You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
This issue will gather NSFS object operations optimizations -
Small objects performance
Actual behavior
As for today, on POSIX file systems, users experience low performance when writing small objects due to rename() syscall taking a lot of time -
2022-01-24 13:56:57.560361 [PID-14/TID-21] [L0] FS::FSWorker::Execute: WARNING Rename _old_path=/nsfs/fs1/cb6/.noobaa-nsfs_61b11a909954d0002a5d14fd/uploads/4a0c95ec-30ad-4161-8fa6-150f800e851e _new_path=/nsfs/fs1/cb6/586_64KB took too long: 881.188 ms
We solved this issue on GPFS filesystems by using libgpfs.
As mentioned on #6858, a POSIX optimization would be to try using link() instead of rename() when the target file is missing, and if there is a race then fallback to rename. That should allow reduce the rename overhead from most of the POSIX calls that do not involve replacing an existing target object.
Expected behavior
Writing of small object performance optimized for POSIX as well.
Environment info
This issue will gather NSFS object operations optimizations -
Small objects performance
Actual behavior
We solved this issue on GPFS filesystems by using libgpfs.
As mentioned on #6858, a POSIX optimization would be to try using link() instead of rename() when the target file is missing, and if there is a race then fallback to rename. That should allow reduce the rename overhead from most of the POSIX calls that do not involve replacing an existing target object.
Expected behavior
Steps to reproduce
The text was updated successfully, but these errors were encountered: