Description
For applications that expect NFS-like behavior with data flush on file close, and do not call fsync() on the file, it would be useful to provide a per-mountpoint tunable parameter and mount option that forces file data to be sync'd to the target, something like "llite.*.sync_on_close" and "-o sync_on_close", or similar. Better would be a per-directory flag that is inherited by files created therein, so that applications or administrators can set this for specific apps/users that need such semantics.
This would have a potentially large impact for workloads that are writing a large number of small files (e.g. tar), so it should default to "off", and only be enabled specifically for workloads that expect it.
Attachments
Issue Links
- is related to
-
LU-17329 Relaxed POSIX Consistency for Lustre
-
- Open
-
Activity
"Andreas Dilger <adilger@whamcloud.com>" uploaded a new patch: https://review.whamcloud.com/c/fs/lustre-release/+/58515
Subject: LU-16049 llite: struct cleanup and option ordering
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 25045bf84fded5ae124a28cba99b41dd403027c2
"Andreas Dilger <adilger@whamcloud.com>" uploaded a new patch: https://review.whamcloud.com/c/fs/lustre-release/+/58508
Subject: LU-16049 llite: add sync_on_close option
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 909ccb828776d3242cdf60c8a05d75d3d9961e0b
"Oleg Drokin <green@whamcloud.com>" merged in patch https://review.whamcloud.com/c/fs/lustre-release/+/58515/
Subject: LU-16049 llite: struct cleanup and option ordering
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: a90bf7aaf224786a4c61c0f0418733086ee29f01