Actions
Feature #21306
closedReduce RBD filestore/bluestore fragmentation throught fallocate
Status:
Rejected
Priority:
Normal
Assignee:
-
Target version:
-
% Done:
0%
Source:
Tags:
Backport:
Reviewed:
Affected Versions:
Pull request ID:
Description
It is well know fact that RBD images make XFS filesystem heavily fragmented on Filestore OSDs.
So I propose a way to eliminate this:
1. rados,osd: implement fallocate() for objects.
2. Implement FALLOC_FL_KEEP_SIZE both in rados,osd
3. Add an option to RBD (clients or maybe image itself) to call fallocate(FALLOC_FL_KEEP_SIZE, chunk_size) on newly created objects.
Updated by Sage Weil about 6 years ago
- Status changed from New to Rejected
We already have a function like this for filestore: librbd does a hint and filestore calls the xfs ioctl to set the default allocation size. By default hte alloc size is capped at 1MB.. see filestore_max_alloc_hint_size
Actions