fix: batch delete document db session block (#32062)
Some checks are pending
autofix.ci / autofix (push) Waiting to run
Build and Push API & Web / build (api, DIFY_API_IMAGE_NAME, linux/amd64, build-api-amd64) (push) Waiting to run
Build and Push API & Web / build (api, DIFY_API_IMAGE_NAME, linux/arm64, build-api-arm64) (push) Waiting to run
Build and Push API & Web / build (web, DIFY_WEB_IMAGE_NAME, linux/amd64, build-web-amd64) (push) Waiting to run
Build and Push API & Web / build (web, DIFY_WEB_IMAGE_NAME, linux/arm64, build-web-arm64) (push) Waiting to run
Build and Push API & Web / create-manifest (api, DIFY_API_IMAGE_NAME, merge-api-images) (push) Blocked by required conditions
Build and Push API & Web / create-manifest (web, DIFY_WEB_IMAGE_NAME, merge-web-images) (push) Blocked by required conditions
Main CI Pipeline / Check Changed Files (push) Waiting to run
Main CI Pipeline / API Tests (push) Blocked by required conditions
Main CI Pipeline / Web Tests (push) Blocked by required conditions
Main CI Pipeline / Style Check (push) Waiting to run
Main CI Pipeline / VDB Tests (push) Blocked by required conditions
Main CI Pipeline / DB Migration Test (push) Blocked by required conditions

This commit is contained in:
Jyong 2026-02-06 18:02:14 +08:00 committed by GitHub
parent 2c9430313d
commit 4430a1b3da
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194

View File

@ -1696,13 +1696,18 @@ class DocumentService:
for document in documents
if document.data_source_type == "upload_file" and document.data_source_info_dict
]
if dataset.doc_form is not None:
batch_clean_document_task.delay(document_ids, dataset.id, dataset.doc_form, file_ids)
# Delete documents first, then dispatch cleanup task after commit
# to avoid deadlock between main transaction and async task
for document in documents:
db.session.delete(document)
db.session.commit()
# Dispatch cleanup task after commit to avoid lock contention
# Task cleans up segments, files, and vector indexes
if dataset.doc_form is not None:
batch_clean_document_task.delay(document_ids, dataset.id, dataset.doc_form, file_ids)
@staticmethod
def rename_document(dataset_id: str, document_id: str, name: str) -> Document:
assert isinstance(current_user, Account)