[LU-11916] LSOM: lazy_stat mount option not recognized Created: 01/Feb/19 Updated: 15/May/20 Resolved: 15/May/20 |
|
| Status: | Resolved |
| Project: | Lustre |
| Component/s: | None |
| Affects Version/s: | Lustre 2.12.0 |
| Fix Version/s: | None |
| Type: | Bug | Priority: | Minor |
| Reporter: | Stephane Thiell | Assignee: | Qian Yingjin |
| Resolution: | Won't Fix | Votes: | 0 |
| Labels: | None | ||
| Environment: |
CentOS 7.6, MOFED 4.5 |
||
| Issue Links: |
|
||||||||||||
| Severity: | 3 | ||||||||||||
| Rank (Obsolete): | 9223372036854775807 | ||||||||||||
| Description |
|
I was trying to use "LSOM enforced" on a 2.12 client (for testing purpose, not very important), and noticed the lazy_stat option doesn't seem to work: [794713.969118] LustreError: 152-6: Unknown option 'lazy_stat', won't mount. This new mount option was mentioned in Li Xi's presentation at LUG'18 (page 7) http://cdn.opensfs.org/wp-content/uploads/2018/04/Xi-Lazy_Size_on_MDS_DDN.pdf Is there a way to enforce LSOM on a Lustre client, or not yet? |
| Comments |
| Comment by Peter Jones [ 01/Feb/19 ] |
|
Qian Can you comment as to whether this was included in the feature as it ended up in 2.12? Peter |
| Comment by Andreas Dilger [ 01/Feb/19 ] |
|
The "lazy_stat" mount option is not in the 2.12 release. The closest is patch https://review.whamcloud.com/33412 The current LSOM implementation isn't just "the size is a bit outdated", it is "the size and blocks stored on the MDS are zero until after the file is closed", so this would likely break many applications, and this can only be set on a per-mountpoint basis. Could you explain more how you want to us LSOM? The current implementation is mostly intended for MDT-level scanning. There is also a patch to add support for "lfs find" that is fairly safe, and discussion around adding support for statx(), which is what I think applications should use when they want the lazy size. For RBH usage, the 33412 patch and a separate mountpoint may be enough, as long as you don't trust the size on recently-created files. |
| Comment by Stephane Thiell [ 01/Feb/19 ] |
|
Thanks Andreas for the information and pointers, this is super helpful. I understand the interest of LSOM for MDT-level scanning, but I wanted to see how it would break (or not) applications if also enforced for (regular) stat(). |
| Comment by Andreas Dilger [ 15/May/20 ] |
|
The statx() API is going to land soon, and "lfs find" already can understand LSOM. |
| Comment by Stephane Thiell [ 15/May/20 ] |
|
Excellent, thanks! |