necesita crear una cuenta de servicio con los permisos finales que se utilizarán para acceder a sus recursos de GCP. The difference is that we will this account using versus using the permanent service account key. impersonate short-lived tokens short-lived tokens Create a service account. gcloud iam service-accounts create "$SERVICE_ACCOUNT_ID"\ --project="$PROJECT_ID" \ --display-name="Hetzner VPS Service Account" \ --description="Service account for Hetzner VPS containers" El ejemplo: gcloud iam service-accounts create "hetzner" \ --project="spiritual-slate-445211-i1" \ --display-name="Hetzner VPS Service Account" \ --description="Service account for Hetzner VPS containers" El PowerShell: gcloud iam service-accounts create "hetzner" ` --project="spiritual-slate-445211-i1" ` --display-name="Hetzner VPS Service Account" ` --description="Service account for Hetzner VPS containers" Assigning the roles to the service account Puede asignar los roles predefinidos del GCP visitando este enlace: https://cloud.google.com/iam/docs/roles-permissions He optado por un enfoque más granular que establece los permisos directamente en un rol personalizado, lo que me ayuda a reducir el área de superficie de ataque. OR - Asignación de roles o permisos granulares: Este enfoque lleva un poco atrás y adelante con GCP (403 errores prohibidos aquí y allá). You can find the granular permissions in this link: https://cloud.google.com/iam/docs/permissions-reference We start by creating a custom role. Create a custom role for the service account - with minimum permissions To create a custom role, we need a YAML file that will hold each permission: Create a file named “roles-hetzner.gcp.yml” and copy and paste the following (Note this file lives locally in your machine or repo) # To update the role: gcloud iam roles update hetzner_role --project=spiritual-slate-445211-i1 --file=./roles-hetzner.gcp.yml title: Hetzner GCP Roles description: | This policy ensures that the Hetzner VPS has the required permissions to access all the Google Cloud services needed for running Docker containers with the same functionality as when they were deployed on Cloud Run. stage: GA # https://cloud.google.com/iam/docs/permissions-reference includedPermissions: # === IAM Permissions === # For service account creation and management - iam.serviceAccounts.getAccessToken - iam.serviceAccounts.signBlob # Required for signed URLs # === Artifact Registry Permissions === # For Docker image storage and management - artifactregistry.repositories.get - artifactregistry.repositories.list - artifactregistry.repositories.downloadArtifacts - artifactregistry.packages.get - artifactregistry.packages.list - artifactregistry.versions.get - artifactregistry.versions.list - artifactregistry.dockerimages.get - artifactregistry.dockerimages.list - artifactregistry.tags.get - artifactregistry.tags.list - artifactregistry.files.get - artifactregistry.files.list - artifactregistry.files.download - artifactregistry.versions.get - artifactregistry.versions.list - resourcemanager.projects.get - artifactregistry.attachments.get - artifactregistry.attachments.list # === Cloud Storage Permissions === # For bucket and object operations - storage.objects.create - storage.objects.delete # Optional: only include if you need to delete images - storage.objects.get - storage.objects.list - storage.objects.update - storage.objects.getIamPolicy - storage.objects.setIamPolicy # === Vertex AI Permissions === # For AI/ML platform operations - aiplatform.endpoints.predict - aiplatform.endpoints.get # Permissions for operations management - aiplatform.operations.list Luego, ejecuta el comando: gcloud iam roles create "$ROLE_NAME" --project="$PROJECT_ID" --file="./roles-hetzner-gcp.yml" gcloud iam roles create hetzner_role --project="spiritual-slate-445211-i1" --file="./roles-hetzner-gcp.yml" Para refrescar: The --file parameter in the command points to a relative path where your terminal is: Como el archivo se encuentra en la misma ubicación, podemos hacer lo siguiente: gcloud iam roles create "herzner_role" --project="spiritual-slate-445211-i1" --file="roles-hetzner.gcp.yml" Para refrescar: The --file parameter in the command points to a relative path where your terminal is: Como el archivo se encuentra en la misma ubicación, podemos hacer lo siguiente: gcloud iam roles create "herzner_role" --project="spiritual-slate-445211-i1" --file="roles-hetzner.gcp.yml" /Users/joseasilis/Documents/programming/alertdown/libs/infraestructura/src/pulumi Actualizar el papel Para realizar cambios en el rol, puede actualizarlo como: gcloud iam roles update hetzner_role --project=spiritual-slate-445211-i1 --file="./roles-hetzner.gcp.yml" How to get granular use it: Cómo obtener granular usarlo: Search for the service I’d like to use, e.g, Vertex AI, and click it: Usando un LLM o los roles de Editor/Viewer, solicite los roles que desee otorgar. Por ejemplo, descubrimos que es has the permission that lets me call Gemini. Vertex AI User aiplatform.endpoints.predict If you mess up with a permission, you will receive a 403 Forbidden from Google Cloud. Add the permission to the YAML File and update the role. Facilitar los servicios que faltan Puede activar los servicios faltantes mediante la ejecución de: gcloud services enable .googleapis.com Asociar la función a la cuenta de servicio gcloud iam service-accounts add-iam-policy-binding \ "$SERVICE_ACCOUNT" \ --project="$PROJECT_ID" \ --role="projects/$PROJECT_ID/roles/$ROLE_NAME" \ --member="serviceAccount:$SERVICE_ACCOUNT" gcloud iam service-accounts add-iam-policy-binding \ "hetzner@spiritual-slate-445211-i1.iam.gserviceaccount.com" \ --project="spiritual-slate-445211-i1" \ --role="projects/spiritual-slate-445211-i1/roles/hetzner_role" \ --member="serviceAccount:hetzner@spiritual-slate-445211-i1.iam.gserviceaccount.comT" Connecting the Service Account to Workload Identity Federation and allowing Service Account Impersonation Independientemente de los dos pasos anteriores, asociamos la función de carga de trabajoIdentityUser directamente al usuario. (Puede optar por agregar los permisos directamente en su lugar) gcloud iam service-accounts add-iam-policy-binding \ "$SERVICE_ACCOUNT" \ --project="$PROJECT_ID" \ --role="roles/iam.workloadIdentityUser" \ --member="principalSet://iam.googleapis.com/projects/$PROJECT_NUMBER/locations/global/workloadIdentityPools/$POOL_ID/*" Which becomes: gcloud iam service-accounts add-iam-policy-binding \ hetzner \ --project="spiritual-slate-445211-i1" \ --role="roles/iam.workloadIdentityUser" \ --member="principalSet://iam.googleapis.com/projects/1016670781645/locations/global/workloadIdentityPools/hetzner-pool/*" Inside the VPS - Docker Compose Si bien recomiendo tener un servidor separado para el proveedor de OpenID (Dex), decidí alojar mi aplicación y Dex en un solo servidor para reducir los costos. Si bien recomiendo tener un servidor separado para el proveedor de OpenID (Dex), decidí alojar mi aplicación y Dex en un solo servidor para reducir los costos. Now we’re moving to the VPS (E.g: Hetzner). We will git pull/or shell copy our scripts to the server (Docker compose and some shell files) that will help us: Bootstrap the Dex service and set up the IDP. Autenticación contra el Registro de Artifact con Docker. Setup a TLS certificate for our HTTPS domain. Configure Nginx con cero implementaciones a una aplicación Remix/React-Router. Docker Componentes: services: # Certificate management (initial + renewal) certbot: image: certbot/dns-cloudflare restart: unless-stopped volumes: # Map certbot data directly to avoid double nesting - ./certbot/conf:/etc/letsencrypt - ./certbot/www:/var/www/certbot - ./certbot/logs:/var/log/letsencrypt - ./certbot-scripts/certbot-manager.sh:/certbot-manager.sh:ro - /run/secrets/cloudflare.ini:/etc/cloudflare/cloudflare.ini:ro environment: - EMAIL=${CERTBOT_EMAIL:-anyemail@mydomain.com} entrypoint: ['/bin/sh', '/certbot-manager.sh'] networks: - shared-network nginx: image: nginx:stable-alpine restart: unless-stopped ports: - '80:80' # HTTP for certbot challenges and redirects - '443:443' # HTTPS for auth.mydomain.com volumes: - ./nginx/nginx.conf:/etc/nginx/nginx.conf:ro - ./certbot/www:/var/www/certbot:ro - ./certbot/conf:/etc/letsencrypt:ro depends_on: - dex networks: - shared-network # Dex OIDC server dex: build: context: ./dex dockerfile: Dockerfile restart: unless-stopped networks: - shared-network volumes: - ./dex:/etc/dex - ./dex/data:/data - ./dex/entrypoint.sh:/entrypoint.sh:ro - ./dex/config.yaml:/etc/dex/config.yaml:ro environment: - DOPPLER_TOKEN=${DOPPLER_TOKEN} - DEX_GCP_STATIC_CLIENT_ID=${DEX_GCP_STATIC_CLIENT_ID} - DEX_GCP_STATIC_CLIENT_SECRET=${DEX_GCP_STATIC_CLIENT_SECRET} - DEX_GCP_STATIC_PASSWORD_EMAIL=${DEX_GCP_STATIC_PASSWORD_EMAIL} - DEX_GCP_URL=${DEX_GCP_URL} healthcheck: test: ['CMD', 'wget', '-qO-', 'http://localhost:5556/healthz'] interval: 5s timeout: 5s retries: 3 start_period: 3s expose: - '5556' # Main application app: image: us-east1-docker.pkg.dev/myfirstproject/remix-app-artifact-repo-production/remix-app-production:latest restart: unless-stopped networks: - shared-network working_dir: /app/apps/remix-app-vite command: ['doppler', 'run', '--', 'bun', 'server.ts'] expose: - '8080' environment: - DOPPLER_TOKEN=${DOPPLER_REMIX_TOKEN} healthcheck: test: [ 'CMD', 'sh', '-c', 'test ! -f /tmp/drain && curl -f --max-time 5 http://localhost:8080/health', ] interval: 5s timeout: 10s retries: 3 start_period: 3s # Allow time for app startup labels: # Configure docker-rollout pre-stop hook for graceful shutdown - 'docker-rollout.pre-stop-hook=sleep 7' networks: shared-network: driver: bridge The docker compose is made of 5 parts: - Used to generate a TLS certificate for us. We need to issue an HTTPS connection because Google will connect to our service and verify the authenticity of the request, and because it’s security 101. Nginx then handles the certificate. I’m using Cloudflare as a proxy. This version of Certbot connects to our Cloudflare account, manages the DNS verification for us, and generates the certificate automatically. Certbot Will consume the TLS certificate for both: Dex (IDP), and the React Router app. It will also serve as a load balancer when I’m trying to perform zero-downtime deployments with . Nginx - Docker Rollout - The star of the show. It will create the Identity Provider that will issue our OIDC tokens that will be used to exchange for short-lived credentials to impersonate the service account. The beastly characteristic of Dex is that it is configured with a single YAML file. Dex App - La imagen del docker de NodeJS Remix/React Router alojado en el Registro de Artifact. Esto se construyó en un tubo de CI/CD y luego se lo empujó (Fuera del alcance de este tutorial). La imagen de Docker alojada en el Registro de Artifact, que se configurará más adelante para sacarlo de Google usando la Federación de Identidad de Carga de Trabajo y OIDC. Este tiene una comprobación de salud que se utiliza por el despliegue de Docker para matar el viejo contenedor (ver abajo). - won’t cut it. We need to make sure certificates are issued first, and it was easier for me to handle the orchestration with a shell script. Some kung-fu was required to have Nginx up and running (As Dex needed to be booted up first before the app) The startup shell script docker compose up Docker Rollout (A bit outside scope, but I wanted to include it anyway) Es un script de archivo único creado por Wowu that helps us have zero-downtime Docker container updates without the need for Kubernetes. https://github.com/Wowu/docker-rollout Para instalar, ejecute en el VPS lo siguiente: # Create directory for Docker cli plugins mkdir -p ~/.docker/cli-plugins # Download docker-rollout script to Docker cli plugins directory curl https://raw.githubusercontent.com/wowu/docker-rollout/main/docker-rollout -o ~/.docker/cli-plugins/docker-rollout # Make the script executable chmod +x ~/.docker/cli-plugins/docker-rollout Using it is very simple. In every place where we would use (note the new syntax), We’d use en su lugar. docker compose docker rollout To deploy an updated image, you the new image from Artifact Registry and execute . It will automatically switch images without downtime. docker compose pull app docker rollout Su servicio no puede tener container_name y puertos definidos en docker-compose.yml, ya que no es posible ejecutar múltiples contenedores con el mismo nombre o mapeo de puertos. Su servicio no puede tener container_name y puertos definidos en docker-compose.yml, ya que no es posible ejecutar múltiples contenedores con el mismo nombre o mapeo de puertos. How does it work? Docker rollout will spin up a new instance of the app (That’s why we don’t use a Una vez que el nuevo servicio está en funcionamiento, añade un archivo sentinel vacío sin extensión (puede ser una llamada de punto final), llamado En el directorio temporal Contenedores de Docker: . container_name drain within the /tmp/drain This forces the health check to fail and signals Docker rollout to kill the old one and keep the new one alive. Parte de la magia se hace a través de Nginx. automáticamente carga los equilibrios de los servidores de arriba (habrá dos instancias de la aplicación al hacer el cambio), y una vez que detecte uno que falla con una comprobación de salud, dejará de servirlo y se redirigirá a la nueva. En otros tipos de aplicaciones, que no se colocan delante de un balanceador de carga, como un trabajador Temporal.io (Fuera del alcance de este tutorial), tendrá que añadir un mecanismo para drenar al trabajador y dejarlo automáticamente. make the change. docker rollout Doppler I stated that Doppler is a service responsible for handling all our environment variables. It’s straightforward to get started: You create an account with them, install the CLI, and add your environments to their service. You'll see down below that Dex will ask for a bcrypt hash. These hashes contain dollar signs ($) which need to be escaped in bash. Don't escape them in Doppler. We will handle them using a file replacement mechanism instead. Esto me tomó 7 horas sólidas para conseguirlo. Verás abajo que Dex pedirá un hash bcrypt. Estos hashes contienen signos de dólar ($) que deben escapar en bash. No los escape en Doppler. This took me a solid 7 hours to get right. CLI installation You will need to install Doppler in the VPS or the server that you are currently using. You can install it using the following command: curl -Ls --tlsv1.2 --proto "=https" --retry 3 https://cli.doppler.com/install.sh | sudo sh Generate a Doppler Token to connect to the Doppler Service Dex Configuration Dex es un “Federated OpenID Connect Provider”.En otras palabras, actúa como un intermediario para conectarse a Proveedores de Identidad (Piensa como estos: Regístrate con Google, Regístrate con GiTHub).OpenID es un protocolo de autenticación que se basa en OAuth 2.0, que permite a las aplicaciones y usuarios obtener información de perfil de usuario con una API similar a REST. It is provided as a Docker image, and we will configure it using Docker Compose. Lo que me gusta de Dex es su simplicidad. sólo necesitas un archivo YAML único que mantendrá toda la configuración. Create a file and place it in your VPS in (Crear el directorio con ) config.yaml ~/dex/config.yaml mkdir ~/dex # dex/config.yaml - Configuration for Google Cloud Workload Identity Federation issuer: $DEX_GCP_URL storage: type: sqlite3 config: file: /var/dex/dex.db web: # Listen on HTTP, assuming a reverse proxy handles TLS termination. http: 0.0.0.0:5556 # Enable the password database to allow authentication with static passwords. enablePasswordDB: true # Define a static user for authentication. # This user will be used to exchange a Dex ID token for a GCP token. staticPasswords: - email: $DEX_GCP_STATIC_PASSWORD_EMAIL hash: $DEX_GCP_STATIC_PASSWORD_SECRET_BCRYPT_HASHED username: $DEX_GCP_STATIC_PASSWORD_EMAIL userID: '722ba69a-3cba-4007-8a24-2611d4c4d5f9' # The `staticClients` list contains OAuth2 clients that can connect to Dex. staticClients: # This is the client for Google Cloud Workload Identity Federation. # The `id` MUST be the full resource name of the GCP Workload Identity Provider. # This value will be the `aud` (audience) claim in the OIDC token. - id: $DEX_GCP_STATIC_CLIENT_ID secret: $DEX_GCP_STATIC_CLIENT_SECRET name: 'Google Cloud Workload Identity Federation' # Redirect URIs are not used in the token-exchange flow. redirectURIs: [] # This section configures OAuth2 behavior. oauth2: # Use the built-in password database as the connector for the password grant type. # This allows the static user defined above to authenticate. passwordConnector: local # By default, Dex supports the necessary grant types, including 'token-exchange' # and the response types 'code', 'token', and 'id_token'. # Explicitly defining them is not necessary unless you need to restrict them. skipApprovalScreen: true Si lo reemplazamos con nuestros entornos, obtendremos: # dex/config.yaml - Configuration for Google Cloud Workload Identity Federation issuer: https://auth.yourdomain.com storage: type: sqlite3 config: file: /var/dex/dex.db web: # Listen on HTTP, assuming a reverse proxy handles TLS termination. http: 0.0.0.0:5556 # Enable the password database to allow authentication with static passwords. enablePasswordDB: true # Define a static user for authentication. # This user will be used to exchange a Dex ID token for a GCP token. staticPasswords: - email: auth@mydomain.com hash: $2y$10$k2IZomh1UUyDlNrsuoNiFuNlOIn5Siw738AdFA6ukcMu07H0uGQ7K username: auth@mydomain.com userID: '722ba69a-3cba-4007-8a24-2611d4c4d5f9' # The `staticClients` list contains OAuth2 clients that can connect to Dex. staticClients: # This is the client for Google Cloud Workload Identity Federation. # The `id` MUST be the full resource name of the GCP Workload Identity Provider. # This value will be the `aud` (audience) claim in the OIDC token. - id: //iam.googleapis.com/projects/1016670781645/locations/global/workloadIdentityPools/hetzner-pool/providers/hetzner-provider secret: 28XhU2xcgQnusLRlG4nZlUZbFdn3lfof21jvvbTG970 name: 'Google Cloud Workload Identity Federation' # Redirect URIs are not used in the token-exchange flow. redirectURIs: [] # This section configures OAuth2 behavior. oauth2: # Use the built-in password database as the connector for the password grant type. # This allows the static user defined above to authenticate. passwordConnector: local # By default, Dex supports the necessary grant types, including 'token-exchange' # and the response types 'code', 'token', and 'id_token'. # Explicitly defining them is not necessary unless you need to restrict them. skipApprovalScreen: true Here’s the breakdown: El emisor issuer: $DEX_GCP_URL Define un camino o subdominio que será único (necesitará actualizar su espacio de nombres para coincidir con esto - lo veremos más adelante) In other words, it will become something like this: issuer: https://auth.mydomain.com The Storage storage: type: sqlite3 config: file: /var/dex/dex.db Dex needs to store its state so it can handle refresh tokens, invalidate current JWTs, and more. Fortunately, we can keep it simple by using SQLite, which it will generate for us. About TLS certificates Dex isn't handling the TLS configuration, Nginx is (You’ll see it in the config). In other deployments, you can let Dex handle the certificates directly by pointing it to the cert files. However, that will be outside the scope of this tutorial. El puerto web web: # Listen on HTTP, assuming a reverse proxy handles TLS termination. http: 0.0.0.0:5556 Este es el puerto que se expone dentro del contenedor Docker. Veremos más adelante sobre cómo lo mapearemos a nuestro contenedor Nginx. Generar una contraseña estática: enablePasswordDB: true # Define a static user for authentication. # This user will be used to exchange a Dex ID token for a GCP token. staticPasswords: - email: $DEX_GCP_STATIC_PASSWORD_EMAIL hash: $DEX_GCP_STATIC_PASSWORD_SECRET_BCRYPT_HASHED username: $DEX_GCP_STATIC_PASSWORD_EMAIL userID: '722ba69a-3cba-4007-8a24-2611d4c4d5f9' # The `staticClients` list contains OAuth2 clients that can connect to Dex. staticClients: # This is the client for Google Cloud Workload Identity Federation. # The `id` MUST be the full resource name of the GCP Workload Identity Provider. # This value will be the `aud` (audience) claim in the OIDC token. - id: $DEX_GCP_STATIC_CLIENT_ID secret: $DEX_GCP_STATIC_CLIENT_SECRET name: 'Google Cloud Workload Identity Federation' # Redirect URIs are not used in the token-exchange flow. redirectURIs: [] # This section configures OAuth2 behavior. oauth2: # Use the built-in password database as the connector for the password grant type. # This allows the static user defined above to authenticate. passwordConnector: local # By default, Dex supports the necessary grant types, including 'token-exchange' # and the response types 'code', 'token', and 'id_token'. # Explicitly defining them is not necessary unless you need to restrict them. skipApprovalScreen: true Esto nos permitirá proporcionar un nombre de usuario y contraseña para autenticar contra Dex, que generará el JWT para que nos autenticemos contra GCP. Esto también se utiliza en otros escenarios para autenticar contra otros proveedores en su nombre. Esto significa que si proporciono el nombre de usuario y contraseña, y tengo una conexión OAuth con GitHub, Microsoft o Google, se conectará a esos servicios en nuestro nombre y devolverá un token que pueda conectarse a esos servicios. This is also used in other scenarios to authenticate against other providers on your behalf. This means that if I provide the username and password, and I have an OAuth connection with GitHub, Microsoft, or Google, it will connect to those services on our behalf and return a token that can connect to those services. But again, this is outside the scope of the tutorial staticClients: ID - la parte mágica: Esto es crucial. Esto especificará el campo de audiencia del JWT generado. Google comparará con esto y validará la solicitud! No puedo decirte cuánto tiempo he gastado en obtener este derecho. This is critical. This will specify the audience field of the generated JWT. Google will compare against this and validate the request! I cannot tell you how much time I spent on getting this right. We need to provide a to our : Tiempo completo hetzner-provider Tiempo completo //iam.googleapis.com/projects/1016670781645/locations/global/workloadIdentityPools/hetzner-pool/providers/hetzner-provider Nota que empezamos con: // Note we start with: // : password Use your password manager and generate a strong password. using any , so escaping isn’t an issue. This will be used to protect your API from anyone accessing your service! Avoid dollar signs a-strong-password-generated-using-a-password-generator Avoid Signos de dólar Usted pasará esta contraseña al enviar una carga útil HTTP a Dex más adelante! Pasas a Dex la cadena hashada de la contraseña de arriba en el field. staticPasswords.password Palabras de Estado: the auth@mydomain.com → This can be anything. Since this will not be propagated upward in the OpenID chain (i.e., used by another provider), it is only used to authenticate against the service. So you can provide something non-existent email: : a random UUID. You can use ULIDs or anything. This field isn’t used in our case. userID It’s a bcrypt hash that you generate using: password: htpasswd -bnBC 10 "" El E.G: htpasswd -bnBC 10 "" a-strong-password-generated-using-a-password-generator It generates: $2y$10$s4ETxkQQeuJu4Kp58O607u.wiqlHnkyV8LkFK1g4cMKGFU959uusq (Note that it starts with a dollar sign) A hashed bcrypt password will have 3 dollar signs as shown above. These need to be escaped correctly in bash/shell if you try to inject them as environment variables. As mentioned in the Dex section, it becomes problematic. I’ll show you how to overcome this using envsubst. Una contraseña bcrypt hashada tendrá 3 signos de dólar como se muestra anteriormente. Estos deben escaparse correctamente en bash/shell si intenta inyectarlos como variables ambientales. Como se mencionó en la sección Dex, se vuelve problemático. te mostraré cómo superar esto usando envsubst. Creating an entry point for Dex: #!/bin/sh set -e echo "🔧 Dex Custom Entrypoint - Starting initialization..." # Export all current environment variables echo "📝 Exporting current environment variables..." export $(printenv | grep -v '^_' | cut -d= -f1) # Check if doppler is configured if [ -z "$DOPPLER_TOKEN" ]; then echo "❌ Error: DOPPLER_TOKEN environment variable is required" exit 1 fi echo "🔐 Fetching DEX_GCP_STATIC_PASSWORD_SECRET_BCRYPT_HASHED from Doppler..." # Get the bcrypt hashed password from doppler DEX_GCP_STATIC_PASSWORD_SECRET_BCRYPT_HASHED=$(doppler secrets get DEX_GCP_STATIC_PASSWORD_SECRET_BCRYPT_HASHED --plain) if [ -z "$DEX_GCP_STATIC_PASSWORD_SECRET_BCRYPT_HASHED" ]; then echo "❌ Error: Failed to retrieve DEX_GCP_STATIC_PASSWORD_SECRET_BCRYPT_HASHED from Doppler" exit 1 fi # Export the retrieved password hash export DEX_GCP_STATIC_PASSWORD_SECRET_BCRYPT_HASHED # Apply environment variable substitution to config echo "🔄 Applying environment variable substitution..." envsubst /tmp/config.yaml # Cleanup function cleanup() { echo "🧹 Cleaning up temporary files..." if [ -f "/tmp/config.yaml" ]; then rm -f /tmp/config.yaml echo "✅ Removed /tmp/config.yaml" fi echo "👋 Dex container shutdown complete" exit 0 } # Set up signal handlers for graceful shutdown trap cleanup SIGTERM SIGINT SIGQUIT echo "🚀 Starting Dex server..." # Start dex server with processed config in background dex serve /tmp/config.yaml & DEX_PID=$! # Wait for dex process to finish wait $DEX_PID This file serves three purposes: Load all the environments from Doppler VPS project Load specifically the Bcrypt password hash in a plain format so it does not add any escaping. DEX_GCP_STATIC_PASSWORD_SECRET_BCRYPT_HASHED=$(doppler secrets get DEX_GCP_STATIC_PASSWORD_SECRET_BCRYPT_HASHED --plain) Use and generate a temporary file which we serve in the with all of our environments replaced. This was the only working mechanism I found that worked. We remove this temporary file in the cleanup function. env_subst config.yaml dex serve /tmp/config.yaml When using an .env file, this script isn’t needed. You can boot Dex directly. However, downloading the hashed bcrypt from Doppler became a mess. This solution nailed it. Cuando se utiliza un archivo .env, este script no es necesario. Puedes iniciar Dex directamente. Sin embargo, descargar el bcrypt hashado de Doppler se convirtió en una confusión. El Nginx: NGINX es un servidor web HTTP, proxy reverso, caché de contenido, balanceador de carga, servidor proxy TCP/UDP y servidor proxy de correo. Es uno de los esqueletos de toda la web. We will use it as the main entry point for our main Docker application and our DEX service. It will also be responsible for handling the TLS connections for us. Sin embargo, esta configuración es un poco desafiante. Nginx no cargará correctamente a menos que todos los servicios estén en funcionamiento. le mostraré a continuación cómo utilizamos un script de automatización que aprovecha dos configuraciones de Nginx, proporcionando tiempo de inactividad cero. Sin embargo, esta configuración es un poco desafiante. Nginx no cargará correctamente a menos que todos los servicios estén en funcionamiento. le mostraré a continuación cómo utilizamos un script de automatización que aprovecha dos configuraciones de Nginx, proporcionando tiempo de inactividad cero. We use these two configurations because NGINX will not load properly unless all the services are available. As you can see, our main application is held in Artifact Registry, which also needs DEX to be authenticated against. To overcome this ,we: Load a Dex-only configuration first. Authenticate Docker against it, generating an OIDC token that authenticates in GCP. Pull the image from Artifact Registry. Provide a second configuration that includes our app, and we reload it in real-time using nginx -s reload nginx.conf.init (Initial Dex only config): user nginx; worker_processes auto; error_log /var/log/nginx/error.log warn; pid /var/run/nginx.pid; events { worker_connections 1024; } http { include /etc/nginx/mime.types; default_type application/octet-stream; # Basic server for returning a maintenance page server { listen 443 ssl http2 default_server; server_name _; # Catch-all ssl_certificate /etc/letsencrypt/live/yourdomain.com/fullchain.pem; ssl_certificate_key /etc/letsencrypt/live/yourdomain.com/privkey.pem; ssl_protocols TLSv1.2 TLSv1.3; # Return a 503 for all requests to the main app location / { return 503 '{"status": "initializing", "message": "Application is starting up, please try again shortly."}'; add_header Content-Type 'application/json'; } } # HTTPS server for auth.yourdomain.com (dex) server { listen 443 ssl http2; server_name auth.yourdomain.com; ssl_certificate /etc/letsencrypt/live/yourdomain.com/fullchain.pem; ssl_certificate_key /etc/letsencrypt/live/yourdomain.com/privkey.pem; ssl_protocols TLSv1.2 TLSv1.3; location / { proxy_pass http://dex:5556; proxy_set_header Host $host; proxy_set_header X-Real-IP $remote_addr; proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for; proxy_set_header X-Forwarded-Proto $scheme; proxy_redirect off; } location /healthz { access_log off; proxy_pass http://dex:5556/healthz; } } # HTTP server for redirects and ACME challenges server { listen 80; server_name www.yourdomain.com yourdomain.com auth.yourdomain.com; location /.well-known/acme-challenge/ { root /var/www/certbot; } location / { return 301 https://$host$request_uri; } } } nginx.conf.prod (Full config) # https://www.digitalocean.com/community/tutorials/how-to-upgrade-nginx-in-place-without-dropping-client-connections # https://www.f5.com/company/blog/nginx/avoiding-top-10-nginx-configuration-mistakes user nginx; # Number of CPU in the server - cat /proc/cpuinfo worker_processes auto; error_log /var/log/nginx/error.log warn; pid /var/run/nginx.pid; events { worker_connections 4096; use epoll; multi_accept on; } http { include /etc/nginx/mime.types; default_type application/octet-stream; # Performance optimizations sendfile on; tcp_nopush on; tcp_nodelay on; keepalive_timeout 30; keepalive_requests 1000; server_tokens off; # Buffer optimizations client_body_buffer_size 128k; client_max_body_size 50m; client_header_buffer_size 1k; large_client_header_buffers 4 4k; output_buffers 1 32k; postpone_output 1460; # Gzip compression gzip on; gzip_vary on; gzip_min_length 1024; gzip_proxied any; gzip_comp_level 6; gzip_types text/plain text/css text/xml text/javascript application/json application/javascript application/xml+rss application/atom+xml image/svg+xml; # Rate limiting limit_req_zone $binary_remote_addr zone=auth:10m rate=10r/m; limit_req_zone $binary_remote_addr zone=api:10m rate=100r/m; # Dynamic upstream for app service (supports docker rollout) upstream app_server { # Use Docker's internal DNS for service discovery # This allows nginx to discover all containers for the 'app' service server app:8080 max_fails=3 fail_timeout=5s; # Configure keepalive connections # Double of the upstream block keepalive 4; keepalive_requests 1000; keepalive_timeout 60s; } # Dynamic upstream for dex service (supports docker rollout) upstream dex_server { server dex:5556 ; } # HTTP server for redirects and ACME challenges server { listen 80; server_name www.yourdomain.com yourdomain.com auth.yourdomain.com; location /.well-known/acme-challenge/ { root /var/www/certbot; } location / { return 301 https://$host$request_uri; } } # HTTPS server for yourdomain.com (main app) server { listen 443 ssl http2; server_name www.yourdomain.com yourdomain.com; ssl_certificate /etc/letsencrypt/live/yourdomain.com/fullchain.pem; ssl_certificate_key /etc/letsencrypt/live/yourdomain.com/privkey.pem; ssl_protocols TLSv1.2 TLSv1.3; ssl_session_cache shared:SSL:50m; ssl_session_timeout 1d; ssl_session_tickets off; ssl_buffer_size 8k; ssl_prefer_server_ciphers on; ssl_ciphers ECDHE-ECDSA-AES128-GCM-SHA256:ECDHE-RSA-AES128-GCM-SHA256:ECDHE-ECDSA-AES256-GCM-SHA384:ECDHE-RSA-AES256-GCM-SHA384:ECDHE-ECDSA-CHACHA20-POLY1305:ECDHE-RSA-CHACHA20-POLY1305:DHE-RSA-AES128-GCM-SHA256:DHE-RSA-AES256-GCM-SHA384; add_header Strict-Transport-Security "max-age=63072000; includeSubDomains; preload" always; limit_req zone=api burst=20 nodelay; # Handle static assets specifically location ~* \.(js|mjs|css|png|jpg|jpeg|gif|ico|svg|woff|woff2|ttf|eot|map)$ { proxy_pass http://app_server; proxy_http_version 1.1; proxy_set_header Connection ""; proxy_set_header Host $host; proxy_set_header X-Real-IP $remote_addr; proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for; proxy_set_header X-Forwarded-Proto $scheme; # Container draining support - retry on different upstream if current fails proxy_next_upstream error timeout http_500 http_502 http_503 http_504; proxy_next_upstream_tries 2; proxy_next_upstream_timeout 10s; # Optimized timeouts for faster failover during rollouts proxy_connect_timeout 3s; proxy_send_timeout 10s; proxy_read_timeout 10s; # Prevent caching of broken responses proxy_buffering off; # Static asset headers expires 1y; add_header Cache-Control "public, max-age=31536000, immutable"; add_header Strict-Transport-Security "max-age=63072000; includeSubDomains; preload" always; } location / { proxy_pass http://app_server; proxy_http_version 1.1; proxy_set_header Connection ""; proxy_set_header Host $host; proxy_set_header X-Real-IP $remote_addr; proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for; proxy_set_header X-Forwarded-Proto $scheme; # Container draining support - retry on different upstream if current fails proxy_next_upstream error timeout http_500 http_502 http_503 http_504; proxy_next_upstream_tries 2; proxy_next_upstream_timeout 15s; # Optimized timeouts for faster failover during rollouts proxy_connect_timeout 5s; proxy_send_timeout 30s; proxy_read_timeout 30s; # Proxy buffering proxy_buffering on; proxy_buffer_size 64k; proxy_buffers 4 64k; proxy_busy_buffers_size 64k; } # Health check endpoints location /health { access_log off; proxy_pass http://app_server/health; # Fast failover for health checks proxy_connect_timeout 2s; proxy_send_timeout 5s; proxy_read_timeout 5s; proxy_next_upstream error timeout http_500 http_502 http_503 http_504; proxy_next_upstream_tries 2; proxy_next_upstream_timeout 5s; } } # HTTPS server for auth.yourdomain.com (dex) server { listen 443 ssl http2; server_name auth.yourdomain.com; ssl_certificate /etc/letsencrypt/live/yourdomain.com/fullchain.pem; ssl_certificate_key /etc/letsencrypt/live/yourdomain.com/privkey.pem; ssl_protocols TLSv1.2 TLSv1.3; ssl_session_cache shared:SSL:50m; ssl_session_timeout 1d; ssl_session_tickets off; ssl_buffer_size 8k; ssl_prefer_server_ciphers on; ssl_ciphers ECDHE-ECDSA-AES128-GCM-SHA256:ECDHE-RSA-AES128-GCM-SHA256:ECDHE-ECDSA-AES256-GCM-SHA384:ECDHE-RSA-AES256-GCM-SHA384:ECDHE-ECDSA-CHACHA20-POLY1305:ECDHE-RSA-CHACHA20-POLY1305:DHE-RSA-AES128-GCM-SHA256:DHE-RSA-AES256-GCM-SHA384; add_header Strict-Transport-Security "max-age=63072000; includeSubDomains; preload" always; limit_req zone=auth burst=5 nodelay; location / { proxy_pass http://dex_server; proxy_set_header Host $host; proxy_set_header X-Real-IP $remote_addr; proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for; proxy_set_header X-Forwarded-Proto $scheme; proxy_redirect off; # Container draining support - retry on different upstream if current fails proxy_next_upstream error timeout http_500 http_502 http_503 http_504; proxy_next_upstream_tries 2; proxy_next_upstream_timeout 10s; # Optimized timeouts for faster failover during rollouts proxy_connect_timeout 5s; proxy_send_timeout 15s; proxy_read_timeout 15s; } # Health check for Dex location /healthz { access_log off; proxy_pass http://dex_server/healthz; # Fast failover for health checks proxy_connect_timeout 2s; proxy_send_timeout 5s; proxy_read_timeout 5s; proxy_next_upstream error timeout http_500 http_502 http_503 http_504; proxy_next_upstream_tries 2; proxy_next_upstream_timeout 5s; } } } ¡Bonus ! F5 es la compañía detrás de NGINX, y tienen un fantástico libro de cocina que te ayuda a comprender mejor cómo configurarlo. https://www.f5.com/content/dam/f5/corp/global/pdf/ebooks/NGINX_Cookbook-final.pdf Bonus! F5 es la compañía detrás de NGINX, y tienen un fantástico libro de cocina que te ayuda a comprender mejor cómo configurarlo. https://www.f5.com/content/dam/f5/corp/global/pdf/ebooks/NGINX_Cookbook-final.pdf Configuring Docker to connect to Artifact Registry using Workload Identity Federation and OIDC. Ahora necesitamos decirle a Docker que se conecte a GCP a través de la Federación de Identidad de Carga de Trabajo. We are going to configure this manually. Credential Helper for that. Se compondrá de tres archivos: fetch-id-token.sh (Esto autentica contra DEX y genera un JWT) (It takes the JWT from fetch-id-token.sh and generates an OIDC token from Google Cloud) fetch-google-oidc-token.sh (It orchestrates the two files above, and sends the OIDC token from fetch-google-oidc-token.sh to GCP in order to impersonate the service account) docker-credential-gcr.sh No copie y pega ciegamente estos archivos, ya que muchos de ellos tienen valores codificados como el usuario actual no-root en el VPS: localuser No copie y pega ciegamente estos archivos, ya que muchos de ellos tienen valores codificados como el usuario actual no-root en el VPS: localuser All of these files need to have execution access: chmod +x ~/scripts/fetch-id-token.sh chmod +x ~/scripts/fetch-google-oidc-token.sh chmod +x /usr/local/bin/docker-credential-gcr Configurar el asistente de credenciales dentro de Docker: Navigate to or create the file if it doesn’t exist, and add the following: ~/.docker/config.json { "credHelpers": { "us-east1-docker.pkg.dev": "gcr" } } This will tell Docker to use the credential helper to authenticate whenever pulling or pushing the image gcr Docker then tries to execute the binary named (extensionless - e.g: ). docker-credential- docker-credential-gcr It is a naming convention used by Docker, and it finds it via $PATH when needed. El sistema de clasificación de ADN-ADN ( ) de : /usr/local/bin/docker-credential-gcr #! /bin/bash # # This file is extensionless. # Update all the /home/localuser/scripts directory with your own! # A custom Docker credential helper for Workload Identity Federation. # This script performs the full OIDC-to-Google-Access-Token exchange. # # Prerequisites: # 1. `curl` and `jq` must be installed. # 2. The custom OIDC token must be available in a file. # 3. The script must be executable (`chmod +x /path/to/this/script.sh`). # 4. This script is copied to /usr/local/bin/docker-credential-gcr (This is a file) # set -euo pipefail $PROJECT_ID="spiritual-slate-445211-i1" # Add logging to a file for debugging exec 2>> /tmp/docker-credential-gcr.log echo "$(date): Starting docker-credential-gcr" >> /tmp/docker-credential-gcr.log set -a source /home/localuser/.env set +a # Step 1: Get the ID Token echo "$(date): Getting ID token" >> /tmp/docker-credential-gcr.log if ! SUBJECT_TOKEN=$(/home/localuser/scripts/fetch-id-token.sh 2>> /tmp/docker-credential-gcr.log); then echo "$(date): ERROR - Failed to get ID token" >> /tmp/docker-credential-gcr.log exit 1 fi echo "$(date): Got ID token: ${SUBJECT_TOKEN:0:20}..." >> /tmp/docker-credential-gcr.log # Step 2: Get federated token echo "$(date): Getting federated token" >> /tmp/docker-credential-gcr.log if ! FEDERATED_ACCESS_TOKEN=$(/home/localuser/scripts/fetch-google-oidc-token.sh "$SUBJECT_TOKEN" 2>> /tmp/docker-credential-gcr.log); then echo "$(date): ERROR - Failed to get federated token" >> /tmp/docker-credential-gcr.log exit 1 fi if [ "${FEDERATED_ACCESS_TOKEN}" == "null" ]; then echo "Error: Failed to get federated token from STS. Response: ${STS_RESPONSE}" >&2 echo "{}" exit 1 fi # Check required environment variable for service account if [[ -z "${GCP_SERVICE_ACCOUNT:-}" ]]; then echo "$(date): ERROR - GCP_SERVICE_ACCOUNT_EMAIL not set" >> /tmp/docker-credential-gcr.log exit 1 fi # Step 4: Use the federated token to impersonate the Service Account. # This calls the IAM Credentials API to get a final, usable Google Cloud access token. IAM_API_URL="https://iamcredentials.googleapis.com/v1/projects/-/serviceAccounts/hetzner@$PROJECT_ID.iam.gserviceaccount.com:generateAccessToken" IAM_PAYLOAD='{"scope": ["https://www.googleapis.com/auth/cloud-platform"]}' IAM_RESPONSE=$(curl -s -X POST "${IAM_API_URL}" \ -H "Accept: application/json" \ -H "Content-Type: application/json" \ -H "Authorization: Bearer ${FEDERATED_ACCESS_TOKEN}" \ -d "${IAM_PAYLOAD}") GOOGLE_ACCESS_TOKEN=$(echo "${IAM_RESPONSE}" | jq -r .accessToken) if [ "${GOOGLE_ACCESS_TOKEN}" == "null" ]; then echo "Error: Failed to get Google access token via impersonation. Response: ${IAM_RESPONSE}" >&2 echo "{}" exit 1 fi # Step 5: Output the credentials in the JSON format that Docker expects. # The "Username" must be "oauth2accesstoken". # The "Secret" is the final Google Cloud access token. cat <<EOF { "Username": "oauth2accesstoken", "Secret": "${GOOGLE_ACCESS_TOKEN}" } EOF This file orchestrates the following two. It also adds some debugging mechanisms to to help you aid in debugging when things fail. /tmp/docker-credential-gcr.log fetch-id-token.sh ( - Crear el Dirección ) : ~/scripts/fetch-id-token.sh scripts #!/usr/bin/env bash # set -euo pipefail # Load environment variables from .env and Doppler set -a source .env 2>/dev/null || true set +a # Check if any required variables are empty if [[ -z "$DEX_GCP_URL" || -z "$DEX_GCP_STATIC_CLIENT_ID" || -z "$DEX_GCP_STATIC_CLIENT_SECRET" || -z "$DEX_GCP_STATIC_PASSWORD_EMAIL" || -z "$DEX_GCP_STATIC_PASSWORD_SECRET" ]]; then echo "Error: One or more required DEX_GCP_* environment variables are empty!" >&2 echo "Make sure DOPPLER_TOKEN is set and you have access to the required secrets." >&2 exit 1 fi BASIC_AUTH_HEADER=$(echo -n "$DEX_GCP_STATIC_CLIENT_ID:$DEX_GCP_STATIC_CLIENT_SECRET" | base64 -w 0) # Request OIDC token from Dex using the correct password grant flow response=$(curl --silent --location "$DEX_GCP_URL/token" \ --header 'Content-Type: application/x-www-form-urlencoded' \ --header "Authorization: Basic $BASIC_AUTH_HEADER" \ --data-urlencode 'grant_type=password' \ --data-urlencode "username=$DEX_GCP_STATIC_PASSWORD_EMAIL" \ --data-urlencode "password=$DEX_GCP_STATIC_PASSWORD_SECRET" \ --data-urlencode 'scope=openid email profile groups' 2>&1) curl_exit_code=$? if [[ $curl_exit_code -ne 0 ]]; then echo "Error: Failed to call Dex API (curl exit code: $curl_exit_code)" >&2 echo "Response: $response" >&2 exit 1 fi # Check if response contains an error if echo "$response" | grep -q '"error"'; then echo "Error: Authentication failed" >&2 echo "Response: $response" >&2 exit 1 fi # Extract the id_token from the JSON response id_token=$(echo "$response" | jq -r '.id_token') # Validate we got a token if [[ -z "$id_token" || "$id_token" == "null" ]]; then echo "Error: Failed to get id_token from Dex. Response: $response" >&2 exit 1 fi # Output the token echo "$id_token" Página de inicio de Google Ads ( ) ~/scripts/fetch-google-oidc-token #!/usr/bin/env bash set -euo pipefail # Check if ID_TOKEN is provided as argument if [[ $# -eq 0 ]]; then echo "Error: ID_TOKEN parameter is required" >&2 echo "Usage: $0 " >&2 exit 1 fi ID_TOKEN="$1" # Validate ID_TOKEN is not empty if [[ -z "$ID_TOKEN" ]]; then echo "Error: ID_TOKEN parameter cannot be empty" >&2 exit 1 fi # Check required environment variables : "${DEX_GCP_STATIC_CLIENT_ID:?DEX_GCP_STATIC_CLIENT_ID environment variable not set}" # Build the audience URL AUDIENCE="$DEX_GCP_STATIC_CLIENT_ID" # Exchange ID_TOKEN for ACCESS_TOKEN using Google STS echo "Calling Google STS with audience: $AUDIENCE" >&2 response=$(curl --silent --location 'https://sts.googleapis.com/v1/token' \ --header 'Content-Type: application/json' \ --data "{ \"grant_type\": \"urn:ietf:params:oauth:grant-type:token-exchange\", \"subject_token_type\": \"urn:ietf:params:oauth:token-type:id_token\", \"subject_token\": \"$ID_TOKEN\", \"audience\": \"$AUDIENCE\", \"requested_token_type\": \"urn:ietf:params:oauth:token-type:access_token\", \"scope\": \"https://www.googleapis.com/auth/cloud-platform\" }") curl_exit_code=$? # Check if curl succeeded if [[ $curl_exit_code -ne 0 ]]; then echo "Error: Failed to call Google STS API (curl exit code: $curl_exit_code)" >&2 echo "STS Response: $response" >&2 exit 1 fi # Extract the access_token from the JSON response access_token=$(echo "$response" | jq -r '.access_token') # Validate we got a token if [[ -z "$access_token" || "$access_token" == "null" ]]; then echo "Error: Failed to get access_token from Google STS. Response: $response" >&2 exit 1 fi # Output the access token echo "$access_token" What these files do in summary: Obtenga un JWT de Dex. Send the JWT to Google to generate an OIDC Token. Utilice el Token OIDC para fingir la cuenta de servicio. Utilice el token como el campo de contraseña del contenedor Docker. Sobre estos archivos: The (Note that this file is extensionless) must be placed in: docker-credential-gcr /usr/local/bin/ The full path should be: /usr/local/bin/docker-credential-gcr This makes it accessible for Docker to authenticate against Artifact Registry In my setup, I have a non-sudo user called . I’ve added the and inside the ’s home directory: localuser fetch-id-token.sh fetch-google-oidc-token.sh localuser /home/localuser/scripts/ The dir is one I created myself. scripts Usted debe actualizar docker-credential-gcr para coincidir con los directorios!!! ( Poniendo todo juntos - El smart-start.sh smart-start.sh As I’ve mentioned, orchestrating this is a bit tricky, as I’m hosting everything within a single server and a single Nginx process, we need to (recap): Generate the TLS certificate Descarga Nginx con Dex. Deja que Docker se autentique contra Dex y GCP - Retire la imagen del Registro de Artifact Reload Nginx config to serve the app. Para ello, hemos creado un ayudante que también requiere acceso ejecutable. smart-start.sh Puedes colocar esto en el or the leading directory next to the El archivo: /home/localuser docker-compose.yml smart-start.sh #!/usr/bin/env bash # # smart-start.sh – one-shot bootstrap for the Hetzner VPS ## # Usage examples: # DOPPLER_TOKEN="dp.st.prd.xyz" ./smart-start.sh # Full deployment # DOPPLER_TOKEN="dp.st.prd.xyz" SKIP_CERT=1 ./smart-start.sh # Skip certificates set -euo pipefail ############################################################################### # Helpers ############################################################################### c_blue='\033[0;34m'; c_green='\033[0;32m'; c_yellow='\033[1;33m'; c_red='\033[0;31m'; c_nc='\033[0m' log(){ printf "${c_blue}ℹ︎ %s${c_nc}\n" "$*"; } ok (){ printf "${c_green}✔ %s${c_nc}\n" "$*"; } warn(){ printf "${c_yellow}⚠ %s${c_nc}\n" "$*"; } die (){ printf "${c_red}✖ %s${c_nc}\n" "$*" >&2; exit 1; } ############################################################################### # Environment & sanity ############################################################################### log "Pre-flight checks" [[ $EUID -eq 0 ]] && die "Run as an unprivileged user with docker group membership." command -v docker >/dev/null || die "Docker missing." command -v doppler >/dev/null || die "Doppler CLI missing." docker compose version >/dev/null || die "Docker Compose plugin missing." docker ps >/dev/null || die "User cannot talk to the Docker socket." # Validate DOPPLER_TOKEN [[ -z "${DOPPLER_TOKEN:-}" ]] && die "DOPPLER_TOKEN environment variable is required." # Helper function to run docker commands with doppler docker_compose() { DOPPLER_TOKEN="$DOPPLER_TOKEN" doppler run -- docker compose "$@" } docker_rollout() { DOPPLER_TOKEN="$DOPPLER_TOKEN" doppler run -- docker rollout "$@" } # # Helper function to check if service is running service_running() { local service="$1" docker_compose ps --services --filter "status=running" | grep -q "^${service}$" } # Helper function to wait for service health wait_for_service_health() { local service="$1" local timeout="${2:-60}" local count=0 log "Waiting for $service health check (timeout: ${timeout}s)..." while [ $count -lt $timeout ]; do if docker_compose ps "$service" --format "table {{.Health}}" 2>/dev/null | grep -q "healthy"; then ok "$service is healthy" return 0 fi # If service is not running, try to check if it's still starting if ! service_running "$service"; then warn "$service is not running, checking if it exited..." docker_compose logs --tail=10 "$service" return 1 fi sleep 2 count=$((count + 2)) done warn "$service health check timeout after ${timeout}s" docker_compose logs --tail=20 "$service" return 1 } ok "Environment looks good." ############################################################################### # TLS (Let's Encrypt via dns-01, can be skipped) ############################################################################### if [[ ${SKIP_CERT:-0} -eq 0 ]]; then log "Phase 1 – TLS certificates" cert_live="certbot/conf/live/alertdown.ai/fullchain.pem" if [[ ! -f $cert_live ]] || ! openssl x509 -checkend 604800 -noout -in "$cert_live" >/dev/null 2>&1 then log "Obtaining / renewing wildcard cert via Cloudflare" [[ -f /run/secrets/cloudflare.ini ]] || die "Cloudflare credentials missing at /run/secrets/cloudflare.ini (chmod 600)." mkdir -p certbot/{conf,logs,www} docker run --rm \ -v "$PWD/certbot/conf:/etc/letsencrypt" \ -v "$PWD/certbot/logs:/var/log/letsencrypt" \ -v /run/secrets/cloudflare.ini:/cloudflare.ini:ro \ certbot/dns-cloudflare certonly \ --dns-cloudflare \ --dns-cloudflare-credentials /cloudflare.ini \ --dns-cloudflare-propagation-seconds 300 \ --email "${CERTBOT_EMAIL:-tech@alertdown.ai}" --agree-tos --no-eff-email \ # We tell Certbot to generate wildcard certificates. -d mydomain.com -d "*.mydomain.com" \ --non-interactive --rsa-key-size 4096 fi ok "TLS certificate ready." else warn "SKIP_CERT=1 → skipping Let's Encrypt" fi ############################################################################### # Bring up auth services (nginx + dex) ############################################################################### log "Phase 2 – Bring up auth services (nginx + dex)" # Use the initial maintenance config first log "Switching to initial NGINX config..." cp nginx/nginx.conf.init nginx/nginx.conf # Start dex first log "Starting dex service..." docker_compose up -d dex if ! wait_for_service_health "dex" 60; then die "Dex failed to become healthy" fi # Start nginx with the init config or reload if already running if service_running "nginx"; then log "Nginx already running, reloading with maintenance config..." if ! docker_compose exec nginx nginx -t -c /etc/nginx/nginx.conf; then die "New nginx config is invalid" fi if ! docker_compose exec nginx nginx -s reload; then die "Nginx reload failed - config may be invalid" fi else # Start nginx normally docker_compose up -d nginx fi # Give nginx a moment to start and verify it's running sleep 5 if ! service_running "nginx"; then docker_compose logs nginx die "Nginx failed to start" fi # Test auth endpoint log "Testing auth endpoint availability..." for i in {1..30}; do if curl -fs "https://auth.mydomain.com/healthz" >/dev/null 2>&1 || \ curl -fs "$DEX_GCP_URL/healthz" >/dev/null 2>&1; then ok "Auth endpoint is accessible" break fi sleep 2 [[ $i -eq 30 ]] && warn "Auth endpoint not yet accessible (may still be starting)" done ############################################################################### # Pull and start app services ############################################################################### log "Phase 3 – Pull and start app service" log "Pulling app image..." if ! docker_compose pull app; then warn "App image pull failed (check WIF setup) – continuing with cached image if available." fi log "Starting app service..." docker_rollout app if ! wait_for_service_health "app" 180; then warn "App service health check failed. The app may not be accessible." # Even if health check fails, we proceed to switch nginx config fi # Switch to the final production nginx config log "Switching to production NGINX config..." cp nginx/nginx.conf.prod nginx/nginx.conf # Reload nginx to apply the new configuration log "Reloading nginx with production config..." if docker_compose exec nginx nginx -s reload; then ok "Nginx reloaded successfully with production config." else warn "Nginx reload failed. Check logs." docker_compose logs nginx fi # Spin up certbot manager to listen for certbot renewals log "Phase 4 – Starting certbot manager" docker_compose up -d certbot ############################################################################### # Health probes ############################################################################### log "Comprehensive health probes" # Internal service health checks declare -A internal_services=( ["dex"]="dex" ["app"]="app" ) for name in "${!internal_services[@]}"; do service="${internal_services[$name]}" if service_running "$service"; then if docker_compose ps "$service" --format "table {{.Health}}" 2>/dev/null | grep -q "healthy"; then ok "Internal $name service ✓" else warn "Internal $name service ✗ (not healthy)" fi else warn "Internal $name service ✗ (not running)" fi done # External endpoint health checks declare -A external_probes=( ["Auth (HTTPS)"]="curl -fs --connect-timeout 10 https://auth.alertdown.ai/healthz" ["App (HTTPS)"]="curl -fs --connect-timeout 10 https://alertdown.ai/health" ["App (HTTPS/nginx-health)"]="curl -fs --connect-timeout 10 https://alertdown.ai/nginx-health" ["HTTP redirect"]="curl -fs --connect-timeout 5 http://alertdown.ai/ | grep -q '301'" ) for name in "${!external_probes[@]}"; do if timeout 15 bash -c "${external_probes[$name]}" >/dev/null 2>&1; then ok "External $name ✓" else warn "External $name ✗" fi done # Service status summary echo log "Service Status Summary:" docker_compose ps --format "table {{.Service}}\t{{.Status}}\t{{.Health}}" echo -e "\n${c_green}🚀 AlertDown stack deployment complete:" echo -e "${c_green} Auth: https://auth.alertdown.ai${c_nc}" echo -e "${c_green} App: https://alertdown.ai${c_nc}" if ! service_running "app" || ! docker_compose ps "app" --format "table {{.Health}}" 2>/dev/null | grep -q "healthy"; then echo -e "${c_yellow} Note: App service may still be starting up${c_nc}" fi Running the script: Para llevar a cabo esto, usted: DOPPLER_TOKEN= bash smart-start.sh E.g: DOPPLER_TOKEN=dp.stasdasdsad22OmPSi bash smart-start.sh Breaking the script apart: #!/usr/bin/env bash # # smart-start.sh – one-shot bootstrap for the Hetzner VPS ## # Usage examples: # DOPPLER_TOKEN="dp.st.prd.xyz" ./smart-start.sh # Full deployment # DOPPLER_TOKEN="dp.st.prd.xyz" SKIP_CERT=1 ./smart-start.sh # Skip certificates set -euo pipefail Selecciona bash, muestra ejemplos y habilita el modo estricto: -e exit on error, -u fail on unset vars, -o pipefail catch pipe falls. 1) helper: colors + logger functions 1) Ayudante: colores + funciones de logger c_blue='\033[0;34m'; c_green='\033[0;32m'; c_yellow='\033[1;33m'; c_red='\033[0;31m'; c_nc='\033[0m' log(){ printf "${c_blue}ℹ︎ %s${c_nc}\n" "$*"; } ok (){ printf "${c_green}✔ %s${c_nc}\n" "$*"; } warn(){ printf "${c_yellow}⚠ %s${c_nc}\n" "$*"; } die (){ printf "${c_red}✖ %s${c_nc}\n" "$*" >&2; exit 1; } Colored output helpers for info/ok/warn/error ( también las salidas). die 2) environment & sanity checks 2) Controles medioambientales y sanitarios log "Pre-flight checks" [[ $EUID -eq 0 ]] && die "Run as an unprivileged user with docker group membership." command -v docker >/dev/null || die "Docker missing." command -v doppler >/dev/null || die "Doppler CLI missing." docker compose version >/dev/null || die "Docker Compose plugin missing." docker ps >/dev/null || die "User cannot talk to the Docker socket." # Validate DOPPLER_TOKEN [[ -z "${DOPPLER_TOKEN:-}" ]] && die "DOPPLER_TOKEN environment variable is required." Ensures: not root (expects user in group), docker , , docker compose plugin available, docker doppler The user can reach the Docker socket, is set. DOPPLER_TOKEN 3) Doppler-wrapped Docker and service helpers 3) Docker envuelto en Doppler y asistentes de servicio docker_compose() { DOPPLER_TOKEN="$DOPPLER_TOKEN" doppler run -- docker compose "$@" } docker_rollout() { DOPPLER_TOKEN="$DOPPLER_TOKEN" doppler run -- docker rollout "$@" } service_running() { local service="$1" docker_compose ps --services --filter "status=running" | grep -q "^${service}$" } wait_for_service_health() { local service="$1" local timeout="${2:-60}" local count=0 log "Waiting for $service health check (timeout: ${timeout}s)..." while [ $count -lt $timeout ]; do if docker_compose ps "$service" --format "table {{.Health}}" 2>/dev/null | grep -q "healthy"; then ok "$service is healthy" return 0 fi if ! service_running "$service"; then warn "$service is not running, checking if it exited..." docker_compose logs --tail=10 "$service" return 1 fi sleep 2 count=$((count + 2)) done warn "$service health check timeout after ${timeout}s" docker_compose logs --tail=20 "$service" return 1 } ok "Environment looks good." y es: / : Runs Docker commands with env injected by Doppler. docker_compose docker_rollout service_running: Verifica si un servicio está en estado de “execución”. Polls Docker Compose health for a service with a timeout; logs on failure. wait_for_service_health 4) TLS via Let’s Encrypt (dns-01) — skippable with SKIP_CERT=1 4) TLS a través de Let’s Encrypt (dns-01) — salpicable con SKIP_CERT=1 if [[ ${SKIP_CERT:-0} -eq 0 ]]; then log "Phase 1 – TLS certificates" cert_live="certbot/conf/live/alertdown.ai/fullchain.pem" if [[ ! -f $cert_live ]] || ! openssl x509 -checkend 604800 -noout -in "$cert_live" >/dev/null 2>&1 then log "Obtaining / renewing wildcard cert via Cloudflare" [[ -f /run/secrets/cloudflare.ini ]] || die "Cloudflare credentials missing at /run/secrets/cloudflare.ini (chmod 600)." mkdir -p certbot/{conf,logs,www} docker run --rm \ -v "$PWD/certbot/conf:/etc/letsencrypt" \ -v "$PWD/certbot/logs:/var/log/letsencrypt" \ -v /run/secrets/cloudflare.ini:/cloudflare.ini:ro \ certbot/dns-cloudflare certonly \ --dns-cloudflare \ --dns-cloudflare-credentials /cloudflare.ini \ --dns-cloudflare-propagation-seconds 300 \ --email "${CERTBOT_EMAIL:-tech@alertdown.ai}" --agree-tos --no-eff-email \ # We tell Certbot to generate wildcard certificates. -d mydomain.com -d "*.mydomain.com" \ --non-interactive --rsa-key-size 4096 fi ok "TLS certificate ready." else warn "SKIP_CERT=1 → skipping Let's Encrypt" fi If not skipping: checks if cert exists and is valid for ≥7 days. alertdown.ai uses to issue/renew wildcard certs via dns-01 with Cloudflare creds at . certbot/dns-cloudflare /run/secrets/cloudflare.ini stores certs under . certbot/conf Nota: los dominios aquí son mydomain.com Pero cert path utiliza alertdown.ai. Asegúrese de que estos coincidan con sus dominios reales. 5) Phase 2 — Bring up the auth stack (Nginx + Dex) 5) Fase 2 – Traer la pila auth (Nginx + Dex) log "Phase 2 – Bring up auth services (nginx + dex)" log "Switching to initial NGINX config..." cp nginx/nginx.conf.init nginx/nginx.conf log "Starting dex service..." docker_compose up -d dex if ! wait_for_service_health "dex" 60; then die "Dex failed to become healthy" fi if service_running "nginx"; then log "Nginx already running, reloading with maintenance config..." if ! docker_compose exec nginx nginx -t -c /etc/nginx/nginx.conf; then die "New nginx config is invalid" fi if ! docker_compose exec nginx nginx -s reload; then die "Nginx reload failed - config may be invalid" fi else docker_compose up -d nginx fi sleep 5 if ! service_running "nginx"; then docker_compose logs nginx die "Nginx failed to start" fi log "Testing auth endpoint availability..." for i in {1..30}; do if curl -fs "https://auth.mydomain.com/healthz" >/dev/null 2>&1 || \ curl -fs "$DEX_GCP_URL/healthz" >/dev/null 2>&1; then ok "Auth endpoint is accessible" break fi sleep 2 [[ $i -eq 30 ]] && warn "Auth endpoint not yet accessible (may still be starting)" done It: Cambiar nginx a una configuración de mantenimiento/inicial. Comienza el desayuno, espera hasta que el contenedor de salud = salud. Asegura que nginx está en funcionamiento; valida la configuración y recarga si ya está en funcionamiento. Pruebas auth salud en https://auth.mydomain.com/healthz o $DEX_GCP_URL/healthz. Nota: de nuevo mydomain.com vs su dominio real; unify. 6) Phase 3 — Pull & start app service, then switch nginx to prod log "Phase 3 – Pull and start app service" log "Pulling app image..." if ! docker_compose pull app; then warn "App image pull failed (check WIF setup) – continuing with cached image if available." fi log "Starting app service..." docker_rollout app if ! wait_for_service_health "app" 180; then warn "App service health check failed. The app may not be accessible." fi log "Switching to production NGINX config..." cp nginx/nginx.conf.prod nginx/nginx.conf log "Reloading nginx with production config..." if docker_compose exec nginx nginx -s reload; then ok "Nginx reloaded successfully with production config." else warn "Nginx reload failed. Check logs." docker_compose logs nginx fi It: Pulsar la imagen de la aplicación (advirtiendo si el arrastre falla; utiliza la imagen en caché). Deploys app via (zero-downtime rollout tool). docker rollout Espera hasta 180s para la salud de la aplicación. Swaps nginx to production config and reloads. 7) Phase 4 — start certbot manager (for renew hooks) 7) Fase 4 – Iniciar el gestor de certbot (para los ganchos de renovación) log "Phase 4 – Starting certbot manager" docker_compose up -d certbot 8) Comprehensive health probes + summary 8) Sondeos de salud completos + resumen log "Comprehensive health probes" declare -A internal_services=( ["dex"]="dex" ["app"]="app" ) for name in "${!internal_services[@]}"; do service="${internal_services[$name]}" if service_running "$service"; then if docker_compose ps "$service" --format "table {{.Health}}" 2>/dev/null | grep -q "healthy"; then ok "Internal $name service ✓" else warn "Internal $name service ✗ (not healthy)" fi else warn "Internal $name service ✗ (not running)" fi done declare -A external_probes=( ["Auth (HTTPS)"]="curl -fs --connect-timeout 10 https://auth.alertdown.ai/healthz" ["App (HTTPS)"]="curl -fs --connect-timeout 10 https://alertdown.ai/health" ["App (HTTPS/nginx-health)"]="curl -fs --connect-timeout 10 https://alertdown.ai/nginx-health" ["HTTP redirect"]="curl -fs --connect-timeout 5 http://alertdown.ai/ | grep -q '301'" ) for name in "${!external_probes[@]}"; do if timeout 15 bash -c "${external_probes[$name]}" >/dev/null 2>&1; then ok "External $name ✓" else warn "External $name ✗" fi done echo log "Service Status Summary:" docker_compose ps --format "table {{.Service}}\t{{.Status}}\t{{.Health}}" echo -e "\n${c_green}🚀 AlertDown stack deployment complete:" echo -e "${c_green} Auth: https://auth.alertdown.ai${c_nc}" echo -e "${c_green} App: https://alertdown.ai${c_nc}" if ! service_running "app" || ! docker_compose ps "app" --format "table {{.Health}}" 2>/dev/null | grep -q "healthy"; then echo -e "${c_yellow} Note: App service may still be starting up${c_nc}" fi It: Comprobar el contenedor “executando” + “saludable” para dex y app. HTTPS externo + HTTP → HTTPS Redirect. Imprima un docker de una línea para componer la tabla de estado y el banner de éxito amistoso. Observa si la aplicación está comenzando. Beware: Por favor, inspeccione este archivo antes de proceder. Hay otros valores codificados dentro de ellos que necesita actualizar. Hooray 🥳 Horacio With this, you’ll have ¡Con una imagen de Docker extraída del Registro Artifact! Dex Up y corriendo Dex Up y corriendo Autenticación con Postman En el caso de que se encuentren con problemas, siempre puede tratar de usar cURL o Postman para deshabilitar los intercambios de token. Necesitamos realizar 3 intercambios de token Dex Google STS (Servicio de Transferencia de Almacenamiento) Google IAM Credenciales Autenticación contra Dex Lo cubriremos más abajo, en shell podemos generar un script auth como: BASIC_AUTH_HEADER=$(echo -n "$DEX_GCP_STATIC_CLIENT_ID:$DEX_GCP_STATIC_CLIENT_SECRET" | base64 -w 0) Lo cubriremos más abajo, en shell podemos generar un script auth como: BASIC_AUTH_HEADER=$(echo -n "$DEX_GCP_STATIC_CLIENT_ID:$DEX_GCP_STATIC_CLIENT_SECRET" | base64 -w 0) The generated fetch request: const myHeaders = new Headers(); myHeaders.append("Content-Type", "application/x-www-form-urlencoded"); myHeaders.append("Authorization", "Basic Ly9pYW0uZ29vZ2xlYXBpcy5jb20vcHJvamVjdHMvMTA4MzM1MDgwNzQ5OC9sb2NhdGlvbnMvZ2xvYmFsL3dvcmtsb2FkSWRlbnRpdHlQb29scy9oZXR6bmVyLXBvb2wvcHJvdmlkZXJzL2hldHpuZXItcHJvdmlkZXI6YS1zdHJvbmctcGFzc3dvcmQtZ2VuZXJhdGVkLXVzaW5nLWEtcGFzc3dvcmQtZ2VuZXJhdG9y"); const urlencoded = new URLSearchParams(); urlencoded.append("grant_type", "password"); urlencoded.append("username", "auth@mydomain.com"); urlencoded.append("password", "pleaseuseastrongerpassword"); urlencoded.append("scope", "openid email profile groups"); const requestOptions = { method: "POST", headers: myHeaders, body: urlencoded, redirect: "follow" }; fetch("https://auth.yourdomain.com", requestOptions) .then((response) => response.text()) .then((result) => console.log(result)) .catch((error) => console.error(error)); Y ahora en CURL: curl --location 'https://auth.yourdomain.com' \ --header 'Content-Type: application/x-www-form-urlencoded' \ --header 'Authorization: Basic Ly9pYW0uZ29vZ2xlYXBpcy5jb20vcHJvamVjdHMvMTA4MzM1MDgwNzQ5OC9sb2NhdGlvbnMvZ2xvYmFsL3dvcmtsb2FkSWRlbnRpdHlQb29scy9oZXR6bmVyLXBvb2wvcHJvdmlkZXJzL2hldHpuZXItcHJvdmlkZXI6YS1zdHJvbmctcGFzc3dvcmQtZ2VuZXJhdGVkLXVzaW5nLWEtcGFzc3dvcmQtZ2VuZXJhdG9y' \ --data-urlencode 'grant_type=password' \ --data-urlencode 'username=auth@mydomain.com' \ --data-urlencode 'password=pleaseuseastrongerpassword' \ --data-urlencode 'scope=openid email profile groups' Authenticating against Google STS Api: With the token in hand, you need to authenticate against Google STS (Storage Transfer Service) which allows us to move data across cloud providers: : Where ALERTDOWN_AUTH_ID_TOKEN => Es el JWT que vino de Dex. AUDIENCE => //iam.googleapis.com/projects/1016670781645/locations/global/workloadIdentityPools/hetzner-pool/providers/hetzner-provider ¡Nota el alcance! Aquí está el comando fetch: const myHeaders = new Headers(); myHeaders.append("Content-Type", "application/json"); const raw = JSON.stringify({ "grant_type": "urn:ietf:params:oauth:grant-type:token-exchange", "subject_token_type": "urn:ietf:params:oauth:token-type:id_token", "subject_token": "eyJhbGciOiJSUzI1NiIsImtpZCI6IjE1NjdmY2MyMTVhMmQxNGY2MzVhZmY1ZjI5MjNmZDkzMTVkZWNkNzQifQ.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.kyI61WGR5m0h4YvflYV4sCHDY0G7ix7R4m59ITvE_Bq3oIwjbH2NxAMzmtPbUp9kCcsbosAeJTcfJWj2n03-LtRZKc1WjELrFytlnSDgt1KeNCqYYWsdG5eUORzYgvfl9ayNqf7QgDPc3Sr7XQElfk07F-uJPAGPssUXY-qxos6lZHrmComzEWkWqfbuq5e-cvLsBP6TmFAt58B2XKAcSLYSuFrp8eMDaCZ7zQ12z9NR9q0N7u7cVKsJT2429I27fh6LrQsthMaaDMEKzfhY-HskbmcvYO_z4U2M1plYvXAqJRJzGGrkArPSGklxvfFS6gIqLSI7MLnzzsJKpTUt1w", "audience": "//iam.googleapis.com/projects/1083350807498/locations/global/workloadIdentityPools/hetzner-pool/providers/hetzner-provider", "requested_token_type": "urn:ietf:params:oauth:token-type:access_token", "scope": "https://www.googleapis.com/auth/cloud-platform", "options": "{\"serviceAccount\": \"hetzner@alertdown.iam.gserviceaccount.com\"}" }); const requestOptions = { method: "POST", headers: myHeaders, body: raw, redirect: "follow" }; fetch("https://sts.googleapis.com/v1/token", requestOptions) .then((response) => response.text()) .then((result) => console.log(result)) .catch((error) => console.error(error)); Aquí está el comando cURL completo: curl --location 'https://sts.googleapis.com/v1/token' \ --header 'Content-Type: application/json' \ --data-raw '{ "grant_type": "urn:ietf:params:oauth:grant-type:token-exchange", "subject_token_type": "urn:ietf:params:oauth:token-type:id_token", "subject_token": "eyJhbGciOiJSUzI1NiIsImtpZCI6IjE1NjdmY2MyMTVhMmQxNGY2MzVhZmY1ZjI5MjNmZDkzMTVkZWNkNzQifQ.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.kyI61WGR5m0h4YvflYV4sCHDY0G7ix7R4m59ITvE_Bq3oIwjbH2NxAMzmtPbUp9kCcsbosAeJTcfJWj2n03-LtRZKc1WjELrFytlnSDgt1KeNCqYYWsdG5eUORzYgvfl9ayNqf7QgDPc3Sr7XQElfk07F-uJPAGPssUXY-qxos6lZHrmComzEWkWqfbuq5e-cvLsBP6TmFAt58B2XKAcSLYSuFrp8eMDaCZ7zQ12z9NR9q0N7u7cVKsJT2429I27fh6LrQsthMaaDMEKzfhY-HskbmcvYO_z4U2M1plYvXAqJRJzGGrkArPSGklxvfFS6gIqLSI7MLnzzsJKpTUt1w", "audience": "//iam.googleapis.com/projects/1083350807498/locations/global/workloadIdentityPools/hetzner-pool/providers/hetzner-provider", "requested_token_type": "urn:ietf:params:oauth:token-type:access_token", "scope": "https://www.googleapis.com/auth/cloud-platform", "options": "{\"serviceAccount\": \"hetzner@alertdown.iam.gserviceaccount.com\"}" }' Google IAM Credenciales Esta es la última parte! Con esto obtenemos el JWT de corta duración que representa nuestra cuenta de servicio El Fetch: const myHeaders = new Headers(); myHeaders.append("Accept", "application/json"); myHeaders.append("Content-Type", "application/json"); // The bearer is the token we receive from Google STS myHeaders.append("Authorization", "Bearer ya29.d.c0ASRK0GZhTkZi8JGyqpYLR_m7ikJXVUiOL0mOA4jtgydbRyRyNusJu0up5PGOebyIYAGZoKHkreDgaukv7DbNZm3OT6JPVB5HRW_KiqZZyUfgnQRloLrcTPOyf6bDl_yHyzEjVlOy-FYPs98RSt2C-POO3R8NZC-7jEYYzqVlimcNgt7E4QBmeSsxhXShegLCjlDcgSBOob3Kb_-Q-xeTvtkflhYo7jPFiE8ChJEqnIIxL6CzritqgOsrGB4VBNuR0BhghbrH_Sa4q7ELTMclLuRb_PuTJ1YtxW1Ia7az3ENOQSF1gxKyeubBAAvKGhtfEDJYVj3mkkPUjXYUTJHPzwV6HRNF2Rls3XWfPZpiTILbbBp1bPt45Up0-9cY6NMie9b1hL9FmtCHxJOzmxDPWWDhi9KDAsRx0OQkJKJWAtl_jNX7HbjNiMRXPF-WMqR3TQia9DEBsEqWNGSMg67uFCbxbtz0DWVjjQS9VlErmK_h1pwdwhaSMWSW5_Qz-VmZ5E2K2uPdPDMLWgdDjk01b4mVXVsDJGpIj2W-9D3C8yzV5avJcAhuWpDCjOtUTUxo5LKDvjZAFbPBaIa5-LfM54UlIvMnem2y9sgIIVxptmR1gslv0PJMgCyeYDjxlCcvQqv8W5IuFw61O7zIp3FlmU8EwcbMGdfs8PRulgLjep2jL9HhxBo8zAcz1aJgVnV0kHFPfYSTVsBUQcNHYCOBmmp_NkrgKW8Tjs1e7buzZpKz4Op6Up2rqMa69ArwasSkTiQ1cGZxM9_hEt9lt9bCjvlyns9JWcSxdBrRptk4fFXBhQyYbZQ90WW2BBxfTCopa_QKcwAiNqVgKHQZlxuGeVgXVuwpBdNqKMGT29BjuiqGlz_64fmLhPr37ivSjdt6YyCKB1S6-_sjDMFUAC73HtCyaMcF8WsNdbQDgBzvNYfZMg_YnsHkQdClj7EwOMymOna9hOZyKHbtFP89IZ852yzTUqEEGAn3WAaneiBbYQjjB1WtMtSGTTGVmggK65NG1HOjiuI8R7Ik0Dgx4kOSQQYD6fHVLcYPignZIzxm8TZhIcuBPIRKm5FHb9LD69lFtpDkJ8w0vfC2bAMz-Y15u-mla8jIkVTDi_3i1mYmJh8cCm0wwEDCBWEtNOJ5zwNFsYMHQwDso-9nc1cM6P57JXsq7g"); const raw = JSON.stringify({ "scope": [ "https://www.googleapis.com/auth/cloud-platform" ] }); const requestOptions = { method: "POST", headers: myHeaders, body: raw, redirect: "follow" }; fetch("https://iamcredentials.googleapis.com/v1/projects/-/serviceAccounts/hetzner@alertdown.iam.gserviceaccount.com:generateAccessToken", requestOptions) .then((response) => response.text()) .then((result) => console.log(result)) .catch((error) => console.error(error)); La curva: curl --location 'https://iamcredentials.googleapis.com/v1/projects/-/serviceAccounts/hetzner@alertdown.iam.gserviceaccount.com:generateAccessToken' \ --header 'Accept: application/json' \ --header 'Content-Type: application/json' \ --header 'Authorization: Bearer ya29.d.c0ASRK0GZhTkZi8JGyqpYLR_m7ikJXVUiOL0mOA4jtgydbRyRyNusJu0up5PGOebyIYAGZoKHkreDgaukv7DbNZm3OT6JPVB5HRW_KiqZZyUfgnQRloLrcTPOyf6bDl_yHyzEjVlOy-FYPs98RSt2C-POO3R8NZC-7jEYYzqVlimcNgt7E4QBmeSsxhXShegLCjlDcgSBOob3Kb_-Q-xeTvtkflhYo7jPFiE8ChJEqnIIxL6CzritqgOsrGB4VBNuR0BhghbrH_Sa4q7ELTMclLuRb_PuTJ1YtxW1Ia7az3ENOQSF1gxKyeubBAAvKGhtfEDJYVj3mkkPUjXYUTJHPzwV6HRNF2Rls3XWfPZpiTILbbBp1bPt45Up0-9cY6NMie9b1hL9FmtCHxJOzmxDPWWDhi9KDAsRx0OQkJKJWAtl_jNX7HbjNiMRXPF-WMqR3TQia9DEBsEqWNGSMg67uFCbxbtz0DWVjjQS9VlErmK_h1pwdwhaSMWSW5_Qz-VmZ5E2K2uPdPDMLWgdDjk01b4mVXVsDJGpIj2W-9D3C8yzV5avJcAhuWpDCjOtUTUxo5LKDvjZAFbPBaIa5-LfM54UlIvMnem2y9sgIIVxptmR1gslv0PJMgCyeYDjxlCcvQqv8W5IuFw61O7zIp3FlmU8EwcbMGdfs8PRulgLjep2jL9HhxBo8zAcz1aJgVnV0kHFPfYSTVsBUQcNHYCOBmmp_NkrgKW8Tjs1e7buzZpKz4Op6Up2rqMa69ArwasSkTiQ1cGZxM9_hEt9lt9bCjvlyns9JWcSxdBrRptk4fFXBhQyYbZQ90WW2BBxfTCopa_QKcwAiNqVgKHQZlxuGeVgXVuwpBdNqKMGT29BjuiqGlz_64fmLhPr37ivSjdt6YyCKB1S6-_sjDMFUAC73HtCyaMcF8WsNdbQDgBzvNYfZMg_YnsHkQdClj7EwOMymOna9hOZyKHbtFP89IZ852yzTUqEEGAn3WAaneiBbYQjjB1WtMtSGTTGVmggK65NG1HOjiuI8R7Ik0Dgx4kOSQQYD6fHVLcYPignZIzxm8TZhIcuBPIRKm5FHb9LD69lFtpDkJ8w0vfC2bAMz-Y15u-mla8jIkVTDi_3i1mYmJh8cCm0wwEDCBWEtNOJ5zwNFsYMHQwDso-9nc1cM6P57JXsq7g' \ --data '{ "scope": [ "https://www.googleapis.com/auth/cloud-platform" ] }' El token Bearer es el token ya29.d que recibimos de Google STS. The Bearer token is the ya29.d token we receive from Google STS. Emitir un token OIDC de NodeJS Este proceso no es tan complicado como el de Docker, pero todavía es peludo y poco documentado. Sin embargo, todavía requiere cierta configuración. google-auth-library Aquí está cómo: La configuración de la OIDC (la pasamos de Doppler): OIDC_CLIENT_ID="//iam.googleapis.com/projects/1016670781645/locations/global/workloadIdentityPools/hetzner-pool/providers/hetzner-provider" OIDC_CLIENT_SECRET="This is the password that we pass in base64 format in the header" OIDC_ISSUER="https://auth.mydomain.com" OIDC_PASSWORD="This is the unhashed bcrypt password" OIDC_POOL_ID="hetzner-pool" OIDC_PROVIDER_ID="hetzner-provider" OIDC_USERNAME="DEX_GCP_STATIC_PASSWORD_EMAIL -> the email address we used in Dex" Instalar los paquetes necesarios: google-auth-library pnpm add google-auth-library @google-cloud/vertexai Crear un Auth.ts: import { ExternalAccountClientOptions, SubjectTokenSupplier, } from 'google-auth-library'; import { GoogleOIDCFetchException } from './google.auth.exception'; export type OIDCToken = { access_token: string; token_type: 'bearer'; expires_in: number; id_token: string; }; export type OidcConfigParams = { oidcIssuer: string; oidcClientId: string; oidcClientSecret: string; oidcUsername: string; oidcPassword: string; googleProjectNumber: string; oidcPoolId: string; oidcProviderId: string; googleServiceAccount: string; }; class OidcTokenSupplier implements SubjectTokenSupplier { readonly #config: OidcConfigParams; constructor(config: OidcConfigParams) { this.#config = config; } async getSubjectToken( ): Promise { // our Dex URL const tokenUrl = `${this.#config.oidcIssuer}/token`; const basicAuthHeader = Buffer.from(`${this.#config.oidcClientId}:${this.#config.oidcClientSecret}`).toString('base64'); const urlencoded = new URLSearchParams(); urlencoded.append("grant_type", "password"); urlencoded.append("username", this.#config.oidcUsername); urlencoded.append("password", this.#config.oidcPassword); urlencoded.append("scope", "openid email profile groups"); const response = await fetch(tokenUrl, { method: 'POST', headers: { 'Content-Type': 'application/x-www-form-urlencoded', 'Authorization': `Basic ${basicAuthHeader}`, }, body: urlencoded, }); if (!response.ok) { throw new GoogleOIDCFetchException( `Failed to get OIDC token: ${response.status} ${response.statusText}`, { status: response.status, statusText: response.statusText, body: await response.text(), } ); } const data = await response.json() as OIDCToken; return data.id_token; } } export function getOidcConfig( params: OidcConfigParams ): ExternalAccountClientOptions { const audience = `//iam.googleapis.com/projects/${params.googleProjectNumber}/locations/global/workloadIdentityPools/${params.oidcPoolId}/providers/${params.oidcProviderId}`; return { type: 'external_account', audience, subject_token_type: 'urn:ietf:params:oauth:token-type:id_token', token_url: 'https://sts.googleapis.com/v1/token', subject_token_supplier: new OidcTokenSupplier(params), service_account_impersonation_url: `https://iamcredentials.googleapis.com/v1/projects/-/serviceAccounts/${params.googleServiceAccount}:generateAccessToken`, scopes: ['https://www.googleapis.com/auth/cloud-platform'], }; } Heredamos de la SubjectTokenSupplier, y la usamos para apuntar a nuestra configuración Dex. El La función transforma nuestra entrada en la forma que Google Auth Credentials espera de cada servicio de paquete de Google. Esto es similar a la configuración de Postman, con la diferencia de que los endpoints STS e IAM están cocidos. getOidcConfig Configuración de Vertex: Podemos consumirlo en Vertex AI: export class GoogleVertexLLM { #client: VertexAI; #projectId: string; #location: string; constructor(config: GoogleVertexLLMConstructorConfig) { this.#projectId = config.projectId; this.#location = config.location; this.#client = new VertexAI({ project: this.#projectId, location: this.#location, googleAuthOptions: { credentials: getOidcConfig(config.oidcConfig) }, }); } async generateChat ( systemPrompt: string ): PromiseExceptionResult { const parts: Part[] = [ { text: systemPrompt, }, ]; const model = this.#client.getGenerativeModel({ model: 'gemini-2.0-flash', systemInstruction: { role: 'system', parts, }, }); const chat = model.startChat({ generationConfig: { temperature: 0, topP: 0.95, maxOutputTokens: 8192, responseMimeType: 'application/json', }, }); } } ¡Entonces realizaría automáticamente todos los intercambios de tokens para nosotros! Lo hicimos Omgosh. Fue loco. Se tomó un tiempo, pero pudimos conectarnos a los servicios y emitir un token rotativo! Hopefully, it won’t take you long. Si hay una parte que no estaba clara del artículo, déjame saber! Siempre puedes encontrarme en y/o Linkedin Twitter y X Vivianidad Vivianidad