Feature #21849
sparse-reads should not be used for small IO requests
Status:
Resolved
Priority:
Normal
Assignee:
Jason Dillaman
Target version:
-
% Done:
0%
Source:
Tags:
Backport:
luminous
Reviewed:
Affected Versions:
Pull request ID:
Description
There is a larger time and space overhead for performing a sparse read so it should be avoided for small IO. Backend support for sparse-read is disabled by default for filestore, and bluestore defaults to 16K allocations for SSDs and 64K allocations for HDDs. Perhaps it would be simplest to just restrict sparse-read to only the cases where the full object is being read and provide an advanced-level config override to lower the requirement if needed.
Related issues
History
#1 Updated by Jason Dillaman over 5 years ago
- Status changed from New to In Progress
- Backport set to luminous
#2 Updated by Jason Dillaman over 5 years ago
- Assignee set to Jason Dillaman
#3 Updated by Jason Dillaman over 5 years ago
- Status changed from In Progress to Fix Under Review
#4 Updated by Mykola Golub over 5 years ago
- Status changed from Fix Under Review to Pending Backport
#5 Updated by Nathan Cutler over 5 years ago
- Copied to Backport #21920: luminous: sparse-reads should not be used for small IO requests added
#6 Updated by Jason Dillaman almost 5 years ago
- Status changed from Pending Backport to Resolved