From c098c4c296b4a0004b08224bef2ca7958584b698 Mon Sep 17 00:00:00 2001
From: balibabu <cike8899@users.noreply.github.com>
Date: Tue, 21 Jul 2026 10:12:41 +0800
Subject: [PATCH] Fix: The form related to switching the dataset pipeline
 configuration was not rendered. (#17124)

---
 web/src/interfaces/database/dataset.ts        |  1 +
 web/src/pages/dataset/compilation/index.tsx   |  1 +
 web/src/pages/dataset/setting/hooks.ts        | 51 +++++++++++++++++++
 web/src/pages/dataset/setting/index.tsx       | 40 +++++++++++++--
 web/src/pages/dataset/setting/utils.ts        | 37 +++++++++-----
 .../datasets/dataset-creating-dialog.tsx      | 14 +++--
 web/src/utils/backend-runtime.ts              |  2 +
 7 files changed, 124 insertions(+), 22 deletions(-)

diff --git a/web/src/interfaces/database/dataset.ts b/web/src/interfaces/database/dataset.ts
index 75925b1f21..98c303a50b 100644
--- a/web/src/interfaces/database/dataset.ts
+++ b/web/src/interfaces/database/dataset.ts
@@ -33,6 +33,7 @@ export interface IDataset {
   nickname: string;
   pagerank: number;
   parser_config: Parserconfig;
+  parser_id?: string;
   permission: string;
   pipeline_id: string;
   raptor_task_finish_at: string;
diff --git a/web/src/pages/dataset/compilation/index.tsx b/web/src/pages/dataset/compilation/index.tsx
index a2b4ae90dd..80054c452b 100644
--- a/web/src/pages/dataset/compilation/index.tsx
+++ b/web/src/pages/dataset/compilation/index.tsx
@@ -54,6 +54,7 @@ export default function Compilation() {
             >
               {t('knowledgeDetails.llmWiki')}
             </Button>
+
             <Button
               variant={viewMode === ViewMode.Skills ? 'default' : 'outline'}
               size="sm"
diff --git a/web/src/pages/dataset/setting/hooks.ts b/web/src/pages/dataset/setting/hooks.ts
index bc00d6bc89..d6f9229363 100644
--- a/web/src/pages/dataset/setting/hooks.ts
+++ b/web/src/pages/dataset/setting/hooks.ts
@@ -15,6 +15,7 @@ import {
   useCallback,
   useEffect,
   useMemo,
+  useRef,
   useState,
 } from 'react';
 import { UseFormReturn } from 'react-hook-form';
@@ -22,6 +23,7 @@ import { useParams, useSearchParams } from 'react-router';
 import { z } from 'zod';
 import { formSchema } from './form-schema';
 import {
+  buildParserConfigFromNodes,
   buildPipelineOperatorNodes,
   getOperatorType,
   transformApiConfigToForm,
@@ -140,6 +142,55 @@ export const usePipelineOperatorNodes = (
   return { operatorNodes, loading };
 };
 
+/**
+ * Resets parser_config when the selected pipeline changes, so stale configs
+ * from the previous pipeline are never submitted. Once the new pipeline's
+ * DSL has loaded, parser_config is seeded with the DSL defaults (i.e. what
+ * the operator tabs display).
+ *
+ * The very first pipeline seen is treated as the initial load: the form was
+ * already reset with the saved parser_config in useFetchDatasetSettingOnMount,
+ * so it is left untouched.
+ */
+export const useResetParserConfigOnPipelineChange = (
+  form: UseFormReturn<z.infer<typeof formSchema>>,
+  pipelineId: string | undefined,
+  savedPipelineId: string | undefined,
+  operatorNodes: RAGFlowNodeType[],
+) => {
+  const previousPipelineIdRef = useRef<string>();
+
+  useEffect(() => {
+    if (!pipelineId) {
+      // Selection cleared (e.g. parse type switched): drop configs that
+      // belonged to the previously selected pipeline.
+      if (previousPipelineIdRef.current) {
+        previousPipelineIdRef.current = '';
+        form.setValue('parser_config', {});
+      }
+      return;
+    }
+
+    const isInitialLoad =
+      previousPipelineIdRef.current === undefined &&
+      pipelineId === savedPipelineId;
+    if (isInitialLoad || previousPipelineIdRef.current === pipelineId) {
+      previousPipelineIdRef.current = pipelineId;
+      return;
+    }
+
+    if (operatorNodes.length === 0) {
+      // The new pipeline's DSL is still loading — clear the previous
+      // pipeline's configs right away; defaults are seeded once it arrives.
+      form.setValue('parser_config', {});
+      return;
+    }
+
+    previousPipelineIdRef.current = pipelineId;
+    form.setValue('parser_config', buildParserConfigFromNodes(operatorNodes));
+  }, [form, pipelineId, savedPipelineId, operatorNodes]);
+};
+
 export const useSaveDatasetSetting = () => {
   const { saveKnowledgeConfiguration, loading } = useUpdateKnowledge();
 
diff --git a/web/src/pages/dataset/setting/index.tsx b/web/src/pages/dataset/setting/index.tsx
index dd5833bd2a..0fdc1fcd57 100644
--- a/web/src/pages/dataset/setting/index.tsx
+++ b/web/src/pages/dataset/setting/index.tsx
@@ -31,6 +31,7 @@ import {
   useFetchDatasetSettingOnMount,
   usePipelineDataList,
   usePipelineOperatorNodes,
+  useResetParserConfigOnPipelineChange,
   useSaveDatasetSetting,
 } from './hooks';
 import PipelineOperatorTabs from './pipeline-operator-tabs';
@@ -92,10 +93,35 @@ export default function DatasetSetting() {
     | Record<string, any>
     | undefined;
 
-  const { operatorNodes } = usePipelineOperatorNodes(
-    parseType === ParseType.Pipeline ? pipelineId : builtinPipelineId,
-    pipelineParserConfig,
-    parseType === ParseType.BuiltIn,
+  const isPipelineMode = parseType === ParseType.Pipeline;
+  const selectedPipelineId = isPipelineMode ? pipelineId : builtinPipelineId;
+  // The saved parser_config only belongs to the pipeline (and parse type) it
+  // was saved with — expose its id only while that exact pipeline is selected,
+  // so that after switching, defaults come purely from the new pipeline's DSL.
+  const savedParseType = knowledgeDetails?.pipeline_id
+    ? ParseType.Pipeline
+    : ParseType.BuiltIn;
+  const savedPipelineId =
+    parseType === savedParseType
+      ? isPipelineMode
+        ? knowledgeDetails?.pipeline_id
+        : knowledgeDetails?.parser_id
+      : undefined;
+
+  const { operatorNodes, loading: operatorNodesLoading } =
+    usePipelineOperatorNodes(
+      selectedPipelineId,
+      selectedPipelineId && selectedPipelineId === savedPipelineId
+        ? pipelineParserConfig
+        : undefined,
+      !isPipelineMode,
+    );
+
+  useResetParserConfigOnPipelineChange(
+    form,
+    selectedPipelineId,
+    savedPipelineId,
+    operatorNodes,
   );
 
   const { activeTab, setActiveTab } = useActiveTab(operatorNodes);
@@ -211,7 +237,11 @@ export default function DatasetSetting() {
 
                 <ButtonLoading
                   type="submit"
-                  loading={datasetSettingLoading || saveLoading}
+                  loading={
+                    datasetSettingLoading ||
+                    saveLoading ||
+                    (operatorNodesLoading && operatorNodes.length === 0)
+                  }
                 >
                   {t('knowledgeConfiguration.save')}
                 </ButtonLoading>
diff --git a/web/src/pages/dataset/setting/utils.ts b/web/src/pages/dataset/setting/utils.ts
index 67e8b7bd57..ceabc7d5a5 100644
--- a/web/src/pages/dataset/setting/utils.ts
+++ b/web/src/pages/dataset/setting/utils.ts
@@ -13,7 +13,7 @@ import {
   transformTitleChunkerParams,
   transformTokenChunkerParams,
 } from '@/pages/agent/utils';
-import { cloneDeep } from 'lodash';
+import { cloneDeep, isEmpty } from 'lodash';
 
 export const FileNodeId = 'File';
 
@@ -278,24 +278,21 @@ export function buildOperatorNode(
   const operatorId = dslNode.id;
   const operatorType = getOperatorType(operatorId);
 
-  // DSL-format config from API's parser_config
-  const configFromApi = operatorId
-    ? pipelineParserConfig[operatorId]
-    : undefined;
-
   // Form-format config from DSL graph
   const configFromDsl = dslNode.data?.form;
 
-  // Convert API config to form format, then merge (DSL template is baseline, API overrides)
-  const convertedApiConfig = transformApiConfigToForm(
-    operatorType,
-    configFromApi,
-  );
   const rawForm = {
     ...configFromDsl, // template defaults from DSL (form format)
-    ...convertedApiConfig, // user overrides from API (now also form format)
   };
 
+  if (!isEmpty(pipelineParserConfig)) {
+    // user overrides from API (now also form format)
+    Object.assign(
+      rawForm,
+      transformApiConfigToForm(operatorType, pipelineParserConfig[operatorId]), // Convert API config to form format, then merge (DSL template is baseline, API overrides)
+    );
+  }
+
   return {
     ...dslNode,
     id: '',
@@ -343,3 +340,19 @@ export function buildPipelineOperatorNodes(
     .filter((node): node is RAGFlowNodeType => node !== undefined)
     .map((node) => buildOperatorNode(node, pipelineParserConfig));
 }
+
+/**
+ * Builds a parser_config record from operator nodes, keyed by operatorId.
+ * Each entry is the node's merged form (DSL template defaults + overrides),
+ * i.e. exactly what the operator tabs display.
+ */
+export function buildParserConfigFromNodes(
+  operatorNodes: RAGFlowNodeType[],
+): Record<string, any> {
+  return Object.fromEntries(
+    operatorNodes.map((node) => [
+      (node.data as Record<string, any>)?.operatorId || node.data?.label || '',
+      node.data?.form ?? {},
+    ]),
+  );
+}
diff --git a/web/src/pages/datasets/dataset-creating-dialog.tsx b/web/src/pages/datasets/dataset-creating-dialog.tsx
index 964fb61fd4..ecb73be5d5 100644
--- a/web/src/pages/datasets/dataset-creating-dialog.tsx
+++ b/web/src/pages/datasets/dataset-creating-dialog.tsx
@@ -29,17 +29,18 @@ import { useTranslation } from 'react-i18next';
 import { z } from 'zod';
 import {
   BuiltinPipelineItem,
+  ChunkMethodItem,
   EmbeddingModelItem,
   ParseTypeItem,
 } from '../dataset/dataset-setting/configuration/common-item';
+import { isGoBackend } from '@/utils/backend-runtime';
 
 const FormId = 'dataset-creating-form';
 
-const ChunkMethodName = 'parser_id';
-
 export function InputForm({ onOk }: IModalProps<any>) {
   const { t } = useTranslation();
   const defaultModelDictionary = useFetchDefaultModelDictionary();
+  const ChunkMethodName = isGoBackend() ? 'parser_id' : 'chunk_method';
 
   const FormSchema = z
     .object({
@@ -141,9 +142,12 @@ export function InputForm({ onOk }: IModalProps<any>) {
 
         <EmbeddingModelItem line={2} isEdit={false} />
         <ParseTypeItem />
-        {parseType === ParseType.BuiltIn && (
-          <BuiltinPipelineItem name={ChunkMethodName} />
-        )}
+        {parseType === ParseType.BuiltIn &&
+          (isGoBackend() ? (
+            <BuiltinPipelineItem name={ChunkMethodName} />
+          ) : (
+            <ChunkMethodItem name={ChunkMethodName}></ChunkMethodItem>
+          ))}
         {parseType === ParseType.Pipeline && (
           <DataFlowSelect
             isMult={false}
diff --git a/web/src/utils/backend-runtime.ts b/web/src/utils/backend-runtime.ts
index 42823078d1..58c751e4c8 100644
--- a/web/src/utils/backend-runtime.ts
+++ b/web/src/utils/backend-runtime.ts
@@ -32,6 +32,8 @@ export const fetchBackendLanguage = (): Promise<string> => promise;
 
 export const getBackendLanguage = (): string | null => backendLanguage;
 
+export const isGoBackend = (): boolean => backendLanguage === 'go';
+
 export const subscribeBackendLanguage = (listener: Listener): (() => void) => {
   listeners.add(listener);
   return () => {