Hey, you are the best.
My `while(true); do ls -al; done` leads to great result. Now I have to
transfer and test it in my "productive" environment to see if it behave
similar. But this take some time.
Two thoughts about the solution:
- Do you have an idea about the performance advantage of the cache or
performance lost?
- What's about setting udba to none in this case?
Best regards and thank you a lot
Tobias Wilken
2009/12/9 <[1][email protected]>
> No, aufs has no timeout for the inode attribute.
I remind myself one thing.
Aufs has one timeout (hidden) option for readdir(3).
When you call readdir(3), aufs internally calls readdir(3) for each
branches, merges the results, and return the merged result to userspace.
Here the merged results is cached for 10 seconds (by default).
If any file is created or removed from the dir in aufs world, then aufs
maintains the cache and you can see the new file or cannot see the
removed one.
But if such creation/deletion happens bypassing aufs and the inotify
event is not be fired, then you will see the obsoleted filename (or
cannot see the filename) for 10 seconds.
If this is your case, try the aufs mount option 'rdcache=0' which will
force aufs not to use the cached readdir result and to call readdir for
each branch everyt ime.
J. R. Okajima
References
1. mailto:[email protected]
------------------------------------------------------------------------------
Return on Information:
Google Enterprise Search pays you back
Get the facts.
http://p.sf.net/sfu/google-dev2dev
