youngkingdom/ragflow
1
0
Fork 0
mirror of https://github.com/infiniflow/ragflow.git synced 2026-05-06 10:17:49 +08:00
Code Issues Packages Projects Releases Wiki Activity

Fix: allow document parsing status recovery after transient errors (#13341)

Browse Source 
### What problem does this PR solve?

Fixes #13285

When an LLM returns a transient error (e.g. overloaded) during parsing,
the task progress is set to -1. Previously, the progress could never be
updated again, leaving the document permanently stuck in FAIL status
even after the task successfully recovered and completed.

Three coordinated changes address this:

1. task_service.update_progress: relax the progress update guard to
accept prog >= 1 even when current progress is -1, so a task that
recovers from a transient failure can report completion.

2. document_service.get_unfinished_docs: include documents that are
marked FAIL (progress == -1) but still have at least one non-failed task
(task.progress >= 0) in the polling set, so their status can be
re-synced once a task recovers. Documents where all tasks have
permanently failed are excluded to avoid unnecessary polling.

3. document_service.update_progress: explicitly set document status to
RUNNING when not all tasks have finished, instead of preserving whatever
stale status (potentially FAIL) the document previously had.

### Type of change

- [x] Bug Fix (non-breaking change which fixes an issue)
This commit is contained in:
Jinghan Xu
2026-03-12 18:02:12 +08:00
committed by GitHub
parent 13a34d7689
commit f6b06fab72
2 changed files with 18 additions and 15 deletions
Download Patch File Download Diff File Expand all files Collapse all files

17
api/db/services/task_service.py
View File

@ -304,9 +304,8 @@ class TaskService(CommonService):
Update Rules:
- progress_msg: Always appends the new message to the existing one, and trims the result to max 3000 lines.
- progress: Only updates if the current progress is not -1 AND
(the new progress is -1 OR greater than the existing progress),
to avoid overwriting valid progress with invalid or regressive values.
- progress: Updates when (a) new progress >= 1 (allows recovery from -1), or
(b) current progress != -1 AND (new progress is -1 OR greater than existing).
Args:
id (str): The unique identifier of the task to update.
@ -327,10 +326,8 @@ class TaskService(CommonService):
prog = info["progress"]
cls.model.update(progress=prog).where(
(cls.model.id == id) &
(
(cls.model.progress != -1) &
((prog == -1) | (prog > cls.model.progress))
)
((prog >= 1) | ((cls.model.progress != -1) &
((prog == -1) | (prog > cls.model.progress))))
).execute()
else:
with DB.lock("update_progress", -1):
@ -341,10 +338,8 @@ class TaskService(CommonService):
prog = info["progress"]
cls.model.update(progress=prog).where(
(cls.model.id == id) &
(
(cls.model.progress != -1) &
((prog == -1) | (prog > cls.model.progress))
)
((prog >= 1) | ((cls.model.progress != -1) &
((prog == -1) | (prog > cls.model.progress))))
).execute()
process_duration = (datetime.now() - task.begin_at).total_seconds()