dtf-system/docker-compose.yml

version: "3.8"

x-app-environment: &app-environment
  APP_ENV: production
  DATABASE_HOST: db
  DATABASE_NAME: dtf
  DATABASE_USER: dtf_app
  # Must differ from POSTGRES_PASSWORD: the runtime role is DML-only, and reusing
  # the administrator credential would make that restriction meaningless.
  DATABASE_PASSWORD: ${APP_DB_PASSWORD:?set APP_DB_PASSWORD}
  S3_ENDPOINT: ${R2_ENDPOINT:?set R2_ENDPOINT}
  S3_PUBLIC_ENDPOINT: ${R2_ENDPOINT:?set R2_ENDPOINT}
  S3_BUCKET: ${R2_BUCKET:?set R2_BUCKET}
  AWS_ACCESS_KEY_ID: ${R2_ACCESS_KEY_ID:?set R2_ACCESS_KEY_ID}
  AWS_SECRET_ACCESS_KEY: ${R2_SECRET_ACCESS_KEY:?set R2_SECRET_ACCESS_KEY}
  AWS_DEFAULT_REGION: auto
  # Optional at deploy time so a missing Kanban credential cannot make the
  # entire Swarm stack invalid. The Kanban login endpoint fails closed until
  # this value is configured.
  OPERATOR_EMAIL: ${OPERATOR_EMAIL:-}
  OPERATOR_PASSWORD: ${OPERATOR_PASSWORD:?set OPERATOR_PASSWORD}
  # fake until Mercado Pago is configured; mercadopago requires MP_ACCESS_TOKEN
  # and MP_WEBHOOK_SECRET or the API and worker refuse to start. Test
  # credentials (TEST-...) until the sandbox flows have passed. The card form
  # appears only with MP_PUBLIC_KEY, and then needs PAYMENT_CSP_SOURCES too.
  PAYMENT_ADAPTER: ${PAYMENT_ADAPTER:-fake}
  MP_ACCESS_TOKEN: ${MP_ACCESS_TOKEN:-}
  # The "assinatura secreta" from the webhook settings in Mercado Pago.
  MP_WEBHOOK_SECRET: ${MP_WEBHOOK_SECRET:-}
  # https://<SITE_DOMAIN>/api/payments/webhook, sent with every payment.
  MP_NOTIFICATION_URL: ${MP_NOTIFICATION_URL:-}
  MP_PUBLIC_KEY: ${MP_PUBLIC_KEY:-}
  # The fake adapter's secret. Optional: without it the webhook verifies
  # nothing and therefore accepts nothing, which is the correct state until a
  # provider is connected. Never set it to a value anyone could guess.
  PAYMENT_WEBHOOK_SECRET: ${PAYMENT_WEBHOOK_SECRET:-}
  # fake offers pickup only. jadlog prices delivery with Jadlog and needs the
  # credentials below and the package weight from the client; it refuses to
  # start without them. The credentials alone are enough for the console
  # check, python -m app.jadlog_probe, on the worker.
  FREIGHT_ADAPTER: ${FREIGHT_ADAPTER:-fake}
  JADLOG_TOKEN: ${JADLOG_TOKEN:-}
  # The "Usuário" Jadlog issued: the CNPJ that contracts the freight.
  JADLOG_CNPJ: ${JADLOG_CNPJ:-}
  JADLOG_CONTA: ${JADLOG_CONTA:-}
  JADLOG_CONTRATO: ${JADLOG_CONTRATO:-}
  # Package weight in kg: a base plus each billed metre of film.
  JADLOG_PESO_BASE_KG: ${JADLOG_PESO_BASE_KG:-}
  JADLOG_PESO_POR_METRO_KG: ${JADLOG_PESO_POR_METRO_KG:-}
  # Working days of production added to Jadlog's delivery time.
  FREIGHT_PRODUCTION_DAYS: ${FREIGHT_PRODUCTION_DAYS:-0}
  # A cart the Site priced is approved at checkout and can be paid at once;
  # orders above QUOTE_AUTO_MAX_METRES, or with a grade the Site did not
  # compute, wait for an operator on the Kanban (app/quote_review.py).
  QUOTE_AUTO_APPROVE: ${QUOTE_AUTO_APPROVE:-true}
  QUOTE_AUTO_MAX_METRES: ${QUOTE_AUTO_MAX_METRES:-50}
  # Order creation in Tiny stays off until it has been tested against the
  # client's account (Tiny has no sandbox). The application credentials can be
  # set now: they let an operator connect Tiny from the Kanban, and the worker
  # keeps that connection alive. Callback: https://<KANBAN_DOMAIN>/api/operator/tiny/callback
  TINY_ADAPTER: fake
  TINY_CLIENT_ID: ${TINY_CLIENT_ID:-}
  TINY_CLIENT_SECRET: ${TINY_CLIENT_SECRET:-}
  TINY_REDIRECT_URI: ${TINY_REDIRECT_URI:-}
  TINY_PRODUCT_TEXTIL_FOLHA: ${TINY_PRODUCT_TEXTIL_FOLHA:-}
  TINY_PRODUCT_TEXTIL_AVULSA: ${TINY_PRODUCT_TEXTIL_AVULSA:-}
  TINY_PRODUCT_UV_FOLHA: ${TINY_PRODUCT_UV_FOLHA:-}
  TINY_PRODUCT_UV_AVULSA: ${TINY_PRODUCT_UV_AVULSA:-}
  # The client's "retirar pessoalmente" forma de envio id, on pickup orders.
  TINY_FORMA_ENVIO_RETIRADA: ${TINY_FORMA_ENVIO_RETIRADA:-}
  # Sets "Aprovada" on paid orders and "Pronto para envio" on finished pickup
  # orders, which the client's Tiny -> n8n notices send to customers. Turn on
  # only together with TINY_ADAPTER=tiny and after n8n stops sending the
  # designer message for DTFIMP products.
  TINY_STATUS_UPDATES: ${TINY_STATUS_UPDATES:-false}
  WHATSAPP_ADAPTER: fake
  STORAGE_ADAPTER: s3-r2
  PUBLIC_ORIGIN: https://${SITE_DOMAIN:?set SITE_DOMAIN}
  PUBLIC_HOST: ${SITE_DOMAIN:?set SITE_DOMAIN}
  ALLOWED_HOSTS: ${SITE_DOMAIN:?set SITE_DOMAIN},${KANBAN_DOMAIN:?set KANBAN_DOMAIN}
  ALLOWED_ORIGINS: https://${SITE_DOMAIN:?set SITE_DOMAIN},https://${KANBAN_DOMAIN:?set KANBAN_DOMAIN}
  COOKIE_SECURE: "true"
  MAX_UPLOAD_BYTES: "5368709120"
  UPLOAD_PART_BYTES: "8388608"
  # Sheets of several GB are the normal order, so room for many of them.
  STORAGE_QUOTA_BYTES: "${STORAGE_QUOTA_BYTES:-536870912000}"
  OWNER_UPLOAD_QUOTA_BYTES: "${OWNER_UPLOAD_QUOTA_BYTES:-53687091200}"
  MAX_PENDING_UPLOADS: "10"
  # ClamAV scans up to this; larger files (up to MAX_UPLOAD_BYTES) are released
  # after a file-format check instead (app/scanning.py).
  SCAN_MAX_BYTES: "2097152000"

services:
  db:
    image: postgres:17-alpine
    environment:
      POSTGRES_DB: dtf
      POSTGRES_USER: dtf_admin
      POSTGRES_PASSWORD: ${POSTGRES_PASSWORD:?set POSTGRES_PASSWORD}
    volumes: [postgres-data:/var/lib/postgresql/data]
    networks: [backend]
    healthcheck:
      test: [CMD-SHELL, 'pg_isready -U "$$POSTGRES_USER" -d "$$POSTGRES_DB"']
      interval: 10s
      timeout: 5s
      retries: 12
      start_period: 20s
    deploy:
      replicas: 1
      restart_policy: {condition: on-failure, delay: 10s}

  db-init:
    image: ${API_IMAGE:-gitea.blyzer.com.br/blyzer/dtf-api}:${IMAGE_TAG:-latest}
    command: python -m app.bootstrap
    environment:
      DATABASE_ADMIN_HOST: db
      DATABASE_ADMIN_NAME: dtf
      DATABASE_ADMIN_USER: dtf_admin
      DATABASE_ADMIN_PASSWORD: ${POSTGRES_PASSWORD:?set POSTGRES_PASSWORD}
      APP_DB_USER: dtf_app
      APP_DB_PASSWORD: ${APP_DB_PASSWORD:?set APP_DB_PASSWORD}
      # The migration job seeds the first operator account from these, so an
      # existing deployment keeps its Kanban login after the accounts table
      # lands. Without them there would be no account at all and login would
      # fail closed with 503.
      OPERATOR_EMAIL: ${OPERATOR_EMAIL:-}
      OPERATOR_PASSWORD: ${OPERATOR_PASSWORD:?set OPERATOR_PASSWORD}
    networks: [backend]
    deploy:
      replicas: 1
      restart_policy: {condition: on-failure, delay: 5s, max_attempts: 20}

  scanner:
    image: clamav/clamav@sha256:9cb27d7660bdf66e9878c832cb433dd8aa152cfbe16f3c2c0084c80b04ae22b4
    entrypoint: [clamd, --foreground=true, --config-file=/etc/clamav/clamd.conf]
    configs:
      - source: clamd_config_2gb
        target: /etc/clamav/clamd.conf
        mode: 0444
    networks: [backend]
    healthcheck:
      test: [CMD, clamdscan, --config-file=/etc/clamav/clamd.conf, --ping, "3"]
      interval: 15s
      timeout: 5s
      retries: 20
      start_period: 90s
    deploy:
      replicas: 1
      restart_policy: {condition: on-failure, delay: 10s}
      resources:
        limits: {memory: 3G}

  api:
    image: ${API_IMAGE:-gitea.blyzer.com.br/blyzer/dtf-api}:${IMAGE_TAG:-latest}
    environment: *app-environment
    networks: [backend, egress]
    read_only: true
    tmpfs: [/tmp]
    healthcheck:
      test: [CMD-SHELL, 'python -c "import os,urllib.request; r=urllib.request.Request(\"http://localhost:8000/health\",headers={\"Host\":os.environ[\"PUBLIC_HOST\"]}); urllib.request.urlopen(r,timeout=3)"']
      interval: 15s
      timeout: 5s
      retries: 12
      start_period: 30s
    deploy:
      replicas: 1
      restart_policy: {condition: on-failure, delay: 5s}

  worker:
    image: ${API_IMAGE:-gitea.blyzer.com.br/blyzer/dtf-api}:${IMAGE_TAG:-latest}
    command: python -m app.worker
    environment: *app-environment
    networks: [backend, egress]
    read_only: true
    tmpfs: [/tmp]
    healthcheck:
      test: [CMD, python, -c, "import urllib.request; urllib.request.urlopen('http://localhost:8002/health',timeout=3)"]
      interval: 15s
      timeout: 5s
      retries: 12
      start_period: 90s
    deploy:
      replicas: 1
      restart_policy: {condition: on-failure, delay: 5s}

  # Daily encrypted copy of the database to a bucket of its own, off this
  # server (ops/db_backup.py). Idle until the BACKUP_* settings are set. The
  # bucket's lifecycle rule removes old copies; this credential never deletes.
  backup:
    image: ${API_IMAGE:-gitea.blyzer.com.br/blyzer/dtf-api}:${IMAGE_TAG:-latest}
    command: python -m ops.db_backup serve
    environment:
      DATABASE_ADMIN_HOST: db
      DATABASE_ADMIN_NAME: dtf
      DATABASE_ADMIN_USER: dtf_admin
      DATABASE_ADMIN_PASSWORD: ${POSTGRES_PASSWORD:?set POSTGRES_PASSWORD}
      # The R2 account endpoint (the same as R2_ENDPOINT) and a bucket and API
      # token for backups only, never the artwork bucket's.
      BACKUP_S3_ENDPOINT: ${BACKUP_S3_ENDPOINT:-}
      BACKUP_BUCKET: ${BACKUP_BUCKET:-}
      BACKUP_ACCESS_KEY_ID: ${BACKUP_ACCESS_KEY_ID:-}
      BACKUP_SECRET_ACCESS_KEY: ${BACKUP_SECRET_ACCESS_KEY:-}
      # The public key (age1...). Its private key stays off this server.
      BACKUP_AGE_RECIPIENT: ${BACKUP_AGE_RECIPIENT:-}
      # Hour of day in Brasília.
      BACKUP_HOUR: ${BACKUP_HOUR:-3}
    networks: [backend, egress]
    read_only: true
    tmpfs: [/tmp]
    deploy:
      replicas: 1
      restart_policy: {condition: on-failure, delay: 30s}

  site:
    image: ${WEB_IMAGE:-gitea.blyzer.com.br/blyzer/dtf-web}:${IMAGE_TAG:-latest}
    environment:
      WEB_INDEX: index.html
      PUBLIC_HOST: ${SITE_DOMAIN:?set SITE_DOMAIN}
      S3_PUBLIC_ENDPOINT: ${R2_ENDPOINT:?set R2_ENDPOINT}
      # Mercado Pago's card form loads from these origins; empty keeps the
      # Site at script-src 'self'. Set together with the Mercado Pago adapter.
      PAYMENT_CSP_SOURCES: ${PAYMENT_CSP_SOURCES:-}
      # The bank's confirmation page for debit and other 3-D Secure cards is
      # on the issuer's own domain, so it cannot be listed: "https:" lets
      # frames and form posts reach it (never scripts). Empty turns it off.
      PAYMENT_CHALLENGE_SOURCES: ${PAYMENT_CHALLENGE_SOURCES:-}
    networks: [backend]
    ports:
      - target: 8080
        published: ${SITE_PORT:-18080}
        protocol: tcp
        mode: ingress
    healthcheck:
      test: [CMD-SHELL, 'wget -q --header="Host: $$PUBLIC_HOST" -O /dev/null http://127.0.0.1:8080/health']
      interval: 15s
      timeout: 5s
      retries: 12
      start_period: 15s
    deploy:
      replicas: 1
      restart_policy: {condition: on-failure, delay: 5s}

  kanban:
    image: ${WEB_IMAGE:-gitea.blyzer.com.br/blyzer/dtf-web}:${IMAGE_TAG:-latest}
    environment:
      WEB_INDEX: kanban.html
      PUBLIC_HOST: ${KANBAN_DOMAIN:?set KANBAN_DOMAIN}
      S3_PUBLIC_ENDPOINT: ${R2_ENDPOINT:?set R2_ENDPOINT}
      PAYMENT_CSP_SOURCES: ""
      PAYMENT_CHALLENGE_SOURCES: ""
    networks: [backend]
    ports:
      - target: 8080
        published: ${KANBAN_PORT:-18081}
        protocol: tcp
        mode: ingress
    healthcheck:
      test: [CMD-SHELL, 'wget -q --header="Host: $$PUBLIC_HOST" -O /dev/null http://127.0.0.1:8080/health']
      interval: 15s
      timeout: 5s
      retries: 12
      start_period: 15s
    deploy:
      replicas: 1
      restart_policy: {condition: on-failure, delay: 5s}

configs:
  # Renamed whenever infra/clamd.conf changes: Swarm cannot update a deployed
  # config in place, and a redeploy with new content under the old name fails.
  clamd_config_2gb:
    file: ./infra/clamd.conf

volumes:
  postgres-data:

networks:
  backend:
    driver: overlay
    internal: true
  egress:
    driver: overlay