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.

fix(api): avoid duplicate streaming app concurrency admission (#45038)

非法操作2626b79e9efb
diff --git a/api/core/app/features/rate_limiting/rate_limit.py b/api/core/app/features/rate_limiting/rate_limit.py
index d495be83a95..93eff83d517 100644
--- a/api/core/app/features/rate_limiting/rate_limit.py
+++ b/api/core/app/features/rate_limiting/rate_limit.py
@@ -1,4 +1,3 @@
-import contextlib
import logging
import time
import uuid
@@ -110,14 +109,6 @@ class RateLimit:
)
-def rate_limit_context(rate_limit: RateLimit, request_id: str | None):
- request_id = rate_limit.enter(request_id)
- yield
- if request_id is not None:
- rate_limit.exit(request_id)
-
-
class RateLimitGenerator:
def __init__(self, rate_limit: RateLimit, generator: Generator[str], request_id: str):
self.rate_limit = rate_limit
diff --git a/api/services/app_generate_service.py b/api/services/app_generate_service.py
index 0dc5d8297ca..c3881dcce80 100644
--- a/api/services/app_generate_service.py
+++ b/api/services/app_generate_service.py
@@ -18,7 +18,6 @@ from core.app.apps.message_based_app_generator import MessageBasedAppGenerator
from core.app.apps.workflow.app_generator import WorkflowAppGenerator
from core.app.entities.app_invoke_entities import InvokeFrom
from core.app.features.rate_limiting import RateLimit
-from core.app.features.rate_limiting.rate_limit import rate_limit_context
from core.app.layers.pause_state_persist_layer import PauseStateLayerConfig
from core.db import session_factory
from enums import DeploymentEdition, QuotaType
@@ -251,18 +250,17 @@ class AppGenerateService:
if streaming:
# Streaming mode: subscribe to SSE and enqueue the execution on first subscriber
- with rate_limit_context(rate_limit, request_id):
- payload = AppExecutionParams.new(
- app_model=app_model,
- workflow=workflow,
- user=user,
- args=args,
- invoke_from=invoke_from,
- streaming=True,
- call_depth=0,
- workflow_run_id=str(uuid.uuid4()),
- )
- payload_json = payload.model_dump_json()
+ payload = AppExecutionParams.new(
+ app_model=app_model,
+ workflow=workflow,
+ user=user,
+ args=args,
+ invoke_from=invoke_from,
+ streaming=True,
+ call_depth=0,
+ workflow_run_id=str(uuid.uuid4()),
+ )
+ payload_json = payload.model_dump_json()
def on_subscribe():
workflow_based_app_execution_task.delay(payload_json)
@@ -307,19 +305,18 @@ class AppGenerateService:
workflow = cls._get_workflow(app_model, invoke_from, workflow_id, session=session)
cls._ensure_workflow_service_mode_available(workflow=workflow, invoke_from=invoke_from)
if streaming:
- with rate_limit_context(rate_limit, request_id):
- payload = AppExecutionParams.new(
- app_model=app_model,
- workflow=workflow,
- user=user,
- args=args,
- invoke_from=invoke_from,
- streaming=True,
- call_depth=0,
- root_node_id=root_node_id,
- workflow_run_id=str(uuid.uuid4()),
- )
- payload_json = payload.model_dump_json()
+ payload = AppExecutionParams.new(
+ app_model=app_model,
+ workflow=workflow,
+ user=user,
+ args=args,
+ invoke_from=invoke_from,
+ streaming=True,
+ call_depth=0,
+ root_node_id=root_node_id,
+ workflow_run_id=str(uuid.uuid4()),
+ )
+ payload_json = payload.model_dump_json()
def on_subscribe():
workflow_based_app_execution_task.delay(payload_json)
diff --git a/api/tests/unit_tests/services/test_app_generate_service.py b/api/tests/unit_tests/services/test_app_generate_service.py
index c74254365e0..7c1795857a9 100644
--- a/api/tests/unit_tests/services/test_app_generate_service.py
+++ b/api/tests/unit_tests/services/test_app_generate_service.py
@@ -2,8 +2,8 @@ import json
import threading
import uuid
from collections import defaultdict, deque
-from collections.abc import Callable
-from contextlib import contextmanager
+from collections.abc import Callable, Generator
+from datetime import timedelta
from typing import Any
from unittest.mock import MagicMock
@@ -15,6 +15,9 @@ import services.app_generate_service as ags_module
from core.app.apps.message_based_app_generator import MessageBasedAppGenerator
from core.app.apps.message_generator import MessageGenerator
from core.app.entities.app_invoke_entities import InvokeFrom
+from core.app.features.rate_limiting import rate_limit as rate_limit_module
+from core.app.features.rate_limiting.rate_limit import RateLimit, RateLimitGenerator
+from core.errors.error import AppInvokeQuotaExceededError
from enums import DeploymentEdition, QuotaType
from models.account import Account
from models.enums import AppStatus
@@ -133,10 +136,35 @@ def _make_workflow(
)
-@contextmanager
-def _noop_rate_limit_context(rate_limit, request_id):
- """Drop-in replacement for rate_limit_context that doesn't touch Redis."""
- yield
+class _RateLimitRedis:
+ def __init__(self) -> None:
+ self.values: dict[str, bytes] = {}
+ self.hashes: dict[str, dict[str, bytes]] = {}
+
+ def exists(self, key: str) -> bool:
+ return key in self.values or key in self.hashes
+
+ def setex(self, key: str, ttl: timedelta, value: int) -> None:
+ self.values[key] = str(value).encode()
+
+ def get(self, key: str) -> bytes:
+ return self.values[key]
+
+ def expire(self, key: str, ttl: timedelta) -> None:
+ pass
+
+ def hgetall(self, key: str) -> dict[str, bytes]:
+ return self.hashes.get(key, {})
+
+ def hlen(self, key: str) -> int:
+ return len(self.hashes.get(key, {}))
+
+ def hset(self, key: str, field: str, value: str) -> None:
+ self.hashes.setdefault(key, {})[field] = value.encode()
+
+ def hdel(self, key: str, *fields: str) -> None:
+ for field in fields:
+ self.hashes.get(key, {}).pop(field, None)
# ---------------------------------------------------------------------------
@@ -304,11 +332,6 @@ class TestAppGenerateServiceCases:
def _common(self, mocker: MockerFixture, config_overrides: Callable[..., None]):
config_overrides(DEPLOYMENT_EDITION=DeploymentEdition.COMMUNITY)
mocker.patch("services.app_generate_service.RateLimit", _DummyRateLimit)
- # Prevent AppExecutionParams.new from touching real models via isinstance
- mocker.patch(
- "services.app_generate_service.rate_limit_context",
- _noop_rate_limit_context,
- )
# -- COMPLETION ---------------------------------------------------------
def test_completion_mode(self, mocker: MockerFixture):
@@ -636,6 +659,73 @@ class TestAppGenerateServiceCases:
session=self.session,
)
+ class TestStreamingRateLimit(_RealSessionTest):
+ @pytest.mark.parametrize("app_mode", [AppMode.ADVANCED_CHAT, AppMode.WORKFLOW])
+ @pytest.mark.parametrize("termination", ["finish", "error", "close"])
+ def test_single_slot_is_held_until_stream_ends(
+ self,
+ monkeypatch: pytest.MonkeyPatch,
+ config_overrides: Callable[..., None],
+ app_mode: AppMode,
+ termination: str,
+ ) -> None:
+ config_overrides(
+ DEPLOYMENT_EDITION=DeploymentEdition.COMMUNITY,
+ APP_MAX_ACTIVE_REQUESTS=0,
+ APP_DEFAULT_ACTIVE_REQUESTS=0,
+ PUBSUB_REDIS_CHANNEL_TYPE="streams",
+ )
+ redis = _RateLimitRedis()
+ monkeypatch.setattr(rate_limit_module, "redis_client", redis)
+ monkeypatch.setattr(RateLimit, "_instance_dict", {})
+ workflow = _make_workflow()
+ monkeypatch.setattr(AppGenerateService, "_get_workflow", lambda *_args, **_kwargs: workflow)
+
+ def retrieve_events(*_args: object, **_kwargs: object) -> Generator[str]:
+ yield "ping"
+ if termination == "error":
+ raise RuntimeError("stream failed")
+
+ monkeypatch.setattr(MessageBasedAppGenerator, "retrieve_events", retrieve_events)
+ app = _make_app(app_mode, max_active_requests=1)
+
+ def generate() -> RateLimitGenerator:
+ response = AppGenerateService.generate(
+ app_model=app,
+ user=_make_user(),
+ args={"inputs": {}, "query": "hi"},
+ invoke_from=InvokeFrom.DEBUGGER,
+ streaming=True,
+ session=self.session,
+ )
+ assert isinstance(response, RateLimitGenerator)
+ return response
+
+ first = generate()
+ active_key = first.rate_limit.active_requests_key
+ assert set(redis.hgetall(active_key)) == {first.request_id}
+
+ with pytest.raises(AppInvokeQuotaExceededError, match="maximum concurrent requests allowed"):
+ generate()
+ assert set(redis.hgetall(active_key)) == {first.request_id}
+
+ assert next(first) == "event: ping\n\n"
+ if termination == "finish":
+ assert list(first) == []
+ elif termination == "error":
+ with pytest.raises(RuntimeError, match="stream failed"):
+ next(first)
+ else:
+ first.close()
+ first.close()
+ assert redis.hlen(active_key) == 0
+
+ following = generate()
+ assert following.request_id != first.request_id
+ assert set(redis.hgetall(active_key)) == {following.request_id}
+ following.close()
+ assert redis.hlen(active_key) == 0
+
# ---------------------------------------------------------------------------
# generate – billing / quota
# ---------------------------------------------------------------------------
@@ -643,10 +733,6 @@ class TestAppGenerateServiceCases:
@pytest.fixture(autouse=True)
def _common(self, mocker: MockerFixture):
mocker.patch("services.app_generate_service.RateLimit", _DummyRateLimit)
- mocker.patch(
- "services.app_generate_service.rate_limit_context",
- _noop_rate_limit_context,
- )
def test_cloud_edition_consumes_quota(self, mocker: MockerFixture, config_overrides: Callable[..., None]):
config_overrides(DEPLOYMENT_EDITION=DeploymentEdition.CLOUD)