Skip to content

RAGFlow is an agent-native database: a Retrieval-Augmented Generation (RAG) engine that grew into context infrastructure.

fix(file): keep polling until cleared datasets disappear from the list (#20423)

chanxdec77fefd0cc
diff --git a/web/src/hooks/use-file-request.ts b/web/src/hooks/use-file-request.ts
index cbb1f36961..4f73611f63 100644
--- a/web/src/hooks/use-file-request.ts
+++ b/web/src/hooks/use-file-request.ts
@@ -329,13 +329,17 @@ export const useRenameFile = () => {
const LinkedDatasetsPollIntervalMs = 500;
const LinkedDatasetsPollMaxAttempts = 6;
-// Returns true when every affected file row on the current page carries all the
-// newly linked datasets. Rows that are not on the current page, or folder rows
-// (which never carry kbs_info), cannot be verified and are treated as done.
+// Returns true when every affected file row on the current page matches the
+// requested link: the selected datasets are present and, in 'replace' mode, no
+// other dataset is left behind — otherwise clearing the selection (empty kbIds)
+// would look done on the stale row. Rows that are not on the current page, or
+// folder rows (which never carry kbs_info), cannot be verified and are treated
+// as done.
const areDatasetsLinked = (
queryClient: QueryClient,
fileIds: string[],
kbIds: string[],
+ mode: ConnectFileToKnowledgeMode,
) => {
const cached = queryClient.getQueriesData<IFetchFileListResult>({
queryKey: [FileApiAction.FetchFileList],
@@ -344,22 +348,27 @@ const areDatasetsLinked = (
(file) => fileIds.includes(file.id) && file.type !== 'folder',
);
if (verifiableFiles.length === 0) return true;
- return verifiableFiles.every((file) =>
- kbIds.every((kbId) => file.kbs_info?.some((kb) => kb.kb_id === kbId)),
- );
+ return verifiableFiles.every((file) => {
+ const linkedIds = (file.kbs_info ?? []).map((kb) => kb.kb_id);
+ return (
+ kbIds.every((kbId) => linkedIds.includes(kbId)) &&
+ (mode === 'add' || linkedIds.length === kbIds.length)
+ );
+ });
};
// Both backends respond to link-to-datasets before the file↔dataset mappings
// are written (the conversion runs in the background), so the refetch right
-// after success can still see stale kbs_info. Poll until the newly linked
-// datasets show up in the list data; give up after a bounded window.
+// after success can still see stale kbs_info. Poll until the change shows up in
+// the list data; give up after a bounded window.
const waitUntilDatasetsLinked = async (
queryClient: QueryClient,
fileIds: string[],
kbIds: string[],
+ mode: ConnectFileToKnowledgeMode,
) => {
for (let attempt = 0; attempt < LinkedDatasetsPollMaxAttempts; attempt++) {
- if (areDatasetsLinked(queryClient, fileIds, kbIds)) return;
+ if (areDatasetsLinked(queryClient, fileIds, kbIds, mode)) return;
await new Promise((resolve) =>
setTimeout(resolve, LinkedDatasetsPollIntervalMs),
);
@@ -398,6 +407,7 @@ export const useConnectToKnowledge = () => {
queryClient,
params.fileIds,
params.kbIds,
+ params.mode,
);
}
return data.code;