Skip to content

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

Fix: throw exception when switching ingestion pipeline (#17833)

Yingfeng302e611a43df
diff --git a/api/db/services/document_counter_service.py b/api/db/services/document_counter_service.py
index d079b5c54d..2e18375ff4 100644
--- a/api/db/services/document_counter_service.py
+++ b/api/db/services/document_counter_service.py
@@ -15,8 +15,7 @@
#
import logging
-from api.db.db_models import DB, Document
-from api.db.services.document_service import DocumentService
+from api.db.db_models import DB, Document, Knowledgebase
def release_reparse_counters(doc_id):
@@ -41,7 +40,35 @@ def release_reparse_counters(doc_id):
if not (fresh.token_num or fresh.chunk_num or fresh.process_duration):
logging.debug("release_reparse_counters: nothing to release for document %s", doc_id)
return
- DocumentService.increment_chunk_num(fresh.id, fresh.kb_id, -fresh.token_num, -fresh.chunk_num, -fresh.process_duration)
+ # Decrement directly inside the outer transaction. Do NOT call
+ # DocumentService.increment_chunk_num here: it is decorated with
+ # @DB.connection_context(), which closes the shared connection on exit.
+ # That would raise "Attempting to close database while transaction is
+ # open." because the outer DB.atomic() block is still open.
+ num = (
+ Document.update(
+ token_num=Document.token_num - fresh.token_num,
+ chunk_num=Document.chunk_num - fresh.chunk_num,
+ process_duration=Document.process_duration - fresh.process_duration,
+ )
+ .where((Document.id == fresh.id) & (Document.kb_id == fresh.kb_id))
+ .execute()
+ )
+ if num == 0:
+ raise LookupError("Document not found which is supposed to be there")
+ # Fetch and lock the knowledgebase row so existence is checked reliably.
+ # A bare UPDATE's num == 0 is ambiguous in MySQL (missing row vs. an
+ # unchanged update both report 0), so rely on the locked fetch instead.
+ kb = Knowledgebase.select().where(Knowledgebase.id == fresh.kb_id).for_update().first()
+ if kb is None:
+ raise LookupError("Knowledgebase not found which is supposed to be there")
+ # Only touch the knowledgebase aggregate when a token/chunk delta exists;
+ # otherwise the update would be a no-op that still reports num == 0.
+ if fresh.token_num or fresh.chunk_num:
+ Knowledgebase.update(
+ token_num=Knowledgebase.token_num - fresh.token_num,
+ chunk_num=Knowledgebase.chunk_num - fresh.chunk_num,
+ ).where(Knowledgebase.id == fresh.kb_id).execute()
logging.debug(
"release_reparse_counters: released document %s (token=%s chunk=%s duration=%s)",
doc_id,