Project

General

Profile

Actions

Fix #6977

closed

objecter: ENOSPC returned by pool quota will cause IO errors for non-cached rbd images

Added by Josh Durgin over 10 years ago. Updated almost 7 years ago.

Status:
Resolved
Priority:
Normal
Assignee:
-
Category:
-
Target version:
-
% Done:

0%

Source:
other
Tags:
Backport:
Reviewed:
Affected Versions:
ceph-qa-suite:
Pull request ID:
Crash signature (v1):
Crash signature (v2):

Description

The client has no idea why it's getting ENOSPC, and block layers turn this into EIO. The only way clients can detect this is by reading the return code, which the kernel client does not currently do except for successful messages. I'm not sure there's a good way to fix this and preserve write ordering, so maybe we just need to document 'don't use pool quotas with rbd'.

Actions #1

Updated by Josh Durgin over 10 years ago

  • Tracker changed from Bug to Fix

One way to fix this is to treat it similar to the full flag, by putting an overquota flag on a pool in the osdmap, and having clients handle it like they do the FULL flag, but just for the over-quota pools.

Actions #2

Updated by Jason Dillaman almost 7 years ago

  • Status changed from New to Resolved

Fixed by #12018

Actions

Also available in: Atom PDF