[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: Some inode questions



On Tue, 2009-05-05 at 15:59 -0600, Andreas Dilger wrote:
> On May 05, 2009  11:40 -0700, Ross Boylan wrote:
> > When I first create /var I took all the defaults.  I have since decided
> > that, since it will hold a cyrus mail spool (each message is a file) I
> > should use something with more inodes.  I created a new (var2) partition
> > and formatted it with 
> > # mkfs.ext3 -T news /dev/mapper/turtle-var2_crypt
> > # news has inode_ratio = 4096
> > 
> > Then I mounted and rsync'd from my existing /var.
> > Afterwords, I get a report that seems to indicate I've used almost no
> > inodes.  It also shows more inodes than blocks; is there any way one
> > could need more than one inode/block?
> 
> Hard links, or empty files...
> 
> > As I read this, 6291445 of 6291456 inodes are free, so 11 are in use.
> > The comparable calculation on the origin file system shows about 8,500
> > inodes in use.
> 
> Indeed, it seems your new filesystem is empty.  That said, the superblock
> contents are not updated on disk while the filesystem is mounted.  I have
> argued that since we are already computing the superblock totals and
> storing them into the superblock it wouldn't be harmful to write the
> superblock to disk occasionally in ext[34]_statfs() by calling at the end:
> 
> 	ext[34]_commit_super(sb, es, 0);
> 
> I don't think there is currently anything in ext[34] that is writing
> the superblock to disk at all, except mount and unmount.
> 
> Using "df -i" should give you accurate numbers for a mounted filesystem.
> 
> Cheers, Andreas
Thank you.  df -i does look sane, and after umount, dumpe2fs does also.
There were actually slightly more inodes in use on the original (var)
than the copy (var2), even right after an rsync.  I'm guessing that
might be files that are open but deleted.
Ross



[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]