filesystem for mail

[Michael Hipp]

Kurt Wall wrote:
> On Mon, Oct 16, 2006 at 10:01:01PM +0100, Jorge Almeida wrote:
>> On Mon, 16 Oct 2006, Kurt Wall wrote:
>>
>>> XFS. I don't think Reiser's arrest will necessarily affect ReiserFS, but
>>> I would opt for XFS for reasons of technical superiority and speed.
>>>
>> Any reliability problems, like the ones mentioned by Michael?
> 
> Not here. At the end of the day, if you're cranking on a file system
> and the power goes away, there will be transactions in core that never
> make it to the on-disk journal. I'm not aware of _any_ file system that
> guarantees data integrity in such a situation. Rather, it will just roll 
> back to the last known good state. What good journaling does is minimize
> the amount of data loss and offer fast recovery when it does happen.
> Nevertheless, I've _never_ lost data on XFS, including when I
> deliberately yanked the powe during a kernel compile to see what would
> happen.

Kurt, you're absolutely correct.

There's no filesystem that can ensure things make it to the disk when 
the power dies. But what I wasn't expecting was that the filesystem 
itself would be corrupted and evidently not able to be rolled back to 
the last known good state.

Correct me if I'm wrong, but the journal feature should automatically 
handle the rollback with no intervention from the user. The xfs_repair 
function is needed when the filesystem is actually corrupt and must be 
repaired.

Michael