Merge branch 'langgenius:main' into add-turbo-pack

api/configs/packaging/__init__.py

@@ -9,7 +9,7 @@ class PackagingInfo(BaseSettings):
 
     CURRENT_VERSION: str = Field(
         description="Dify version",
-        default="1.4.1",
+        default="1.4.2",
     )
 
     COMMIT_SHA: str = Field(

api/controllers/inner_api/plugin/plugin.py

@@ -29,7 +29,7 @@ from core.plugin.entities.request import (
     RequestRequestUploadFile,
 )
 from core.tools.entities.tool_entities import ToolProviderType
-from libs.helper import compact_generate_response
+from libs.helper import length_prefixed_response
 from models.account import Account, Tenant
 from models.model import EndUser
 
@@ -44,7 +44,7 @@ class PluginInvokeLLMApi(Resource):
             response = PluginModelBackwardsInvocation.invoke_llm(user_model.id, tenant_model, payload)
             return PluginModelBackwardsInvocation.convert_to_event_stream(response)
 
-        return compact_generate_response(generator())
+        return length_prefixed_response(0xF, generator())
 
 
 class PluginInvokeTextEmbeddingApi(Resource):
@@ -101,7 +101,7 @@ class PluginInvokeTTSApi(Resource):
             )
             return PluginModelBackwardsInvocation.convert_to_event_stream(response)
 
-        return compact_generate_response(generator())
+        return length_prefixed_response(0xF, generator())
 
 
 class PluginInvokeSpeech2TextApi(Resource):
@@ -162,7 +162,7 @@ class PluginInvokeToolApi(Resource):
                 ),
             )
 
-        return compact_generate_response(generator())
+        return length_prefixed_response(0xF, generator())
 
 
 class PluginInvokeParameterExtractorNodeApi(Resource):
@@ -228,7 +228,7 @@ class PluginInvokeAppApi(Resource):
             files=payload.files,
         )
 
-        return compact_generate_response(PluginAppBackwardsInvocation.convert_to_event_stream(response))
+        return length_prefixed_response(0xF, PluginAppBackwardsInvocation.convert_to_event_stream(response))
 
 
 class PluginInvokeEncryptApi(Resource):

api/core/app/apps/base_app_runner.py

@@ -1,3 +1,4 @@
+import logging
 import time
 from collections.abc import Generator, Mapping, Sequence
 from typing import TYPE_CHECKING, Any, Optional, Union
@@ -33,6 +34,8 @@ from models.model import App, AppMode, Message, MessageAnnotation
 if TYPE_CHECKING:
     from core.file.models import File
 
+_logger = logging.getLogger(__name__)
+
 
 class AppRunner:
     def get_pre_calculate_rest_tokens(
@@ -298,7 +301,7 @@ class AppRunner:
         )
 
     def _handle_invoke_result_stream(
-        self, invoke_result: Generator, queue_manager: AppQueueManager, agent: bool
+        self, invoke_result: Generator[LLMResultChunk, None, None], queue_manager: AppQueueManager, agent: bool
     ) -> None:
         """
         Handle invoke result
@@ -317,18 +320,28 @@ class AppRunner:
             else:
                 queue_manager.publish(QueueAgentMessageEvent(chunk=result), PublishFrom.APPLICATION_MANAGER)
 
-            text += result.delta.message.content
+            message = result.delta.message
+            if isinstance(message.content, str):
+                text += message.content
+            elif isinstance(message.content, list):
+                for content in message.content:
+                    if not isinstance(content, str):
+                        # TODO(QuantumGhost): Add multimodal output support for easy ui.
+                        _logger.warning("received multimodal output, type=%s", type(content))
+                        text += content.data
+                    else:
+                        text += content  # failback to str
 
             if not model:
                 model = result.model
 
             if not prompt_messages:
-                prompt_messages = result.prompt_messages
+                prompt_messages = list(result.prompt_messages)
 
             if result.delta.usage:
                 usage = result.delta.usage
 
-        if not usage:
+        if usage is None:
             usage = LLMUsage.empty_usage()
 
         llm_result = LLMResult(

api/core/app/task_pipeline/easy_ui_based_generate_task_pipeline.py

@@ -48,6 +48,7 @@ from core.model_manager import ModelInstance
 from core.model_runtime.entities.llm_entities import LLMResult, LLMResultChunk, LLMResultChunkDelta, LLMUsage
 from core.model_runtime.entities.message_entities import (
     AssistantPromptMessage,
+    TextPromptMessageContent,
 )
 from core.model_runtime.model_providers.__base.large_language_model import LargeLanguageModel
 from core.ops.entities.trace_entity import TraceTaskName
@@ -309,6 +310,23 @@ class EasyUIBasedGenerateTaskPipeline(BasedGenerateTaskPipeline):
                 delta_text = chunk.delta.message.content
                 if delta_text is None:
                     continue
+                if isinstance(chunk.delta.message.content, list):
+                    delta_text = ""
+                    for content in chunk.delta.message.content:
+                        logger.debug(
+                            "The content type %s in LLM chunk delta message content.: %r", type(content), content
+                        )
+                        if isinstance(content, TextPromptMessageContent):
+                            delta_text += content.data
+                        elif isinstance(content, str):
+                            delta_text += content  # failback to str
+                        else:
+                            logger.warning(
+                                "Unsupported content type %s in LLM chunk delta message content.: %r",
+                                type(content),
+                                content,
+                            )
+                            continue
 
                 if not self._task_state.llm_result.prompt_messages:
                     self._task_state.llm_result.prompt_messages = chunk.prompt_messages

api/core/plugin/backwards_invocation/base.py

@@ -11,14 +11,12 @@ class BaseBackwardsInvocation:
             try:
                 for chunk in response:
                     if isinstance(chunk, BaseModel | dict):
-                        yield BaseBackwardsInvocationResponse(data=chunk).model_dump_json().encode() + b"\n\n"
-                    elif isinstance(chunk, str):
-                        yield f"event: {chunk}\n\n".encode()
+                        yield BaseBackwardsInvocationResponse(data=chunk).model_dump_json().encode()
             except Exception as e:
                 error_message = BaseBackwardsInvocationResponse(error=str(e)).model_dump_json()
-                yield f"{error_message}\n\n".encode()
+                yield error_message.encode()
         else:
-            yield BaseBackwardsInvocationResponse(data=response).model_dump_json().encode() + b"\n\n"
+            yield BaseBackwardsInvocationResponse(data=response).model_dump_json().encode()
 
 
 T = TypeVar("T", bound=dict | Mapping | str | bool | int | BaseModel)

api/core/repositories/sqlalchemy_workflow_execution_repository.py

@@ -6,7 +6,7 @@ import json
 import logging
 from typing import Optional, Union
 
-from sqlalchemy import select
+from sqlalchemy import func, select
 from sqlalchemy.engine import Engine
 from sqlalchemy.orm import sessionmaker
 
@@ -151,11 +151,11 @@ class SQLAlchemyWorkflowExecutionRepository(WorkflowExecutionRepository):
             existing = session.scalar(select(WorkflowRun).where(WorkflowRun.id == domain_model.id_))
             if not existing:
                 # For new records, get the next sequence number
-                stmt = select(WorkflowRun.sequence_number).where(
+                stmt = select(func.max(WorkflowRun.sequence_number)).where(
                     WorkflowRun.app_id == self._app_id,
                     WorkflowRun.tenant_id == self._tenant_id,
                 )
-                max_sequence = session.scalar(stmt.order_by(WorkflowRun.sequence_number.desc()))
+                max_sequence = session.scalar(stmt)
                 db_model.sequence_number = (max_sequence or 0) + 1
             else:
                 # For updates, keep the existing sequence number

api/core/workflow/nodes/llm/node.py

@@ -525,6 +525,8 @@ class LLMNode(BaseNode[LLMNodeData]):
                 # Set appropriate response format based on model capabilities
                 self._set_response_format(completion_params, model_schema.parameter_rules)
         model_config_with_cred.parameters = completion_params
+        # NOTE(-LAN-): This line modify the `self.node_data.model`, which is used in `_invoke_llm()`.
+        node_data_model.completion_params = completion_params
         return model, model_config_with_cred
 
     def _fetch_prompt_messages(

api/libs/helper.py

@@ -3,6 +3,7 @@ import logging
 import re
 import secrets
 import string
+import struct
 import subprocess
 import time
 import uuid
@@ -14,6 +15,7 @@ from zoneinfo import available_timezones
 
 from flask import Response, stream_with_context
 from flask_restful import fields
+from pydantic import BaseModel
 
 from configs import dify_config
 from core.app.features.rate_limiting.rate_limit import RateLimitGenerator
@@ -183,7 +185,7 @@ def generate_string(n):
 
 def extract_remote_ip(request) -> str:
     if request.headers.get("CF-Connecting-IP"):
-        return cast(str, request.headers.get("Cf-Connecting-Ip"))
+        return cast(str, request.headers.get("CF-Connecting-IP"))
     elif request.headers.getlist("X-Forwarded-For"):
         return cast(str, request.headers.getlist("X-Forwarded-For")[0])
     else:
@@ -206,6 +208,60 @@ def compact_generate_response(response: Union[Mapping, Generator, RateLimitGener
         return Response(stream_with_context(generate()), status=200, mimetype="text/event-stream")
 
 
+def length_prefixed_response(magic_number: int, response: Union[Mapping, Generator, RateLimitGenerator]) -> Response:
+    """
+    This function is used to return a response with a length prefix.
+    Magic number is a one byte number that indicates the type of the response.
+
+    For a compatibility with latest plugin daemon https://github.com/langgenius/dify-plugin-daemon/pull/341
+    Avoid using line-based response, it leads a memory issue.
+
+    We uses following format:
+    | Field         | Size     | Description                     |
+    |---------------|----------|---------------------------------|
+    | Magic Number  | 1 byte   | Magic number identifier         |
+    | Reserved      | 1 byte   | Reserved field                  |
+    | Header Length | 2 bytes  | Header length (usually 0xa)    |
+    | Data Length   | 4 bytes  | Length of the data              |
+    | Reserved      | 6 bytes  | Reserved fields                 |
+    | Data          | Variable | Actual data content             |
+
+    | Reserved Fields | Header   | Data     |
+    |-----------------|----------|----------|
+    | 4 bytes total   | Variable | Variable |
+
+    all data is in little endian
+    """
+
+    def pack_response_with_length_prefix(response: bytes) -> bytes:
+        header_length = 0xA
+        data_length = len(response)
+        # | Magic Number 1byte | Reserved 1byte | Header Length 2bytes | Data Length 4bytes | Reserved 6bytes | Data
+        return struct.pack("<BBHI", magic_number, 0, header_length, data_length) + b"\x00" * 6 + response
+
+    if isinstance(response, dict):
+        return Response(
+            response=pack_response_with_length_prefix(json.dumps(jsonable_encoder(response)).encode("utf-8")),
+            status=200,
+            mimetype="application/json",
+        )
+    elif isinstance(response, BaseModel):
+        return Response(
+            response=pack_response_with_length_prefix(response.model_dump_json().encode("utf-8")),
+            status=200,
+            mimetype="application/json",
+        )
+
+    def generate() -> Generator:
+        for chunk in response:
+            if isinstance(chunk, str):
+                yield pack_response_with_length_prefix(chunk.encode("utf-8"))
+            else:
+                yield pack_response_with_length_prefix(chunk)
+
+    return Response(stream_with_context(generate()), status=200, mimetype="text/event-stream")
+
+
 class TokenManager:
     @classmethod
     def generate_token(

docker/docker-compose-template.yaml

@@ -2,7 +2,7 @@ x-shared-env: &shared-api-worker-env
 services:
   # API service
   api:
-    image: langgenius/dify-api:1.4.1
+    image: langgenius/dify-api:1.4.2
     restart: always
     environment:
       # Use the shared environment variables.
@@ -31,7 +31,7 @@ services:
   # worker service
   # The Celery worker for processing the queue.
   worker:
-    image: langgenius/dify-api:1.4.1
+    image: langgenius/dify-api:1.4.2
     restart: always
     environment:
       # Use the shared environment variables.
@@ -57,7 +57,7 @@ services:
 
   # Frontend web application.
   web:
-    image: langgenius/dify-web:1.4.1
+    image: langgenius/dify-web:1.4.2
     restart: always
     environment:
       CONSOLE_API_URL: ${CONSOLE_API_URL:-}
@@ -142,7 +142,7 @@ services:
 
   # plugin daemon
   plugin_daemon:
-    image: langgenius/dify-plugin-daemon:0.1.1-local
+    image: langgenius/dify-plugin-daemon:0.1.2-local
     restart: always
     environment:
       # Use the shared environment variables.

docker/docker-compose.middleware.yaml

@@ -71,7 +71,7 @@ services:
 
   # plugin daemon
   plugin_daemon:
-    image: langgenius/dify-plugin-daemon:0.1.1-local
+    image: langgenius/dify-plugin-daemon:0.1.2-local
     restart: always
     env_file:
       - ./middleware.env

docker/docker-compose.yaml

@@ -508,7 +508,7 @@ x-shared-env: &shared-api-worker-env
 services:
   # API service
   api:
-    image: langgenius/dify-api:1.4.1
+    image: langgenius/dify-api:1.4.2
     restart: always
     environment:
       # Use the shared environment variables.
@@ -537,7 +537,7 @@ services:
   # worker service
   # The Celery worker for processing the queue.
   worker:
-    image: langgenius/dify-api:1.4.1
+    image: langgenius/dify-api:1.4.2
     restart: always
     environment:
       # Use the shared environment variables.
@@ -563,7 +563,7 @@ services:
 
   # Frontend web application.
   web:
-    image: langgenius/dify-web:1.4.1
+    image: langgenius/dify-web:1.4.2
     restart: always
     environment:
       CONSOLE_API_URL: ${CONSOLE_API_URL:-}
@@ -648,7 +648,7 @@ services:
 
   # plugin daemon
   plugin_daemon:
-    image: langgenius/dify-plugin-daemon:0.1.1-local
+    image: langgenius/dify-plugin-daemon:0.1.2-local
     restart: always
     environment:
       # Use the shared environment variables.

web/app/(shareLayout)/layout.tsx

@@ -12,12 +12,18 @@ const Layout: FC<{
 }> = ({ children }) => {
   const isGlobalPending = useGlobalPublicStore(s => s.isGlobalPending)
   const setWebAppAccessMode = useGlobalPublicStore(s => s.setWebAppAccessMode)
+  const systemFeatures = useGlobalPublicStore(s => s.systemFeatures)
   const pathname = usePathname()
   const searchParams = useSearchParams()
   const redirectUrl = searchParams.get('redirect_url')
   const [isLoading, setIsLoading] = useState(true)
   useEffect(() => {
     (async () => {
+      if (!systemFeatures.webapp_auth.enabled) {
+        setIsLoading(false)
+        return
+      }
+
       let appCode: string | null = null
       if (redirectUrl)
         appCode = redirectUrl?.split('/').pop() || null

web/app/components/base/markdown-blocks/button.tsx

@@ -14,7 +14,7 @@ const MarkdownButton = ({ node }: any) => {
     size={size}
     className={cn('!h-auto min-h-8 select-none whitespace-normal !px-3')}
     onClick={() => {
-      if (isValidUrl(link)) {
+      if (link && isValidUrl(link)) {
         window.open(link, '_blank')
         return
       }

web/package.json

@@ -1,6 +1,6 @@
 {
   "name": "dify-web",
-  "version": "1.4.1",
+  "version": "1.4.2",
   "private": true,
   "engines": {
     "node": ">=v22.11.0"