Steve Lord [lord@xxxxxxx] wrote:
> On Tue, 2003-07-29 at 14:12, Steve Lord wrote:
> > I think you got bitten by stripe alignment. Inode clusters are allocated
> > on stripe boundaries. You probably have no boundaries left free, so
> > it cannot allocate any inode space.
> Actually, you are out of inode room. We actually allocate inodes in
> blocks of 64 - which is 4 fs blocks in the default setup. We deal
> with them in memory in chunks of 2 fs blocks, but on disk we ask for
> 64 at a time.
It would be nice if it can fallback to smaller chunks. Ideally down to 1 fs
block. Or reduced the report of free inodes (df -i).
It is somewhat ugly to get a file system full error when df and df -i
reports free space avialable. There are applications (caches) which dont
like such situations.
But i consider that to be a small bug when it occurred at 98% usage. But a
big one at 80% usage. I was bitten by this in the past (solaris ufs).
> So, working out how not to fragment so much in the first place would
> be the best solution here.
Fragmention of data or free space?