UN-4622 [FIX] Stop publishing the x2text service's port on the host

docker/docker-compose.yaml

@@ -134,8 +134,11 @@ services:
     image: unstract/x2text-service:${VERSION}
     container_name: unstract-x2text-service
     restart: unless-stopped
-    ports:
-      - "3004:3004"
+    # Internal service: not published on the host. Other services reach it on the
+    # compose network as http://unstract-x2text-service:3004.
+    # For local development only, uncomment to expose it on loopback:
+    # ports:
+    #   - "127.0.0.1:3004:3004"
     env_file:
       - ../x2text-service/.env
     depends_on:

docs/local-dev-setup-executor-migration.md

@@ -210,7 +210,7 @@ cp sample.env .env
 
 Then edit `workers/.env` — change all Docker hostnames to `localhost`:
 
-> The runner's port 5002 is not published on the host by default. Uncomment the loopback `ports` entry for `runner` in `docker/docker-compose.yaml` and recreate it (`docker compose up -d runner` from `docker/`), or workers on the host cannot reach it.
+> The runner's port 5002 and the x2text service's port 3004 are not published on the host by default. Uncomment the loopback `ports` entries for `runner` and `x2text-service` in `docker/docker-compose.yaml` then recreate them from `docker/` with the same tag you started the stack with (`VERSION=<tag> docker compose up -d runner x2text-service`), or workers on the host cannot reach them.
 
 ```ini
 # === CRITICAL CHANGES FOR LOCAL DEV ===
@@ -375,7 +375,7 @@ cd workers && ./run-worker.sh all
 | Backend API | 8000 | http://localhost:8000/api/v1/ |
 | Frontend | 3000 | http://frontend.unstract.localhost |
 | Platform Service | 3001 | http://localhost:3001 |
-| X2Text Service | 3004 | http://localhost:3004 |
+| X2Text Service | 3004 | http://localhost:3004 (not published by default; uncomment the loopback `ports` entry for `x2text-service` in `docker/docker-compose.yaml`) |
 | Runner | 5002 | http://localhost:5002 (not published by default; uncomment the loopback `ports` entry for `runner` in `docker/docker-compose.yaml`) |
 | Celery Flower (optional) | 5555 | http://localhost:5555 |
 

tests/README.md

@@ -201,7 +201,7 @@ Three modes behind one protocol, chosen by `--runtime` or `UNSTRACT_E2E_RUNTIME`
 |---|---|---|
 | `compose` | CI; testing the prod image. | `docker compose -f docker/docker-compose.yaml -f tests/compose/docker-compose.test.yaml up -d --wait`, then HTTP. Teardown wipes volumes. |
 | `testcontainers` | Local iteration on infra-only groups. | Stands up Postgres/Redis/RabbitMQ/MinIO via testcontainers and exposes their handles on `PlatformEndpoints.infra`. **Stub today**: does NOT auto-launch backend/prompt-service/etc. as subprocesses — full-platform e2e on testcontainers will need that wiring added. Use `compose` for now if you need the whole stack. |
-| `local` | After `./run-platform.sh`. | Assume a developer-managed stack; read URLs from env. |
+| `local` | After `./run-platform.sh`. | Assume a developer-managed stack; read URLs from env. The smoke gate probes x2text-service at `localhost:3004`, which `docker/docker-compose.yaml` does not publish: uncomment its loopback `ports` entry there, then re-run `./run-platform.sh` with the options you started it with, so the container is recreated with it. |
 
 The rig brings the platform up **once** per `run` invocation (if any selected group has `requires_platform: true`) and exports its URLs via env vars (`UNSTRACT_BACKEND_URL`, `UNSTRACT_PROMPT_SERVICE_URL`, etc.). The rig uses `env.setdefault(...)` so a pre-set value (e.g. from `local` runtime or a developer override) wins over the runtime's default — useful when iterating against a custom stack, but means stale shell env can mask wiring bugs. The smoke test asserts the fixture's URL matches the env var to catch this.
 

tests/compose/docker-compose.test.yaml

@@ -33,6 +33,12 @@ services:
   platform-service:
     image: unstract/platform-service:${UNSTRACT_TEST_VERSION:-latest}
 
+  # Not published by docker/docker-compose.yaml; the rig's readiness probe and
+  # e2e smoke test reach its health endpoint from the host (UNSTRACT_X2TEXT_URL).
+  x2text-service:
+    ports:
+      - "127.0.0.1:3004:3004"
+
   runner:
     image: unstract/runner:${UNSTRACT_TEST_VERSION:-latest}
 

tests/rig/runtime.py

@@ -137,7 +137,9 @@ def from_env(cls, *, infra: InfraEndpoints | None = None) -> PlatformEndpoints:
         """Build a ``PlatformEndpoints`` from ``UNSTRACT_*`` env vars.
 
         Used by every runtime so the lookup logic stays single-sourced.
-        Defaults match the dev compose stack at ``docker/docker-compose.yaml``.
+        Defaults are localhost URLs for a developer-managed stack. Not every
+        one is published by ``docker/docker-compose.yaml`` as shipped; see the
+        commented-out loopback ``ports`` entries there.
         """
         return cls(
             backend_url=os.environ.get("UNSTRACT_BACKEND_URL", "http://localhost:8000"),
@@ -315,6 +317,9 @@ class TestcontainersRuntime:
 
     For full platform e2e against testcontainers, use ``ComposeRuntime`` for now
     or set ``UNSTRACT_E2E_RUNTIME=local`` after running ``run-platform.sh``.
+    With ``local``, x2text-service must be reachable at ``UNSTRACT_X2TEXT_URL``
+    (default ``localhost:3004``), which ``docker/docker-compose.yaml`` does not
+    publish: uncomment its loopback ``ports`` entry there.
     """
 
     name: ClassVar[str] = "testcontainers"

workers/sample.env

@@ -533,7 +533,7 @@ GOOGLE_OAUTH2_SECRET=
 # REDIS_HOST=localhost
 # CACHE_REDIS_HOST=localhost
 # PLATFORM_SERVICE_HOST=http://localhost
-# X2TEXT_HOST=http://localhost
+# X2TEXT_HOST=http://localhost  # also uncomment the x2text-service loopback port in docker/docker-compose.yaml
 # UNSTRACT_RUNNER_HOST=http://localhost  # also uncomment the runner's loopback port in docker/docker-compose.yaml
 # WORKFLOW_EXECUTION_FILE_STORAGE_CREDENTIALS={"provider": "minio", "credentials": {"endpoint_url": "http://localhost:9000", "key": "minio", "secret": "minio123"}}
 # API_FILE_STORAGE_CREDENTIALS={"provider": "minio", "credentials": {"endpoint_url": "http://localhost:9000", "key": "minio", "secret": "minio123"}}