estel/dify
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

feat: decouple WorkflowAppRunner from AppRunner (#21739)

Browse Source 
Signed-off-by: -LAN- <laipz8200@outlook.com>
This commit is contained in:
-LAN-
2025-07-23 14:56:06 +08:00
committed by GitHub
parent 60c37fe492
commit 6b544aa0b9
6 changed files with 198 additions and 171 deletions
Download Patch File Download Diff File Expand all files Collapse all files

48
api/core/app/apps/README.md
View File

@@ -1,48 +0,0 @@
## Guidelines for Database Connection Management in App Runner and Task Pipeline
Due to the presence of tasks in App Runner that require long execution times, such as LLM generation and external requests, Flask-Sqlalchemy's strategy for database connection pooling is to allocate one connection (transaction) per request. This approach keeps a connection occupied even during non-DB tasks, leading to the inability to acquire new connections during high concurrency requests due to multiple long-running tasks.
Therefore, the database operations in App Runner and Task Pipeline must ensure connections are closed immediately after use, and it's better to pass IDs rather than Model objects to avoid detach errors.
Examples:
1. Creating a new record:
```python
app = App(id=1)
db.session.add(app)
db.session.commit()
db.session.refresh(app) # Retrieve table default values, like created_at, cached in the app object, won't affect after close
# Handle non-long-running tasks or store the content of the App instance in memory (via variable assignment).
db.session.close()
return app.id
```
2. Fetching a record from the table:
```python
app = db.session.query(App).filter(App.id == app_id).first()
created_at = app.created_at
db.session.close()
# Handle tasks (include long-running).
```
3. Updating a table field:
```python
app = db.session.query(App).filter(App.id == app_id).first()
app.updated_at = time.utcnow()
db.session.commit()
db.session.close()
return app_id
```

38
api/core/app/apps/advanced_chat/app_generator.py
View File

@@ -7,7 +7,8 @@ from typing import Any, Literal, Optional, Union, overload
from flask import Flask, current_app
from pydantic import ValidationError
from sqlalchemy.orm import sessionmaker
from sqlalchemy import select
from sqlalchemy.orm import Session, sessionmaker
import contexts
from configs import dify_config
@@ -486,12 +487,39 @@ class AdvancedChatAppGenerator(MessageBasedAppGenerator):
"""
with preserve_flask_contexts(flask_app, context_vars=context):
try:
# get conversation and message
conversation = self._get_conversation(conversation_id)
message = self._get_message(message_id)
# chatbot app
with Session(db.engine, expire_on_commit=False) as session:
workflow = session.scalar(
select(Workflow).where(
Workflow.tenant_id == application_generate_entity.app_config.tenant_id,
Workflow.app_id == application_generate_entity.app_config.app_id,
Workflow.id == application_generate_entity.app_config.workflow_id,
)
)
if workflow is None:
raise ValueError("Workflow not found")
# Determine system_user_id based on invocation source
is_external_api_call = application_generate_entity.invoke_from in {
InvokeFrom.WEB_APP,
InvokeFrom.SERVICE_API,
}
if is_external_api_call:
# For external API calls, use end user's session ID
end_user = session.scalar(select(EndUser).where(EndUser.id == application_generate_entity.user_id))
system_user_id = end_user.session_id if end_user else ""
else:
# For internal calls, use the original user ID
system_user_id = application_generate_entity.user_id
app = session.scalar(select(App).where(App.id == application_generate_entity.app_config.app_id))
if app is None:
raise ValueError("App not found")
runner = AdvancedChatAppRunner(
application_generate_entity=application_generate_entity,
queue_manager=queue_manager,
@@ -499,8 +527,12 @@ class AdvancedChatAppGenerator(MessageBasedAppGenerator):
message=message,
dialogue_count=self._dialogue_count,
variable_loader=variable_loader,
workflow=workflow,
system_user_id=system_user_id,
app=app,
)
try:
runner.run()
except GenerateTaskStoppedError:
pass

115
api/core/app/apps/advanced_chat/app_runner.py
View File

@@ -1,6 +1,6 @@
import logging
from collections.abc import Mapping
from typing import Any, cast
from typing import Any, Optional, cast
from sqlalchemy import select
from sqlalchemy.orm import Session
@@ -9,13 +9,19 @@ from configs import dify_config
from core.app.apps.advanced_chat.app_config_manager import AdvancedChatAppConfig
from core.app.apps.base_app_queue_manager import AppQueueManager
from core.app.apps.workflow_app_runner import WorkflowBasedAppRunner
from core.app.entities.app_invoke_entities import AdvancedChatAppGenerateEntity, InvokeFrom
from core.app.entities.app_invoke_entities import (
AdvancedChatAppGenerateEntity,
AppGenerateEntity,
InvokeFrom,
)
from core.app.entities.queue_entities import (
QueueAnnotationReplyEvent,
QueueStopEvent,
QueueTextChunkEvent,
)
from core.app.features.annotation_reply.annotation_reply import AnnotationReplyFeature
from core.moderation.base import ModerationError
from core.moderation.input_moderation import InputModeration
from core.variables.variables import VariableUnion
from core.workflow.callbacks import WorkflowCallback, WorkflowLoggingCallback
from core.workflow.entities.variable_pool import VariablePool
@@ -23,8 +29,9 @@ from core.workflow.system_variable import SystemVariable
from core.workflow.variable_loader import VariableLoader
from core.workflow.workflow_entry import WorkflowEntry
from extensions.ext_database import db
from models import Workflow
from models.enums import UserFrom
from models.model import App, Conversation, EndUser, Message
from models.model import App, Conversation, Message, MessageAnnotation
from models.workflow import ConversationVariable, WorkflowType
logger = logging.getLogger(__name__)
@@ -37,21 +44,29 @@ class AdvancedChatAppRunner(WorkflowBasedAppRunner):
def __init__(
self,
*,
application_generate_entity: AdvancedChatAppGenerateEntity,
queue_manager: AppQueueManager,
conversation: Conversation,
message: Message,
dialogue_count: int,
variable_loader: VariableLoader,
workflow: Workflow,
system_user_id: str,
app: App,
) -> None:
super().__init__(queue_manager, variable_loader)
super().__init__(
queue_manager=queue_manager,
variable_loader=variable_loader,
app_id=application_generate_entity.app_config.app_id,
)
self.application_generate_entity = application_generate_entity
self.conversation = conversation
self.message = message
self._dialogue_count = dialogue_count
def _get_app_id(self) -> str:
return self.application_generate_entity.app_config.app_id
self._workflow = workflow
self.system_user_id = system_user_id
self._app = app
def run(self) -> None:
app_config = self.application_generate_entity.app_config
@@ -61,18 +76,6 @@ class AdvancedChatAppRunner(WorkflowBasedAppRunner):
if not app_record:
raise ValueError("App not found")
workflow = self.get_workflow(app_model=app_record, workflow_id=app_config.workflow_id)
if not workflow:
raise ValueError("Workflow not initialized")
user_id: str | None = None
if self.application_generate_entity.invoke_from in {InvokeFrom.WEB_APP, InvokeFrom.SERVICE_API}:
end_user = db.session.query(EndUser).filter(EndUser.id == self.application_generate_entity.user_id).first()
if end_user:
user_id = end_user.session_id
else:
user_id = self.application_generate_entity.user_id
workflow_callbacks: list[WorkflowCallback] = []
if dify_config.DEBUG:
workflow_callbacks.append(WorkflowLoggingCallback())
@@ -80,14 +83,14 @@ class AdvancedChatAppRunner(WorkflowBasedAppRunner):
if self.application_generate_entity.single_iteration_run:
# if only single iteration run is requested
graph, variable_pool = self._get_graph_and_variable_pool_of_single_iteration(
workflow=workflow,
workflow=self._workflow,
node_id=self.application_generate_entity.single_iteration_run.node_id,
user_inputs=dict(self.application_generate_entity.single_iteration_run.inputs),
)
elif self.application_generate_entity.single_loop_run:
# if only single loop run is requested
graph, variable_pool = self._get_graph_and_variable_pool_of_single_loop(
workflow=workflow,
workflow=self._workflow,
node_id=self.application_generate_entity.single_loop_run.node_id,
user_inputs=dict(self.application_generate_entity.single_loop_run.inputs),
)
@@ -98,7 +101,7 @@ class AdvancedChatAppRunner(WorkflowBasedAppRunner):
# moderation
if self.handle_input_moderation(
app_record=app_record,
app_record=self._app,
app_generate_entity=self.application_generate_entity,
inputs=inputs,
query=query,
@@ -108,7 +111,7 @@ class AdvancedChatAppRunner(WorkflowBasedAppRunner):
# annotation reply
if self.handle_annotation_reply(
app_record=app_record,
app_record=self._app,
message=self.message,
query=query,
app_generate_entity=self.application_generate_entity,
@@ -128,7 +131,7 @@ class AdvancedChatAppRunner(WorkflowBasedAppRunner):
ConversationVariable.from_variable(
app_id=self.conversation.app_id, conversation_id=self.conversation.id, variable=variable
)
for variable in workflow.conversation_variables
for variable in self._workflow.conversation_variables
]
session.add_all(db_conversation_variables)
# Convert database entities to variables.
@@ -141,7 +144,7 @@ class AdvancedChatAppRunner(WorkflowBasedAppRunner):
query=query,
files=files,
conversation_id=self.conversation.id,
user_id=user_id,
user_id=self.system_user_id,
dialogue_count=self._dialogue_count,
app_id=app_config.app_id,
workflow_id=app_config.workflow_id,
@@ -152,25 +155,25 @@ class AdvancedChatAppRunner(WorkflowBasedAppRunner):
variable_pool = VariablePool(
system_variables=system_inputs,
user_inputs=inputs,
environment_variables=workflow.environment_variables,
environment_variables=self._workflow.environment_variables,
# Based on the definition of `VariableUnion`,
# `list[Variable]` can be safely used as `list[VariableUnion]` since they are compatible.
conversation_variables=cast(list[VariableUnion], conversation_variables),
)
# init graph
graph = self._init_graph(graph_config=workflow.graph_dict)
graph = self._init_graph(graph_config=self._workflow.graph_dict)
db.session.close()
# RUN WORKFLOW
workflow_entry = WorkflowEntry(
tenant_id=workflow.tenant_id,
app_id=workflow.app_id,
workflow_id=workflow.id,
workflow_type=WorkflowType.value_of(workflow.type),
tenant_id=self._workflow.tenant_id,
app_id=self._workflow.app_id,
workflow_id=self._workflow.id,
workflow_type=WorkflowType.value_of(self._workflow.type),
graph=graph,
graph_config=workflow.graph_dict,
graph_config=self._workflow.graph_dict,
user_id=self.application_generate_entity.user_id,
user_from=(
UserFrom.ACCOUNT
@@ -241,3 +244,51 @@ class AdvancedChatAppRunner(WorkflowBasedAppRunner):
self._publish_event(QueueTextChunkEvent(text=text))
self._publish_event(QueueStopEvent(stopped_by=stopped_by))
def query_app_annotations_to_reply(
self, app_record: App, message: Message, query: str, user_id: str, invoke_from: InvokeFrom
) -> Optional[MessageAnnotation]:
"""
Query app annotations to reply
:param app_record: app record
:param message: message
:param query: query
:param user_id: user id
:param invoke_from: invoke from
:return:
"""
annotation_reply_feature = AnnotationReplyFeature()
return annotation_reply_feature.query(
app_record=app_record, message=message, query=query, user_id=user_id, invoke_from=invoke_from
)
def moderation_for_inputs(
self,
*,
app_id: str,
tenant_id: str,
app_generate_entity: AppGenerateEntity,
inputs: Mapping[str, Any],
query: str | None = None,
message_id: str,
) -> tuple[bool, Mapping[str, Any], str]:
"""
Process sensitive_word_avoidance.
:param app_id: app id
:param tenant_id: tenant id
:param app_generate_entity: app generate entity
:param inputs: inputs
:param query: query
:param message_id: message id
:return:
"""
moderation_feature = InputModeration()
return moderation_feature.check(
app_id=app_id,
tenant_id=tenant_id,
app_config=app_generate_entity.app_config,
inputs=dict(inputs),
query=query or "",
message_id=message_id,
trace_manager=app_generate_entity.trace_manager,
)

38
api/core/app/apps/workflow/app_generator.py
View File

@@ -7,7 +7,8 @@ from typing import Any, Literal, Optional, Union, overload
from flask import Flask, current_app
from pydantic import ValidationError
from sqlalchemy.orm import sessionmaker
from sqlalchemy import select
from sqlalchemy.orm import Session, sessionmaker
import contexts
from configs import dify_config
@@ -445,17 +446,44 @@ class WorkflowAppGenerator(BaseAppGenerator):
"""
with preserve_flask_contexts(flask_app, context_vars=context):
try:
# workflow app
with Session(db.engine, expire_on_commit=False) as session:
workflow = session.scalar(
select(Workflow).where(
Workflow.tenant_id == application_generate_entity.app_config.tenant_id,
Workflow.app_id == application_generate_entity.app_config.app_id,
Workflow.id == application_generate_entity.app_config.workflow_id,
)
)
if workflow is None:
raise ValueError("Workflow not found")
# Determine system_user_id based on invocation source
is_external_api_call = application_generate_entity.invoke_from in {
InvokeFrom.WEB_APP,
InvokeFrom.SERVICE_API,
}
if is_external_api_call:
# For external API calls, use end user's session ID
end_user = session.scalar(select(EndUser).where(EndUser.id == application_generate_entity.user_id))
system_user_id = end_user.session_id if end_user else ""
else:
# For internal calls, use the original user ID
system_user_id = application_generate_entity.user_id
runner = WorkflowAppRunner(
application_generate_entity=application_generate_entity,
queue_manager=queue_manager,
workflow_thread_pool_id=workflow_thread_pool_id,
variable_loader=variable_loader,
workflow=workflow,
system_user_id=system_user_id,
)
try:
runner.run()
except GenerateTaskStoppedError:
except GenerateTaskStoppedError as e:
logger.warning(f"Task stopped: {str(e)}")
pass
except InvokeAuthorizationError:
queue_manager.publish_error(
@@ -471,8 +499,6 @@ class WorkflowAppGenerator(BaseAppGenerator):
except Exception as e:
logger.exception("Unknown Error when generating")
queue_manager.publish_error(e, PublishFrom.APPLICATION_MANAGER)
finally:
db.session.close()
def _handle_response(
self,

61
api/core/app/apps/workflow/app_runner.py
View File

@@ -14,10 +14,8 @@ from core.workflow.entities.variable_pool import VariablePool
from core.workflow.system_variable import SystemVariable
from core.workflow.variable_loader import VariableLoader
from core.workflow.workflow_entry import WorkflowEntry
from extensions.ext_database import db
from models.enums import UserFrom
from models.model import App, EndUser
from models.workflow import WorkflowType
from models.workflow import Workflow, WorkflowType
logger = logging.getLogger(__name__)
@@ -29,22 +27,23 @@ class WorkflowAppRunner(WorkflowBasedAppRunner):
def __init__(
self,
*,
application_generate_entity: WorkflowAppGenerateEntity,
queue_manager: AppQueueManager,
variable_loader: VariableLoader,
workflow_thread_pool_id: Optional[str] = None,
workflow: Workflow,
system_user_id: str,
) -> None:
"""
:param application_generate_entity: application generate entity
:param queue_manager: application queue manager
:param workflow_thread_pool_id: workflow thread pool id
"""
super().__init__(queue_manager, variable_loader)
super().__init__(
queue_manager=queue_manager,
variable_loader=variable_loader,
app_id=application_generate_entity.app_config.app_id,
)
self.application_generate_entity = application_generate_entity
self.workflow_thread_pool_id = workflow_thread_pool_id
def _get_app_id(self) -> str:
return self.application_generate_entity.app_config.app_id
self._workflow = workflow
self._sys_user_id = system_user_id
def run(self) -> None:
"""
@@ -53,24 +52,6 @@ class WorkflowAppRunner(WorkflowBasedAppRunner):
app_config = self.application_generate_entity.app_config
app_config = cast(WorkflowAppConfig, app_config)
user_id = None
if self.application_generate_entity.invoke_from in {InvokeFrom.WEB_APP, InvokeFrom.SERVICE_API}:
end_user = db.session.query(EndUser).filter(EndUser.id == self.application_generate_entity.user_id).first()
if end_user:
user_id = end_user.session_id
else:
user_id = self.application_generate_entity.user_id
app_record = db.session.query(App).filter(App.id == app_config.app_id).first()
if not app_record:
raise ValueError("App not found")
workflow = self.get_workflow(app_model=app_record, workflow_id=app_config.workflow_id)
if not workflow:
raise ValueError("Workflow not initialized")
db.session.close()
workflow_callbacks: list[WorkflowCallback] = []
if dify_config.DEBUG:
workflow_callbacks.append(WorkflowLoggingCallback())
@@ -79,14 +60,14 @@ class WorkflowAppRunner(WorkflowBasedAppRunner):
if self.application_generate_entity.single_iteration_run:
# if only single iteration run is requested
graph, variable_pool = self._get_graph_and_variable_pool_of_single_iteration(
workflow=workflow,
workflow=self._workflow,
node_id=self.application_generate_entity.single_iteration_run.node_id,
user_inputs=self.application_generate_entity.single_iteration_run.inputs,
)
elif self.application_generate_entity.single_loop_run:
# if only single loop run is requested
graph, variable_pool = self._get_graph_and_variable_pool_of_single_loop(
workflow=workflow,
workflow=self._workflow,
node_id=self.application_generate_entity.single_loop_run.node_id,
user_inputs=self.application_generate_entity.single_loop_run.inputs,
)
@@ -98,7 +79,7 @@ class WorkflowAppRunner(WorkflowBasedAppRunner):
system_inputs = SystemVariable(
files=files,
user_id=user_id,
user_id=self._sys_user_id,
app_id=app_config.app_id,
workflow_id=app_config.workflow_id,
workflow_execution_id=self.application_generate_entity.workflow_execution_id,
@@ -107,21 +88,21 @@ class WorkflowAppRunner(WorkflowBasedAppRunner):
variable_pool = VariablePool(
system_variables=system_inputs,
user_inputs=inputs,
environment_variables=workflow.environment_variables,
environment_variables=self._workflow.environment_variables,
conversation_variables=[],
)
# init graph
graph = self._init_graph(graph_config=workflow.graph_dict)
graph = self._init_graph(graph_config=self._workflow.graph_dict)
# RUN WORKFLOW
workflow_entry = WorkflowEntry(
tenant_id=workflow.tenant_id,
app_id=workflow.app_id,
workflow_id=workflow.id,
workflow_type=WorkflowType.value_of(workflow.type),
tenant_id=self._workflow.tenant_id,
app_id=self._workflow.app_id,
workflow_id=self._workflow.id,
workflow_type=WorkflowType.value_of(self._workflow.type),
graph=graph,
graph_config=workflow.graph_dict,
graph_config=self._workflow.graph_dict,
user_id=self.application_generate_entity.user_id,
user_from=(
UserFrom.ACCOUNT

39
api/core/app/apps/workflow_app_runner.py
View File

@@ -1,8 +1,7 @@
from collections.abc import Mapping
from typing import Any, Optional, cast
from typing import Any, cast
from core.app.apps.base_app_queue_manager import AppQueueManager, PublishFrom
from core.app.apps.base_app_runner import AppRunner
from core.app.entities.queue_entities import (
AppQueueEvent,
QueueAgentLogEvent,
@@ -65,18 +64,20 @@ from core.workflow.nodes.node_mapping import NODE_TYPE_CLASSES_MAPPING
from core.workflow.system_variable import SystemVariable
from core.workflow.variable_loader import DUMMY_VARIABLE_LOADER, VariableLoader, load_into_variable_pool
from core.workflow.workflow_entry import WorkflowEntry
from extensions.ext_database import db
from models.model import App
from models.workflow import Workflow
class WorkflowBasedAppRunner(AppRunner):
def __init__(self, queue_manager: AppQueueManager, variable_loader: VariableLoader = DUMMY_VARIABLE_LOADER) -> None:
self.queue_manager = queue_manager
class WorkflowBasedAppRunner:
def __init__(
self,
*,
queue_manager: AppQueueManager,
variable_loader: VariableLoader = DUMMY_VARIABLE_LOADER,
app_id: str,
) -> None:
self._queue_manager = queue_manager
self._variable_loader = variable_loader
def _get_app_id(self) -> str:
raise NotImplementedError("not implemented")
self._app_id = app_id
def _init_graph(self, graph_config: Mapping[str, Any]) -> Graph:
"""
@@ -693,21 +694,5 @@ class WorkflowBasedAppRunner(AppRunner):
)
)
def get_workflow(self, app_model: App, workflow_id: str) -> Optional[Workflow]:
"""
Get workflow
"""
# fetch workflow by workflow_id
workflow = (
db.session.query(Workflow)
.filter(
Workflow.tenant_id == app_model.tenant_id, Workflow.app_id == app_model.id, Workflow.id == workflow_id
)
.first()
)
# return workflow
return workflow
def _publish_event(self, event: AppQueueEvent) -> None:
self.queue_manager.publish(event, PublishFrom.APPLICATION_MANAGER)
self._queue_manager.publish(event, PublishFrom.APPLICATION_MANAGER)