Skip to content

armgit / dify

Public Imported from GitHub
0

Build Agentic workflows, RAG pipelines, with rich AI model and tool support on one collaborative workspace. Deploy on cloud, VPC, or self-hosted, so teams move from prototype to production without rebuilding the stack.

refactor(api): adopt with_session in audio controllers (#37403) (#43322)

KenXiong30eadcc9e536
diff --git a/api/controllers/service_api/app/audio.py b/api/controllers/service_api/app/audio.py
index 2db10dcf42b..01adcbda6d4 100644
--- a/api/controllers/service_api/app/audio.py
+++ b/api/controllers/service_api/app/audio.py
@@ -2,12 +2,14 @@ import logging
from flask import request
from flask_restx import Resource
+from sqlalchemy.orm import Session
from werkzeug.exceptions import InternalServerError
import services
from controllers.common.controller_schemas import TextToAudioPayload
from controllers.common.fields import AudioBinaryResponse, AudioTranscriptResponse
from controllers.common.schema import register_response_schema_models, register_schema_model
+from controllers.common.session import with_session
from controllers.console.wraps import model_validate
from controllers.service_api import service_api_ns
from controllers.service_api.app.error import (
@@ -26,7 +28,6 @@ from controllers.service_api.schema import binary_response, expect_with_user, mu
from controllers.service_api.wraps import FetchUserArg, WhereisUserArg, validate_app_token
from core.base.tts.audio_mime import SUPPORTED_TTS_AUDIO_MIME_TYPES
from core.errors.error import ModelCurrentlyNotSupportError, ProviderTokenNotInitError, QuotaExceededError
-from extensions.ext_database import db
from graphon.model_runtime.errors.invoke import InvokeError
from libs.helper import dump_response
from models.model import App, EndUser
@@ -98,7 +99,8 @@ class AudioApi(Resource):
service_api_ns.models[AudioTranscriptResponse.__name__],
)
@validate_app_token(fetch_user_arg=FetchUserArg(fetch_from=WhereisUserArg.FORM))
- def post(self, app_model: App, end_user: EndUser):
+ @with_session(write=False)
+ def post(self, session: Session, app_model: App, end_user: EndUser):
"""Convert audio to text using speech-to-text.
Accepts an audio file upload and returns the transcribed text.
@@ -109,7 +111,7 @@ class AudioApi(Resource):
response = AudioService.transcript_asr(
app_model=app_model,
file=file,
- session=db.session(),
+ session=session,
end_user=end_user.id,
)
@@ -182,8 +184,9 @@ class TextApi(Resource):
# TTS returns provider audio bytes, so the success response is intentionally schema-less.
@service_api_ns.response(200, "Text successfully converted to audio")
@validate_app_token(fetch_user_arg=FetchUserArg(fetch_from=WhereisUserArg.JSON))
+ @with_session(write=False)
@model_validate(TextToAudioPayload)
- def post(self, payload: TextToAudioPayload, app_model: App, end_user: EndUser):
+ def post(self, payload: TextToAudioPayload, session: Session, app_model: App, end_user: EndUser):
"""Convert text to audio using text-to-speech.
Converts the provided text to audio using the specified voice.
@@ -202,7 +205,7 @@ class TextApi(Resource):
)
return AudioService.transcript_tts(
app_model=app_model,
- session=db.session(),
+ session=session,
text=text,
voice=voice,
end_user=end_user.external_user_id,
diff --git a/api/controllers/web/audio.py b/api/controllers/web/audio.py
index 9a5dc0e6f4b..cbb058e5429 100644
--- a/api/controllers/web/audio.py
+++ b/api/controllers/web/audio.py
@@ -2,10 +2,12 @@ import logging
from flask import request
from pydantic import field_validator
+from sqlalchemy.orm import Session
from werkzeug.exceptions import InternalServerError
import services
from controllers.common.controller_schemas import TextToAudioPayload as TextToAudioPayloadBase
+from controllers.common.session import with_session
from controllers.console.wraps import model_validate
from controllers.web import web_ns
from controllers.web.error import (
@@ -22,7 +24,6 @@ from controllers.web.error import (
)
from controllers.web.wraps import WebApiResource
from core.errors.error import ModelCurrentlyNotSupportError, ProviderTokenNotInitError, QuotaExceededError
-from extensions.ext_database import db
from fields.base import ResponseModel
from graphon.model_runtime.errors.invoke import InvokeError
from libs.helper import dump_response, uuid_value
@@ -75,7 +76,8 @@ class AudioApi(WebApiResource):
}
)
@web_ns.response(200, "Success", web_ns.models[AudioToTextResponse.__name__])
- def post(self, app_model: App, end_user: EndUser):
+ @with_session(write=False)
+ def post(self, session: Session, app_model: App, end_user: EndUser):
"""Convert audio to text"""
file = request.files.get("file")
@@ -83,7 +85,7 @@ class AudioApi(WebApiResource):
response = AudioService.transcript_asr(
app_model=app_model,
file=file,
- session=db.session(),
+ session=session,
end_user=end_user.external_user_id,
)
@@ -132,8 +134,9 @@ class TextApi(WebApiResource):
)
# response-contract:ignore provider audio bytes; TODO: model binary audio response if shape is standardized.
@web_ns.response(200, "Success")
+ @with_session(write=False)
@model_validate(TextToAudioPayload)
- def post(self, payload: TextToAudioPayload, app_model: App, end_user: EndUser):
+ def post(self, payload: TextToAudioPayload, session: Session, app_model: App, end_user: EndUser):
"""Convert text to audio"""
try:
message_id = payload.message_id
@@ -149,7 +152,7 @@ class TextApi(WebApiResource):
)
return AudioService.transcript_tts(
app_model=app_model,
- session=db.session(),
+ session=session,
text=text,
voice=voice,
end_user=end_user.external_user_id,
diff --git a/api/tests/unit_tests/controllers/service_api/app/test_audio.py b/api/tests/unit_tests/controllers/service_api/app/test_audio.py
index 325c5d33f77..ece7d3381b5 100644
--- a/api/tests/unit_tests/controllers/service_api/app/test_audio.py
+++ b/api/tests/unit_tests/controllers/service_api/app/test_audio.py
@@ -210,17 +210,24 @@ class TestAudioServiceMockedBehavior:
class TestAudioApi:
- def test_success(self, app: Flask, monkeypatch: pytest.MonkeyPatch) -> None:
- monkeypatch.setattr(AudioService, "transcript_asr", lambda **_kwargs: {"text": "ok"})
+ def test_success(self, app: Flask, monkeypatch: pytest.MonkeyPatch, sqlite_session: Session) -> None:
+ calls: dict[str, object] = {}
+
+ def fake_transcript_asr(**kwargs: object) -> dict[str, str]:
+ calls.update(kwargs)
+ return {"text": "ok"}
+
+ monkeypatch.setattr(AudioService, "transcript_asr", fake_transcript_asr)
api = AudioApi()
handler = unwrap(api.post)
app_model = _app()
end_user = _end_user()
with app.test_request_context("/audio-to-text", method="POST", data={"file": _file_data()}):
- response = handler(api, app_model=app_model, end_user=end_user)
+ response = handler(api, sqlite_session, app_model=app_model, end_user=end_user)
assert response == {"text": "ok"}
+ assert calls["session"] is sqlite_session
@pytest.mark.parametrize(
("exc", "expected"),
@@ -237,7 +244,14 @@ class TestAudioApi:
(InvokeError("invoke"), CompletionRequestError),
],
)
- def test_error_mapping(self, app: Flask, monkeypatch: pytest.MonkeyPatch, exc, expected) -> None:
+ def test_error_mapping(
+ self,
+ app: Flask,
+ monkeypatch: pytest.MonkeyPatch,
+ sqlite_session: Session,
+ exc: Exception,
+ expected: type[Exception],
+ ) -> None:
monkeypatch.setattr(AudioService, "transcript_asr", lambda **_kwargs: (_ for _ in ()).throw(exc))
api = AudioApi()
handler = unwrap(api.post)
@@ -246,9 +260,9 @@ class TestAudioApi:
with app.test_request_context("/audio-to-text", method="POST", data={"file": _file_data()}):
with pytest.raises(expected):
- handler(api, app_model=app_model, end_user=end_user)
+ handler(api, sqlite_session, app_model=app_model, end_user=end_user)
- def test_unhandled_error(self, app: Flask, monkeypatch: pytest.MonkeyPatch) -> None:
+ def test_unhandled_error(self, app: Flask, monkeypatch: pytest.MonkeyPatch, sqlite_session: Session) -> None:
monkeypatch.setattr(
AudioService, "transcript_asr", lambda **_kwargs: (_ for _ in ()).throw(RuntimeError("boom"))
)
@@ -259,11 +273,11 @@ class TestAudioApi:
with app.test_request_context("/audio-to-text", method="POST", data={"file": _file_data()}):
with pytest.raises(InternalServerError):
- handler(api, app_model=app_model, end_user=end_user)
+ handler(api, sqlite_session, app_model=app_model, end_user=end_user)
class TestTextApi:
- def test_success(self, app: Flask, monkeypatch: pytest.MonkeyPatch) -> None:
+ def test_success(self, app: Flask, monkeypatch: pytest.MonkeyPatch, sqlite_session: Session) -> None:
monkeypatch.setattr(AudioService, "transcript_tts", lambda **_kwargs: {"audio": "ok"})
api = TextApi()
@@ -277,11 +291,13 @@ class TestTextApi:
json={"text": "hello", "voice": "v"},
):
payload = TextToAudioPayload.model_validate(request.get_json() or {})
- response = handler(api, payload, app_model=app_model, end_user=end_user)
+ response = handler(api, payload, sqlite_session, app_model=app_model, end_user=end_user)
assert response == {"audio": "ok"}
- def test_success_with_message_ref(self, app: Flask, monkeypatch: pytest.MonkeyPatch) -> None:
+ def test_success_with_message_ref(
+ self, app: Flask, monkeypatch: pytest.MonkeyPatch, sqlite_session: Session
+ ) -> None:
calls = {}
def fake_transcript_tts(**kwargs):
@@ -301,12 +317,13 @@ class TestTextApi:
json={"text": "hello", "message_id": "message-1"},
):
payload = TextToAudioPayload.model_validate(request.get_json() or {})
- response = handler(api, payload, app_model=app_model, end_user=end_user)
+ response = handler(api, payload, sqlite_session, app_model=app_model, end_user=end_user)
assert response == {"audio": "ok"}
assert calls["message_ref"] == MessageRef(AppRef("tenant-1", "a1"), "message-1", end_user_id="end-user-1")
+ assert calls["session"] is sqlite_session
- def test_error_mapping(self, app: Flask, monkeypatch: pytest.MonkeyPatch) -> None:
+ def test_error_mapping(self, app: Flask, monkeypatch: pytest.MonkeyPatch, sqlite_session: Session) -> None:
monkeypatch.setattr(
AudioService, "transcript_tts", lambda **_kwargs: (_ for _ in ()).throw(QuotaExceededError())
)
@@ -319,4 +336,4 @@ class TestTextApi:
with app.test_request_context("/text-to-audio", method="POST", json={"text": "hello"}):
payload = TextToAudioPayload.model_validate(request.get_json() or {})
with pytest.raises(ProviderQuotaExceededError):
- handler(api, payload, app_model=app_model, end_user=end_user)
+ handler(api, payload, sqlite_session, app_model=app_model, end_user=end_user)
diff --git a/api/tests/unit_tests/controllers/web/test_audio.py b/api/tests/unit_tests/controllers/web/test_audio.py
index 78f40d6dcb9..c6a36803f19 100644
--- a/api/tests/unit_tests/controllers/web/test_audio.py
+++ b/api/tests/unit_tests/controllers/web/test_audio.py
@@ -2,13 +2,15 @@
from __future__ import annotations
+import inspect
from io import BytesIO
from unittest.mock import MagicMock, patch
import pytest
from flask import Flask
+from sqlalchemy.orm import Session
-from controllers.web.audio import AudioApi, TextApi
+from controllers.web.audio import AudioApi, TextApi, TextToAudioPayload
from controllers.web.error import (
AudioTooLargeError,
CompletionRequestError,
@@ -54,6 +56,13 @@ def _end_user() -> EndUser:
return make_end_user(end_user_id="eu-1", app_id="app-1", external_user_id="ext-1", name="Web User")
+# The @with_session decorator opens a request-scoped session and injects it as
+# the first argument after self; these undecorated handlers let tests pass a
+# session explicitly and assert it is forwarded to AudioService.
+_audio_post = inspect.unwrap(AudioApi.post)
+_text_post = inspect.unwrap(TextApi.post)
+
+
# ---------------------------------------------------------------------------
# AudioApi (audio-to-text)
# ---------------------------------------------------------------------------
@@ -66,6 +75,16 @@ class TestAudioApi:
result = AudioApi().post(_app_model(), _end_user())
assert result == {"text": "hello"}
+ assert isinstance(mock_asr.call_args.kwargs["session"], Session)
+
+ @patch("controllers.web.audio.AudioService.transcript_asr", return_value={"text": "hello"})
+ def test_forwards_injected_session(self, mock_asr: MagicMock, app: Flask, sqlite_session: Session) -> None:
+ data = {"file": (BytesIO(b"fake-audio"), "test.mp3")}
+ with app.test_request_context("/audio-to-text", method="POST", data=data, content_type="multipart/form-data"):
+ result = _audio_post(AudioApi(), sqlite_session, _app_model(), _end_user())
+
+ assert result == {"text": "hello"}
+ assert mock_asr.call_args.kwargs["session"] is sqlite_session
@patch("controllers.web.audio.AudioService.transcript_asr", side_effect=NoAudioUploadedServiceError())
def test_no_audio_uploaded(self, mock_asr: MagicMock, app: Flask) -> None:
@@ -144,6 +163,16 @@ class TestTextApi:
assert result == "audio-bytes"
mock_tts.assert_called_once()
+ assert isinstance(mock_tts.call_args.kwargs["session"], Session)
+
+ @patch("controllers.web.audio.AudioService.transcript_tts", return_value="audio-bytes")
+ def test_forwards_injected_session(self, mock_tts: MagicMock, app: Flask, sqlite_session: Session) -> None:
+ payload = TextToAudioPayload.model_validate({"text": "hello", "voice": "alloy"})
+ with app.test_request_context("/text-to-audio", method="POST"):
+ result = _text_post(TextApi(), payload, sqlite_session, _app_model(), _end_user())
+
+ assert result == "audio-bytes"
+ assert mock_tts.call_args.kwargs["session"] is sqlite_session
@patch("controllers.web.audio.AudioService.transcript_tts", return_value="audio-bytes")
def test_happy_path_with_message_ref(self, mock_tts: MagicMock, app: Flask) -> None: