Re: [PATCH 2/2] xfs: don't truncate prealloc from frequently accessed inodes

Re: [PATCH 2/2] xfs: don't truncate prealloc from frequently accessed inodes

[Andi Kleen]

Dave Chinner <david@fromorbit.com> writes:
>
> To avoid this problem, keep a count of the number of ->release calls
> made on an inode. For most cases, an inode is only going to be opened
> once for writing and then closed again during it's lifetime in
> cache. Hence if there are multiple ->release calls, there is a good
> chance that the inode is being accessed by the NFS server. Hence
> count up every time ->release is called while there are delalloc
> blocks still outstanding on the inode.

Seems like a hack. It would be cleaner and less fragile to add a
explicit VFS hint that is passed down from the nfs server, similar
to the existing open intents.

-Andi

-- 
ak@linux.intel.com -- Speaking for myself only.

Re: [PATCH 2/2] xfs: don't truncate prealloc from frequently accessed inodes

[Dave Chinner]

On Mon, Nov 29, 2010 at 10:42:29AM +0100, Andi Kleen wrote:
> Dave Chinner <david@fromorbit.com> writes:
> >
> > To avoid this problem, keep a count of the number of ->release calls
> > made on an inode. For most cases, an inode is only going to be opened
> > once for writing and then closed again during it's lifetime in
> > cache. Hence if there are multiple ->release calls, there is a good
> > chance that the inode is being accessed by the NFS server. Hence
> > count up every time ->release is called while there are delalloc
> > blocks still outstanding on the inode.
> 
> Seems like a hack. It would be cleaner and less fragile to add a
> explicit VFS hint that is passed down from the nfs server, similar
> to the existing open intents.

Agreed.

However, we've been asking for the nfsd to change it's behaviour for
various operations for quite some time (i.e. years) to help
filesystems behave better, but and we're no closer to having it
fixed now than we were 3 or 4 years ago. What the nfsd really needs
is an an open file cache so that IO looks like normal file IO rather
than every write being an "open-write-close" operation....

While we wait for nfsd to be fixed, we've still got people reporting
excessive fragmentation during concurrent sequential writes to nfs
servers running XFS, so we really need some kind of fix for the
problem...

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com