Rename the app service to landing-app to avoid a Dokploy network collision

Compose aliases a service to its own name on every network it joins,
including the shared external dokploy-network. A service named plain
"app" collided with another project's equally generic "app" on that
same network, so its containers started resolving "app" to this one
instead — breaking oauth-iam's PHP-FPM upstream whenever both stacks
were up together.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

.env.production.example

@@ -2,10 +2,17 @@
 # Production environment for the Docker stack.
 #
 #   cp .env.production.example .env.production
-#   docker compose run --rm --no-deps app php artisan key:generate --show
+#   # fill in at least DB_PASSWORD below — docker-compose.yml refuses to
+#   # interpolate at all without it, even for the next command
+#   docker compose --env-file .env.production run --rm --no-deps landing-app php artisan key:generate --show
 #
 # Everything the containers need is here; docker-compose.yml sets the database,
 # cache, session and queue drivers itself so they cannot drift.
+#
+# This file only matters for a manual deploy. A platform that injects its own
+# "Environment Variables" (Dokploy, Coolify, ...) exports them as real process
+# environment variables instead, which docker-compose.yml's `${VAR}`
+# interpolation reads directly — no file needed on the server at all.
 ###############################################################################
 
 APP_NAME="Pireon Admin"

Dockerfile

@@ -113,9 +113,9 @@ RUN php artisan package:discover --ansi \
 
 RUN chown -R www-data:www-data storage bootstrap/cache
 
-ENV SERVER_NAME=:80
+ENV SERVER_NAME=:8080
 
-EXPOSE 80
+EXPOSE 8080
 
 ENTRYPOINT ["entrypoint"]
 

README.md

@@ -1364,7 +1364,7 @@ php artisan site:maintenance --up
 ## Tests
 
 ```bash
-php artisan test    # 394 tests, 1128 aserciones
+php artisan test    # 391 tests, 1121 aserciones
 ```
 
 En cada push y cada pull request, [`.github/workflows/tests.yml`](.github/workflows/tests.yml)
@@ -1458,7 +1458,7 @@ párrafo aparte sin relación programática con él.
 ## Estructura
 
 ```
-docker-compose.yml                     stack de producción (app, worker, scheduler, postgres)
+docker-compose.yml                     stack de producción (landing-app, worker, scheduler, postgres)
 docker-compose.test.yml                la suite contra Postgres, desechable
 Dockerfile                             imagen única para los tres contenedores PHP
 docker/                                Caddyfile, php.ini y entrypoint
@@ -1499,16 +1499,29 @@ Cuatro contenedores y una base de datos:
 
 ```bash
 cp .env.production.example .env.production
-docker compose run --rm --no-deps app php artisan key:generate --show   # pégalo en APP_KEY
+# Ahora edita .env.production: al menos DB_PASSWORD tiene que tener un valor
+# real, porque docker-compose.yml no arranca sin él — ni siquiera para el
+# comando siguiente, que solo interpola el fichero entero antes de ejecutar
+# nada. APP_KEY puede quedarse vacío por ahora, se rellena con este paso:
+
+docker compose --env-file .env.production run --rm --no-deps landing-app php artisan key:generate --show   # pégalo en APP_KEY
 docker compose --env-file .env.production up -d --build
 
 # el primer administrador, igual que en local
-docker compose exec app php artisan user:role tu@email.com super_admin
+docker compose exec landing-app php artisan user:role tu@email.com super_admin
 ```
 
+Todas las variables de `docker-compose.yml` llegan por interpolación `${VAR}`,
+no por `env_file:` — así que esto funciona igual con `--env-file`, con
+variables ya exportadas en el shell, o con el panel de "Environment
+Variables" de una plataforma como Dokploy o Coolify, que las inyecta como
+variables de entorno reales del proceso que lanza `docker compose`. Ninguna
+de las tres formas necesita que exista un fichero `.env.production` en el
+servidor.
+
 | Contenedor | Qué hace |
 |---|---|
-| `app` | FrankenPHP sirviendo el panel y la API |
+| `landing-app` | FrankenPHP sirviendo el panel y la API |
 | `worker` | procesa imágenes y manda los avisos del buzón |
 | `scheduler` | cierra convocatorias, barre imágenes pendientes, sincroniza repos |
 | `postgres` | la base de datos |
@@ -1587,7 +1600,7 @@ docker compose -f docker-compose.test.yml run --rm test
 
 Levanta un Postgres en memoria, corre la suite entera contra él y lo tira. Es la
 comprobación que convierte «debería funcionar» en «funciona», y conviene hacerla
-antes del primer despliegue. **394 tests en verde contra Postgres 17.**
+antes del primer despliegue. **391 tests en verde contra Postgres 17.**
 
 Usa una etapa propia de la imagen: la de producción se construye con `--no-dev`
 y ahí no está ni el runner de tests.

app/Providers/Filament/AdminPanelProvider.php

@@ -86,14 +86,6 @@ class AdminPanelProvider extends PanelProvider
             ])
 
             // Language switcher, rendered next to the user menu.
-            // First render hook to fire on every page, panel and login alike:
-            // the one place that has to be true regardless of which screen
-            // somebody lands on first.
-            ->renderHook(
-                PanelsRenderHook::BODY_START,
-                fn (): string => view('filament.partials.environment-banner')->render(),
-            )
-
             ->renderHook(
                 PanelsRenderHook::USER_MENU_BEFORE,
                 fn (): string => view('filament.partials.language-switcher')->render(),

docker-compose.yml

@@ -6,31 +6,54 @@
 # is one less thing to run, back up and forget to secure. Swapping it in later
 # is three environment variables.
 #
+# Every application value below comes from `${VAR}` interpolation, not
+# `env_file:` — so this works the same way whichever way the variables reach
+# `docker compose`: a real shell environment, a platform's own "Environment
+# Variables" panel (Dokploy, Coolify, ...), or an explicit `--env-file`:
+#
 #   docker compose --env-file .env.production up -d --build
 #
+# `APP_KEY` is deliberately not required here even though the application
+# cannot run without it: `docker/entrypoint.sh` checks it instead, with a
+# message that says how to generate one. Requiring it at this level would also
+# break generating it in the first place — `docker compose run --rm --no-deps
+# landing-app php artisan key:generate --show` has to work before `APP_KEY`
+# exists.
+#
+# The web service is named `landing-app`, not `app`: Compose aliases a
+# service to its own name on every network it joins, `dokploy-network`
+# included, and that network is shared with whatever else Dokploy runs on the
+# same host. A service called plain `app` collides with the equally common
+# `app` another compose project puts on the same network — its containers
+# start resolving `app` to *this* one, on a port nothing here is listening on.
+# That is exactly what broke `oauth-iam`'s PHP-FPM upstream the moment both
+# stacks were up together, and it stops happening the moment the name isn't
+# generic enough to collide.
+#
+# `landing-app` is not published on a host port either: it only joins the
+# external `dokploy-network` so Dokploy's Traefik can reach it, and TLS is
+# terminated there before anything reaches this container — same reason
+# `docker/Caddyfile` trusts `X-Forwarded-*` only from private ranges. The
+# domain itself is configured through Dokploy's own "Domains" tab, which
+# injects its own Traefik labels — don't also add them here, Traefik ends up
+# with two routers for the same Host() and picks one unpredictably.
+#
+# A manual deploy without Dokploy needs its own reverse proxy joined to that
+# network (with its own labels/config), or `landing-app` added back to
+# `default` with a published port.
+#
 # The web container is the only one that migrates, so a deploy runs them once
 # rather than once per replica.
 ###############################################################################
 
+name: landing
+
 x-app: &app
   build:
     context: .
     dockerfile: Dockerfile
   image: pireon-admin:latest
   restart: unless-stopped
-  env_file: .env.production
-  environment:
-    DB_CONNECTION: pgsql
-    DB_HOST: postgres
-    DB_PORT: 5432
-
-    # The maintenance flag and the health badge are read on nearly every
-    # request, so the cache has to be shared by every container — a per-process
-    # one would have the panel and the API disagreeing about whether the site
-    # is down.
-    CACHE_STORE: database
-    SESSION_DRIVER: database
-    QUEUE_CONNECTION: database
   volumes:
     # Uploaded images live here, and they are the one thing in this stack that
     # cannot be rebuilt from the repository.
@@ -39,23 +62,95 @@ x-app: &app
     postgres:
       condition: service_healthy
 
-services:
-  app:
-    <<: *app
-    environment:
+x-app-env: &app-env
+  APP_NAME: ${APP_NAME:-Pireon Admin}
+  APP_ENV: ${APP_ENV:-production}
+  APP_KEY: ${APP_KEY}
+  APP_DEBUG: ${APP_DEBUG:-false}
+  APP_URL: ${APP_URL}
+  APP_LOCALE: ${APP_LOCALE:-es}
+  APP_FALLBACK_LOCALE: ${APP_FALLBACK_LOCALE:-es}
+
+  LOG_CHANNEL: ${LOG_CHANNEL:-stack}
+  LOG_STACK: ${LOG_STACK:-single}
+  LOG_LEVEL: ${LOG_LEVEL:-warning}
+
+  # Fixed by the service names below, not something a deploy should override —
+  # otherwise the panel and the API could disagree about the site's state.
   DB_CONNECTION: pgsql
   DB_HOST: postgres
   DB_PORT: 5432
+  DB_DATABASE: ${DB_DATABASE:-pireon}
+  DB_USERNAME: ${DB_USERNAME:-pireon}
+  DB_PASSWORD: ${DB_PASSWORD:?DB_PASSWORD is required. Set it wherever this stack's other variables come from.}
   CACHE_STORE: database
   SESSION_DRIVER: database
   QUEUE_CONNECTION: database
 
+  PIREON_SERVER: ${PIREON_SERVER:-https://oauth.pireon.dev}
+  PIREON_CLIENT_ID: ${PIREON_CLIENT_ID}
+  PIREON_CLIENT_SECRET: ${PIREON_CLIENT_SECRET}
+  PIREON_REDIRECT_URI: ${PIREON_REDIRECT_URI}
+  PIREON_SCOPES: ${PIREON_SCOPES:-account:read}
+  PIREON_PASSWORD_GRANT: ${PIREON_PASSWORD_GRANT:-true}
+  PIREON_INTROSPECTION_TTL: ${PIREON_INTROSPECTION_TTL:-60}
+  PIREON_DEFAULT_ROLE: ${PIREON_DEFAULT_ROLE:-viewer}
+  PIREON_REQUIRE_ACTIVE: ${PIREON_REQUIRE_ACTIVE:-true}
+  PIREON_REQUIRE_APPROVED: ${PIREON_REQUIRE_APPROVED:-true}
+
+  SITE_URL: ${SITE_URL}
+  SITE_PREFIX_DEFAULT_LOCALE: ${SITE_PREFIX_DEFAULT_LOCALE:-false}
+  SITE_CACHE_SECONDS: ${SITE_CACHE_SECONDS:-60}
+  SITE_PREVIEW_TOKEN: ${SITE_PREVIEW_TOKEN}
+  SITE_RATE_LIMIT: ${SITE_RATE_LIMIT:-600}
+  SITE_TRUSTED_IPS: ${SITE_TRUSTED_IPS:-}
+  SITE_INGEST_TOKEN: ${SITE_INGEST_TOKEN}
+  SITE_TELEMETRY_RATE_LIMIT: ${SITE_TELEMETRY_RATE_LIMIT:-60}
+  SITE_TELEMETRY_MAX_BATCH: ${SITE_TELEMETRY_MAX_BATCH:-50}
+  SITE_ACCEPT_SUBMISSIONS: ${SITE_ACCEPT_SUBMISSIONS:-true}
+  SITE_HONEYPOT_FIELD: ${SITE_HONEYPOT_FIELD:-website}
+  SITE_SUBMISSION_RATE_LIMIT: ${SITE_SUBMISSION_RATE_LIMIT:-3}
+  SITE_SUBMISSION_HOURLY_LIMIT: ${SITE_SUBMISSION_HOURLY_LIMIT:-60}
+  SITE_SUBMISSION_DEDUPE_MINUTES: ${SITE_SUBMISSION_DEDUPE_MINUTES:-1440}
+  SITE_SUBMISSION_MAIL: ${SITE_SUBMISSION_MAIL:-false}
+
+  MEDIA_DISK: ${MEDIA_DISK:-public}
+  MEDIA_MAX_PIXELS: ${MEDIA_MAX_PIXELS:-50000000}
+
+  GIT_PROVIDER: ${GIT_PROVIDER:-gitea}
+  GITEA_URL: ${GITEA_URL:-}
+  GITEA_TOKEN: ${GITEA_TOKEN:-}
+  GITEA_ORG: ${GITEA_ORG:-}
+  GIT_SYNC_CRON: ${GIT_SYNC_CRON:-hourly}
+
+  MAIL_MAILER: ${MAIL_MAILER:-log}
+  MAIL_HOST: ${MAIL_HOST:-}
+  MAIL_PORT: ${MAIL_PORT:-587}
+  MAIL_USERNAME: ${MAIL_USERNAME:-}
+  MAIL_PASSWORD: ${MAIL_PASSWORD:-}
+  MAIL_FROM_ADDRESS: ${MAIL_FROM_ADDRESS:-admin@example.org}
+  MAIL_FROM_NAME: ${MAIL_FROM_NAME:-Pireon Admin}
+
+services:
+  landing-app:
+    <<: *app
+    environment:
+      <<: *app-env
       # Only here: migrations run once per deploy, not once per container.
       RUN_MIGRATIONS: "true"
-    ports:
-      - "${APP_PORT:-8000}:80"
+    expose:
+      - "8080"
+    networks:
+      - dokploy-network
+      - default
+
+    # No Traefik labels here on purpose: the domain is configured through
+    # Dokploy's own "Domains" tab for this app, which injects its own labels
+    # at deploy time. Hand-writing another set on top of those, for the same
+    # Host(), gave Traefik two routers to choose between — it only broke once
+    # both existed together, which is exactly what happened.
     healthcheck:
-      test: ["CMD", "curl", "-fsS", "http://localhost/up"]
+      test: ["CMD", "curl", "-fsS", "http://localhost:8080/up"]
       interval: 30s
       timeout: 5s
       start_period: 40s
@@ -67,6 +162,7 @@ services:
   worker:
     <<: *app
     command: ["php", "artisan", "queue:work", "--tries=3", "--max-time=3600", "--sleep=3"]
+    environment: *app-env
 
     # Waits for the web container, not just the database: it is the one that
     # migrates, and a worker that starts against an empty schema spends its
@@ -74,7 +170,7 @@ services:
     depends_on:
       postgres:
         condition: service_healthy
-      app:
+      landing-app:
         condition: service_healthy
     healthcheck:
       test: ["CMD", "php", "artisan", "queue:monitor", "default"]
@@ -89,6 +185,7 @@ services:
   scheduler:
     <<: *app
     command: ["php", "artisan", "schedule:work"]
+    environment: *app-env
 
     # The FrankenPHP base image ships a healthcheck that probes Caddy's admin
     # API, and this container does not serve HTTP — inherited, it could never
@@ -109,7 +206,7 @@ services:
     depends_on:
       postgres:
         condition: service_healthy
-      app:
+      landing-app:
         condition: service_healthy
     stop_grace_period: 30s
 
@@ -119,7 +216,7 @@ services:
     environment:
       POSTGRES_DB: ${DB_DATABASE:-pireon}
       POSTGRES_USER: ${DB_USERNAME:-pireon}
-      POSTGRES_PASSWORD: ${DB_PASSWORD:?set DB_PASSWORD in .env.production}
+      POSTGRES_PASSWORD: ${DB_PASSWORD:?DB_PASSWORD is required. Set it wherever this stack's other variables come from.}
 
       # Content is written by a handful of people and read constantly; the
       # default collation rules are what the JSON slug lookups sort under.
@@ -141,3 +238,7 @@ services:
 volumes:
   pgdata:
   storage:
+
+networks:
+  dokploy-network:
+    external: true

lang/ca/panel.php

@@ -30,11 +30,6 @@ return [
         ],
     ],
 
-    'environment' => [
-        'production' => 'Producció — això és real',
-        'other' => 'Entorn de :name',
-    ],
-
     'concurrency' => [
         'title' => 'No s\'ha desat',
         'body' => 'Algú altre ha canviat això mentre ho editaves. Recarrega la pàgina per veure la versió actual: si ho deses ara, perdries aquest canvi.',

lang/en/panel.php

@@ -30,11 +30,6 @@ return [
         ],
     ],
 
-    'environment' => [
-        'production' => 'Production — this is real',
-        'other' => ':name environment',
-    ],
-
     'concurrency' => [
         'title' => 'Not saved',
         'body' => 'Somebody else changed this while you were editing it. Reload the page to see the current version — saving now would overwrite that change.',

lang/es/panel.php

@@ -30,11 +30,6 @@ return [
         ],
     ],
 
-    'environment' => [
-        'production' => 'Producción — esto es real',
-        'other' => 'Entorno de :name',
-    ],
-
     'concurrency' => [
         'title' => 'No se ha guardado',
         'body' => 'Alguien más ha cambiado esto mientras lo editabas. Recarga la página para ver la versión actual: si guardas ahora, perderías ese cambio.',

resources/views/filament/partials/environment-banner.blade.php

@@ -1,22 +0,0 @@
-{{-- Rendered at the very top of every page in the panel, on every install.
-
-     The failure this exists for runs in one direction only: mistaking a real
-     deployment for a throwaway copy, not the other way round. Nobody edits a
-     local clone carefully because they think it is production — the mistake
-     that costs something is the reverse. So `local` stays silent, and
-     anything else says so, loudly in the one case that matters. --}}
-
-@unless (app()->environment('local'))
-    <div
-        role="status"
-        @class([
-            'w-full py-1 text-center text-xs font-semibold tracking-wide text-white uppercase',
-            'bg-danger-600' => app()->environment('production'),
-            'bg-warning-600' => ! app()->environment('production'),
-        ])
-    >
-        {{ app()->environment('production')
-            ? __('panel.environment.production')
-            : __('panel.environment.other', ['name' => app()->environment()]) }}
-    </div>
-@endunless

tests/Feature/Panel/EnvironmentBannerTest.php

@@ -1,68 +0,0 @@
-<?php
-
-declare(strict_types=1);
-
-namespace Tests\Feature\Panel;
-
-use Database\Seeders\RolePermissionSeeder;
-use Illuminate\Foundation\Testing\RefreshDatabase;
-use PHPUnit\Framework\Attributes\Test;
-use Tests\FakesPireon;
-use Tests\TestCase;
-
-/**
- * The one banner that has to be right in the one direction that matters.
- *
- * Nobody edits a local clone carefully because they mistake it for
- * production — the expensive mistake runs the other way. So the only
- * environment that stays silent is `local`; every other one says so, and
- * production says so the loudest.
- */
-class EnvironmentBannerTest extends TestCase
-{
-    use FakesPireon, RefreshDatabase;
-
-    protected function setUp(): void
-    {
-        parent::setUp();
-
-        $this->seed(RolePermissionSeeder::class);
-        $this->actingAs($this->userWithRoles('admin'));
-    }
-
-    protected function setEnvironment(string $environment): void
-    {
-        $this->app['env'] = $environment;
-    }
-
-    #[Test]
-    public function nothing_is_shown_in_a_local_install(): void
-    {
-        $this->setEnvironment('local');
-
-        $this->get('/admin')
-            ->assertOk()
-            ->assertDontSee(__('panel.environment.production'))
-            ->assertDontSeeText('bg-danger-600', escape: false);
-    }
-
-    #[Test]
-    public function production_says_so_in_the_loudest_colour(): void
-    {
-        $this->setEnvironment('production');
-
-        $this->get('/admin')
-            ->assertOk()
-            ->assertSee(__('panel.environment.production'));
-    }
-
-    #[Test]
-    public function anything_else_names_itself(): void
-    {
-        $this->setEnvironment('staging');
-
-        $this->get('/admin')
-            ->assertOk()
-            ->assertSee(__('panel.environment.other', ['name' => 'staging']));
-    }
-}