[Lustre-discuss] inode size with mkfs

[Andreas Dilger]

On 2010-03-09, at 17:20, Stuart Midgley wrote:
> After installing e2fsprogs-1.41.6.sun1-0redhat it appears the  
> default inode size for ext file systems changes from 128 to 256  
> (see /etc/mke2fs.conf).  This caused us several days of "head  
> scratching" as we were trying to figure out why grub wouldn't work  
> on our oss's where we had this package installed.
>
> To resolve the issue, we had to build our /boot partition with
>
>   mkfs.ext2 -I 128 ...
>
> which set the inode size back to 128 for this partition and grub was  
> happy again.

This seems a bit confusing, since I would imagine that /boot is  
formatted with the default OS version of e2fsprogs, and the Lustre- 
patched version is installed after initial installation?

> So... why was it changed as the default for ALL ext file systems?   
> Surely, an entry could have been put in for lustre with the correct  
> settings?  ie. mkfs.ext3 -T lustre ....
>
> Stopping something as fundamental as grub from working doesn't seem  
> like a good thing for a package to do.


This is actually a change in the upstream e2fsprogs-1.41, and was not  
introduced in the -sun1 patched version.  It was done because ext4 is  
now using the extra inode space for nanosecond timestamps and xattrs,  
though I grant that both of these features were first developed for  
Lustre.

What version of GRUB are you using?  I know that FC12 is installing  
ext4 by default, so I imagine that they are using 256-byte inodes with  
GRUB.

Cheers, Andreas
--
Andreas Dilger
Sr. Staff Engineer, Lustre Group
Sun Microsystems of Canada, Inc.