refactor: resolve summary models through manager (#40696)

api/core/rag/index_processor/processor/paragraph_index_processor.py

@@ -9,12 +9,10 @@ from sqlalchemy import select
 from sqlalchemy.orm import Session
 
 from core.app.file_access import DatabaseFileAccessController
-from core.app.llm import deduct_llm_quota
 from core.db.session_factory import session_factory
 from core.entities.knowledge_entities import PreviewDetail
 from core.llm_generator.prompts import DEFAULT_GENERATOR_SUMMARY_PROMPT
-from core.model_manager import ModelInstance
-from core.plugin.impl.model_runtime_factory import create_plugin_provider_manager
+from core.model_manager import ModelManager
 from core.rag.cleaner.clean_processor import CleanProcessor
 from core.rag.datasource.keyword.keyword_factory import Keyword
 from core.rag.datasource.vdb.vector_factory import Vector
@@ -388,7 +386,7 @@ class ParagraphIndexProcessor(BaseIndexProcessor):
         session: Session,
     ) -> tuple[str, LLMUsage]:
         """
-        Generate summary for the given text using ModelInstance.invoke_llm and the default or custom summary prompt,
+        Generate summary for the given text using the configured LLM and the default or custom summary prompt,
         and supports vision models by including images from the segment attachments or text content.
 
         Args:
@@ -431,11 +429,13 @@ class ParagraphIndexProcessor(BaseIndexProcessor):
                 # If default prompt doesn't have {language} placeholder, use it as-is
                 pass
 
-        provider_manager = create_plugin_provider_manager(tenant_id=tenant_id)
-        provider_model_bundle = provider_manager.get_provider_model_bundle(
-            tenant_id, model_provider_name, ModelType.LLM
+        model_manager = ModelManager.for_tenant(tenant_id=tenant_id)
+        model_instance = model_manager.get_model_instance(
+            tenant_id=tenant_id,
+            provider=model_provider_name,
+            model_type=ModelType.LLM,
+            model=model_name,
         )
-        model_instance = ModelInstance(provider_model_bundle, model_name)
 
         # Get model schema to check if vision is supported
         model_schema = model_instance.model_type_instance.get_model_schema(model_name, model_instance.credentials)
@@ -496,13 +496,6 @@ class ParagraphIndexProcessor(BaseIndexProcessor):
         summary_content = result.message.get_text_content()
         usage = result.usage
 
-        # Deduct quota for summary generation (same as workflow nodes)
-        try:
-            deduct_llm_quota(tenant_id=tenant_id, model_instance=model_instance, usage=usage)
-        except Exception as e:
-            # Log but don't fail summary generation if quota deduction fails
-            logger.warning("Failed to deduct quota for summary generation: %s", str(e))
-
         return summary_content, usage
 
     @staticmethod

api/tests/unit_tests/core/rag/indexing/processor/test_paragraph_index_processor.py

@@ -16,7 +16,7 @@ from core.rag.models.document import AttachmentDocument, Document
 from extensions.storage.storage_type import StorageType
 from graphon.model_runtime.entities.llm_entities import LLMResult, LLMUsage
 from graphon.model_runtime.entities.message_entities import AssistantPromptMessage, ImagePromptMessageContent
-from graphon.model_runtime.entities.model_entities import ModelFeature
+from graphon.model_runtime.entities.model_entities import ModelFeature, ModelType
 from models.dataset import DocumentSegment, SegmentAttachmentBinding
 from models.enums import CreatorUserRole
 from models.model import UploadFile
@@ -519,41 +519,51 @@ class TestParagraphIndexProcessor:
         with pytest.raises(ValueError, match="model_name and model_provider_name"):
             ParagraphIndexProcessor.generate_summary("tenant-1", "text", {"enable": True}, session=self.session)
 
-    def test_generate_summary_text_only_flow(self, caplog: pytest.LogCaptureFixture) -> None:
+    def test_generate_summary_text_only_flow(self) -> None:
         model_instance = Mock()
         model_instance.credentials = {"k": "v"}
         model_instance.model_type_instance.get_model_schema.return_value = SimpleNamespace(features=[])
         model_instance.invoke_llm.return_value = self._llm_result("text summary")
 
-        with (
-            patch(
-                "core.rag.index_processor.processor.paragraph_index_processor.create_plugin_provider_manager"
-            ) as mock_provider_manager,
-            patch(
-                "core.rag.index_processor.processor.paragraph_index_processor.ModelInstance",
-                return_value=model_instance,
-            ),
-            patch(
-                "core.rag.index_processor.processor.paragraph_index_processor.deduct_llm_quota",
-                side_effect=RuntimeError("quota"),
-            ),
-        ):
-            mock_provider_manager.return_value.get_provider_model_bundle.return_value = Mock()
-            with caplog.at_level(
-                logging.WARNING, logger="core.rag.index_processor.processor.paragraph_index_processor"
-            ):
-                summary, usage = ParagraphIndexProcessor.generate_summary(
-                    "tenant-1",
-                    "text content",
-                    {"enable": True, "model_name": "model-a", "model_provider_name": "provider-a"},
-                    document_language="English",
-                    session=self.session,
-                )
+        with patch(
+            "core.rag.index_processor.processor.paragraph_index_processor.ModelManager.for_tenant"
+        ) as mock_model_manager:
+            mock_model_manager.return_value.get_model_instance.return_value = model_instance
+            summary, usage = ParagraphIndexProcessor.generate_summary(
+                "tenant-1",
+                "text content",
+                {"enable": True, "model_name": "model-a", "model_provider_name": "provider-a"},
+                document_language="English",
+                session=self.session,
+            )
 
         assert summary == "text summary"
         assert isinstance(usage, LLMUsage)
-        assert sum(1 for r in caplog.records if r.levelno == logging.WARNING) == 1
-        assert any("Failed to deduct quota for summary generation" in record.message for record in caplog.records)
+        mock_model_manager.assert_called_once_with(tenant_id="tenant-1")
+        mock_model_manager.return_value.get_model_instance.assert_called_once_with(
+            tenant_id="tenant-1",
+            provider="provider-a",
+            model_type=ModelType.LLM,
+            model="model-a",
+        )
+
+    def test_generate_summary_propagates_model_invocation_errors(self) -> None:
+        model_instance = Mock()
+        model_instance.credentials = {"k": "v"}
+        model_instance.model_type_instance.get_model_schema.return_value = SimpleNamespace(features=[])
+        model_instance.invoke_llm.side_effect = RuntimeError("invocation failed")
+
+        with patch(
+            "core.rag.index_processor.processor.paragraph_index_processor.ModelManager.for_tenant"
+        ) as mock_model_manager:
+            mock_model_manager.return_value.get_model_instance.return_value = model_instance
+            with pytest.raises(RuntimeError, match="invocation failed"):
+                ParagraphIndexProcessor.generate_summary(
+                    "tenant-1",
+                    "text content",
+                    {"enable": True, "model_name": "model-a", "model_provider_name": "provider-a"},
+                    session=self.session,
+                )
 
     def test_generate_summary_handles_vision_and_image_conversion(self) -> None:
         model_instance = Mock()
@@ -567,12 +577,8 @@ class TestParagraphIndexProcessor:
 
         with (
             patch(
-                "core.rag.index_processor.processor.paragraph_index_processor.create_plugin_provider_manager"
-            ) as mock_provider_manager,
-            patch(
-                "core.rag.index_processor.processor.paragraph_index_processor.ModelInstance",
-                return_value=model_instance,
-            ),
+                "core.rag.index_processor.processor.paragraph_index_processor.ModelManager.for_tenant"
+            ) as mock_model_manager,
             patch.object(
                 ParagraphIndexProcessor, "_extract_images_from_segment_attachments", return_value=[image_file]
             ),
@@ -581,9 +587,8 @@ class TestParagraphIndexProcessor:
                 "core.rag.index_processor.processor.paragraph_index_processor.file_manager.to_prompt_message_content",
                 return_value=image_content,
             ),
-            patch("core.rag.index_processor.processor.paragraph_index_processor.deduct_llm_quota"),
         ):
-            mock_provider_manager.return_value.get_provider_model_bundle.return_value = Mock()
+            mock_model_manager.return_value.get_model_instance.return_value = model_instance
             summary, _ = ParagraphIndexProcessor.generate_summary(
                 "tenant-1",
                 "text content",
@@ -606,12 +611,8 @@ class TestParagraphIndexProcessor:
 
         with (
             patch(
-                "core.rag.index_processor.processor.paragraph_index_processor.create_plugin_provider_manager"
-            ) as mock_provider_manager,
-            patch(
-                "core.rag.index_processor.processor.paragraph_index_processor.ModelInstance",
-                return_value=model_instance,
-            ),
+                "core.rag.index_processor.processor.paragraph_index_processor.ModelManager.for_tenant"
+            ) as mock_model_manager,
             patch(
                 "core.rag.index_processor.processor.paragraph_index_processor.DEFAULT_GENERATOR_SUMMARY_PROMPT",
                 "Prompt {missing}",
@@ -623,7 +624,7 @@ class TestParagraphIndexProcessor:
                 side_effect=RuntimeError("bad image"),
             ),
         ):
-            mock_provider_manager.return_value.get_provider_model_bundle.return_value = Mock()
+            mock_model_manager.return_value.get_model_instance.return_value = model_instance
             with pytest.raises(ValueError, match="Expected LLMResult"):
                 with caplog.at_level(
                     logging.WARNING, logger="core.rag.index_processor.processor.paragraph_index_processor"