Re: Best file system

[Chris Knadle <Chris.Knadle@coredump.us>]

On Wednesday 01 February 2006 12:12, Sylvestre Ledru wrote:
> While we are talking about filesystems, I am using a server with an ext3
> filesystem. Due to a bug (software, me, client... I don't know), the
> zeodb database was stored into the temporary directory (/tmp/) and
> deleted...
> As there is no way of doing an undelete with ext3, I dump the partition
> with dd (dd if=/dev/hda7 of=/home/dump_temp.dump bs=1048576 count=1024).

   At least one admin was able to mount an ext3 partion as ext2 and then 
undelete files from there.  Sounds like a crazy idea, and it might be too 
late depending  on what you've attempted already:
	http://lists.debian.org/debian-user/2005/01/msg00840.html
	http://lists.debian.org/debian-user/2005/01/msg00885.html

   Also, the straight 'dd' doesn't actually copy everything, but hopefully it 
got what you needed.  For next time, have a look at dcfldd, which supposedly 
does a much better job (and thus takes longer to copy):
	http://dcfldd.sourceforge.net/



Throwing in my anecdote on the subject of ext3:

   Sometime in 2003 the company I worked for had a server room with an 
overloaded circuit which resulted in repeated power failures when the circuit 
breaker tripped.  The mail server that ran ext3 failed to come back up after 
one too many of these events, causing file system corruption in /var such 
that accessing a particular section would cause a kernel panic, which 
occurred on every full reboot.
   We were able to use debugfs to slowly copy most of the data that was 
in /var, but even debugfs would generate a kernel panic when the troubled 
section was hit.  /var/spool/mail was a softlink to another partition, so 
that wasn't a concern.

   The point here is that ext3 is pretty good, but it too is not invulnerable 
when it comes to unexpected power failures.

	- Chris

-- 

Chris Knadle
Chris.Knadle@coredump.us