Fix: Clearing the files of the dataset selected in the chat should result in an error message appearing on the chat page. (#18510)

web/src/components/knowledge-base-item.tsx

@@ -104,38 +104,43 @@ export function useDisableDifferenceEmbeddingDataset(name: string) {
   }, [datasetId, datasetList]);
 
   const nextOptions = useMemo(() => {
-    const datasetListMap = datasetList.map((item: IDataset) => {
-      return {
-        label: item.name,
-        icon: () => (
-          <RAGFlowAvatar
-            className="size-4"
-            avatar={item.avatar}
-            name={item.name}
-          />
-        ),
-        suffix: (
-          <section className="flex gap-2">
-            <DatasetLabel text={item.nickname} />
-            <DatasetLabel
-              text={
-                item.embedding_model_name
-                  ? item.embedding_model_name
-                  : item.embedding_model
-              }
-            />
-          </section>
-        ),
-        value: item.id,
-        disabled:
-          item.chunk_count <= 0 ||
-          item.chunk_method === DocumentParserType.Tag ||
-          (selectedEmbedBaseName !== '' &&
-            getEmbeddingBaseName(item.embedding_model) !== selectedEmbedBaseName),
-      };
-    });
-
-    return datasetListMap;
+    return (
+      datasetList
+        // Datasets without chunks are not selectable. A stale selected value
+        // (emptied or deleted dataset) is excluded as well — the MultiSelect
+        // badge falls back to rendering its raw id and stays removable.
+        .filter((item) => item.chunk_count > 0)
+        .map((item: IDataset) => {
+          return {
+            label: item.name,
+            icon: () => (
+              <RAGFlowAvatar
+                className="size-4"
+                avatar={item.avatar}
+                name={item.name}
+              />
+            ),
+            suffix: (
+              <section className="flex gap-2">
+                <DatasetLabel text={item.nickname} />
+                <DatasetLabel
+                  text={
+                    item.embedding_model_name
+                      ? item.embedding_model_name
+                      : item.embedding_model
+                  }
+                />
+              </section>
+            ),
+            value: item.id,
+            disabled:
+              item.chunk_method === DocumentParserType.Tag ||
+              (selectedEmbedBaseName !== '' &&
+                getEmbeddingBaseName(item.embedding_model) !==
+                  selectedEmbedBaseName),
+          };
+        })
+    );
   }, [datasetList, selectedEmbedBaseName]);
 
   const handleSearchChange = useCallback((value: string) => {

web/src/components/ui/multi-select.tsx

@@ -5,6 +5,7 @@ import { cva, type VariantProps } from 'class-variance-authority';
 import {
   CheckIcon,
   ChevronDown,
+  TriangleAlert,
   WandSparkles,
   XCircle,
   XIcon,
@@ -377,8 +378,20 @@ export const MultiSelect = React.forwardRef<
                           {IconComponent && (
                             <IconComponent className="h-4 w-4" />
                           )}
-                          <div className="max-w-28 text-ellipsis overflow-hidden">
-                            {option?.label}
+                          {/* A selected value with no matching option (e.g. the
+                              entity no longer exists) gets a warning marker and
+                              falls back to the raw value so the badge stays
+                              readable and removable. */}
+                          {!option && (
+                            <TriangleAlert className="h-4 w-4 flex-shrink-0 text-text-disabled" />
+                          )}
+                          <div
+                            className={cn(
+                              'max-w-28 text-ellipsis overflow-hidden',
+                              { 'text-text-disabled': !option },
+                            )}
+                          >
+                            {option?.label ?? value}
                           </div>
                           {canRemoveValue(value) && (
                             <XCircle

web/src/hooks/use-knowledge-request.ts

@@ -1145,9 +1145,9 @@ export const useSelectKnowledgeOptions = () => {
 };
 
 /**
- * Fetch datasets by a set of IDs. Used to resolve the names of
- * already-selected datasets that are not present in the first page of
- * the paginated list so they can be echoed back in the form field.
+ * Fetch datasets by a set of IDs. Used to resolve already-selected datasets
+ * that are not present in the first page of the paginated list, e.g. to echo
+ * their names in a form field. For staleness checks see `useStaleDatasetIds`.
  */
 export const useFetchDatasetsByIds = (ids: string[]) => {
   const sortedIds = useMemo(() => [...ids].sort(), [ids]);
@@ -1164,6 +1164,29 @@ export const useFetchDatasetsByIds = (ids: string[]) => {
   return { data, loading };
 };
 
+/**
+ * Resolve which of the given persisted dataset ids have gone stale — the
+ * dataset no longer exists or has been emptied of chunks. The set stays
+ * empty while the lookup is in flight so consumers can hold off validation
+ * until it settles; `settled` flips true once the lookup has finished.
+ */
+export const useStaleDatasetIds = (datasetIds?: string[]) => {
+  const persistedIds = useMemo(() => datasetIds ?? [], [datasetIds]);
+  const { data: datasets, loading } = useFetchDatasetsByIds(persistedIds);
+
+  const staleDatasetIds = useMemo(() => {
+    if (loading) {
+      return new Set<string>();
+    }
+    const usableIds = new Set(
+      (datasets ?? []).filter((x) => x.chunk_count > 0).map((x) => x.id),
+    );
+    return new Set(persistedIds.filter((id) => !usableIds.has(id)));
+  }, [datasets, loading, persistedIds]);
+
+  return { staleDatasetIds, settled: !loading };
+};
+
 //#region tags
 export const useRenameTag = () => {
   const knowledgeBaseId = useKnowledgeBaseId();

web/src/locales/en.ts

@@ -1040,6 +1040,8 @@ This auto-tagging feature enhances retrieval by adding another layer of domain-s
       knowledgeBases: 'Datasets',
       knowledgeBasesPlaceholder: 'Select value',
       knowledgeBasesMessage: 'Please select',
+      datasetUnavailable:
+        'The selected knowledge base is unavailable (deleted or has no chunks), please re-select',
       knowledgeBasesTip:
         'Select the datasets to associate with this chat assistant. An empty dataset will not appear in the dropdown list.',
       system: 'System prompt',

web/src/locales/zh.ts

@@ -938,6 +938,7 @@ NER：使用 spaCy NER 和基于规则的关键词提取来抽取 Entities 和 R
       knowledgeBases: '知识库',
       knowledgeBasesPlaceholder: '请选择',
       knowledgeBasesMessage: '请选择',
+      datasetUnavailable: '所选知识库不可用（已删除或无 chunk），请重新选择',
       knowledgeBasesTip:
         '选择关联的知识库。新建或空知识库不会在下拉菜单中显示。',
       system: '系统提示词',

web/src/pages/next-chats/chat/app-settings/chat-settings.tsx

@@ -4,6 +4,7 @@ import { ScrollArea } from '@/components/ui/scroll-area';
 import { DatasetMetadata } from '@/constants/chat';
 import { useSetModalState } from '@/hooks/common-hooks';
 import { useFetchChat, useUpdateChat } from '@/hooks/use-chat-request';
+import { useStaleDatasetIds } from '@/hooks/use-knowledge-request';
 import { useFindLlmByUuid } from '@/hooks/use-llm-request';
 import { cn } from '@/lib/utils';
 import {
@@ -14,7 +15,7 @@ import { zodResolver } from '@hookform/resolvers/zod';
 import { isEmpty, omit } from 'lodash';
 import { LucidePanelRightClose, LucideSettings } from 'lucide-react';
 import { useEffect } from 'react';
-import { useForm } from 'react-hook-form';
+import { useForm, useWatch } from 'react-hook-form';
 import { useTranslation } from 'react-i18next';
 import { useParams } from 'react-router';
 import { z } from 'zod';
@@ -27,8 +28,16 @@ import { getWebSearchProvider } from '../web-search-api-key';
 type ChatSettingsProps = { hasSingleChatBox: boolean };
 
 export function ChatSettings({ hasSingleChatBox }: ChatSettingsProps) {
-  const formSchema = useChatSettingSchema();
   const { data } = useFetchChat();
+
+  // Only the persisted ids need validation: ids picked from the dataset
+  // select are valid by construction, while a persisted id may reference a
+  // dataset that has since been deleted or emptied of chunks.
+  const { staleDatasetIds, settled: datasetsFetched } = useStaleDatasetIds(
+    data?.dataset_ids,
+  );
+
+  const formSchema = useChatSettingSchema(staleDatasetIds);
   const { updateChat, loading } = useUpdateChat();
   const findLlmByUuid = useFindLlmByUuid();
   const { id } = useParams();
@@ -42,6 +51,7 @@ export function ChatSettings({ hasSingleChatBox }: ChatSettingsProps) {
   const form = useForm<FormSchemaType>({
     resolver: zodResolver(formSchema),
     shouldUnregister: false,
+    mode: 'onChange',
     defaultValues: {
       name: '',
       icon: '',
@@ -145,6 +155,20 @@ export function ChatSettings({ hasSingleChatBox }: ChatSettingsProps) {
     }
   }, [data, form]);
 
+  const datasetIds = useWatch({ control: form.control, name: 'dataset_ids' });
+  const trigger = form.trigger;
+
+  // A persisted dataset_ids value never fires onChange validation, so once
+  // the lookup of those ids has settled, revalidate explicitly — it may
+  // reference datasets that have since been deleted or emptied of chunks.
+  useEffect(() => {
+    if (!datasetsFetched || !datasetIds?.length) {
+      return;
+    }
+
+    trigger('dataset_ids');
+  }, [trigger, datasetsFetched, datasetIds.length]);
+
   return (
     <>
       {settingVisible || (

web/src/pages/next-chats/chat/app-settings/use-chat-setting-schema.tsx

@@ -13,7 +13,7 @@ import { WebSearchProvider } from '@/constants/chat';
 import { useTranslate } from '@/hooks/common-hooks';
 import { z } from 'zod';
 
-export function useChatSettingSchema() {
+export function useChatSettingSchema(staleDatasetIds: Set<string>) {
   const { t } = useTranslate('chat');
 
   const promptConfigSchema = z.object({
@@ -65,5 +65,15 @@ export function useChatSettingSchema() {
     ...MetadataFilterSchema,
   });
 
-  return formSchema;
+  // A persisted dataset_ids value may reference datasets that have since been
+  // deleted or emptied of chunks — those stale ids are flagged here.
+  return formSchema.superRefine((data, ctx) => {
+    if (data.dataset_ids.some((id) => staleDatasetIds.has(id))) {
+      ctx.addIssue({
+        path: ['dataset_ids'],
+        message: t('datasetUnavailable'),
+        code: z.ZodIssueCode.custom,
+      });
+    }
+  });
 }