[Linux-cluster] ext3 filesystem

[Robert Peterson]

Graham Wood wrote:
>
>> You're right--the other nodes won't be informed of file system changes,
>> so they should all mount ext3 read-only.  Better to use GFS.
> Unless I'm really missing something - this still isn't going to work.
>
> To give a simple example:
>
> 1) Create a 10GB file on the filesystem using node1
> 2) Mount it read only on node2, and look at the start of the file.
> 3) Delete the file on node1, and copy /etc/shadow onto the disk 200 times
> 4) less the file again on node2 (and look further in) - you're 
> reasonably likely to see the contents of /etc/shadow
>
> Whenever I've seen this sort of thing discussed in the past, the 
> comment has been that it has to be mounted read-only EVERYWHERE.
>
> Graham
Hi Graham,

That's what I was trying to say in the "they should all" quote above. 
If it's ext3, it has to be mounted read-only everywhere because changes 
to the file system will pull the
rug out from under ext3 on the other nodes, with unpredictable results. 
If it's GFS, no problem sharing the file system.

Regards,

Bob Peterson
Red Hat Cluster Suite