Re: Which file-system is good for power down?

[Adam Hamsik <haaaad%gmail.com@localhost>]

On Jan,Monday 7 2008, at 1:18 PM, Christian Baer wrote:

> On Mon, 7 Jan 2008 09:46:56 +0100 Adam Hamsik wrote:
>
> > > The difference is that soft updates (softdeps) will save any data  
> > > that
> > > actually did make it to the disk prior to the power going down. In
> > > case of ext3 that is not necessarily the case. Especially if the  
> > > load
> > > on the  drive is heavy and it doesn't support tagged command  
> > > queuing. I
> > > could happen that data is actually written to the disk but can't be
> > > found because  the meta data wasn't written in time.
>
> > This is not true metadata must be committed before system can write
> > any data to fs(in ordered mode data are keeped in transactions and  
> > can
> > be written to disk only when transaction is in commiting state).
>
> Ok, I guess I used the word "metadata" the wrong way. I also  
> considered
> the journal to be metadata.
>
> The way I understand journaling is (in a very primitive modell). A  
> file is
> to be created and written to disk. The system does these steps:

You can create journal on external device, too.

> - An entry is made in the journal stating that the given file is to be
>  created.

Entry in journal is created from modified filesystem blocks e.g.  
directory inode and other fs-information.  you can journal non fs data  
with ext3 ,too

> - The file is created (written to disk), the directory is updated as  
> are
>  the inode and the other fs-information.

yes
> - An entry is made in the journal stating that the file was created
>  successfully.

yes

If you have had hard crash after entry 1 is done but 2 not you don't  
care because filesystem is consistent and nonfs data wasn't written yet.

If you have had hard crash after entry 1 is done and 2 partially you  
have to redo this journal entry and all entries after it.

If you have had hard crash after entry 1 and 2 was done  don't care  
because filesystem is consistent.

> This list is (as stated) very rough and the steps are probably  
> broken down
> far more like updating the journal for the entry in the directory and
> inodes etc.
>
> The way I understand it, any data that is actually written to the disk
> without the journal being updated is lost. Or did I get that wrong?
>
> > Journaling and softdeps are different techniques for keeping
> > filesystem in consitent state after reboot, their task is not protect
> > data.
>
> Depends on how you define "protection". :-) If a filesystem is  
> completely
> broken after a crash you could lose a lot more data than was still  
> in the
> memory. Being able to restore the file system and the stored data is
> protection in a way. :-)

Regards

Adam.