Re: Change Filesystem
From: Lee Sau Dan (danlee_at_informatik.uni-freiburg.de)
Date: 12/09/04
- Next message: Mike Mol: "Re: Data recovery/kernel hangs"
- Previous message: Lee Sau Dan: "Re: Change Filesystem"
- In reply to: Dances With Crows: "Re: Change Filesystem"
- Next in thread: John Thompson: "Re: Change Filesystem"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Date: 09 Dec 2004 18:02:33 +0800
>>>>> "Dances" == Dances With Crows <danSPANceswitTRAPhcrows@gmail.com> writes:
>> Mh, you are aware that xfs is a journaling fs, recovering in a
>> second from an unplanned shutdown.
Dances> That's the theory.
No, it isn't. The theory (of journaling *filesystem meta-data*) only
says that the structure of the filesystem can be recovered easily
after a crash. It doesn't say that the file CONTENTS can be
recovered. If you want the file contents to survive a crash, you need
a fs that supports *data-journaling* and you usually need to
explicitly enable it.
Dances> ISTR reading something somewhere (one of the
Dances> comp.os.linux.* groups, or possibly Slashdot) written by
Dances> someone who kept a lot of data on XFS partitions and lost
Dances> most of it to an unplanned shutdown.
Not surprising, if data-journaling is not enabled.
Did that someone tell you the filesystem becomes inconsistent (i.e.
with unclaimed blocks, cross-linked blocks, or references to
non-existent blocks) after the crash? I guess if he did, the answer
would be NO. The meta-data journal only guarantees that the data
structure can be recovered. It can do nothing with the file CONTENTS.
So, that guy should not have lost any files. He only had corrupted
file contents. This is consistent with the theory: you journal the
meta-data: the meta-data (==files and directories and
permissions/ownerships) can be recovered. You don't journal the data
(==file contents) : you can't always have the file contents recovered.
Dances> Maybe the commit interval was set too high, maybe it was a
Dances> bug in XFS that has since been fixed. People have made
Dances> the same data-loss complaints about ReiserFS.
Has anybody report a corrupted filesystem, i.e. data *structures* in
the filesystem to be in an inconsistent state? If not, then you can't
say the fs code is buggy based on this.
Again, journal (of meta-data only) guarantees the the filesystem has a
consistent data structure, but not correct file contents.
Dances> IME, ReiserFS and ext3 deal with power loss situations
Dances> very well, though. 3 power outages this year at my
Dances> apartment, ReiserFS and ext3 partitions on 2 desktop
Dances> machines, no data loss ever.
That also agrees with my personal experience. However, knowing that
I'm not using data-journal (not available in reiser3, not enabled in
ext3 by default), I still know that I'm vulnerable to loss of *file
contents*. I know the journal (of meta-data) is only good to
guarantee that the filesystem *data structures* are safe against
crashes, which is quite enough for me.
--
Lee Sau Dan §õ¦u´° ~{@nJX6X~}
E-mail: danlee@informatik.uni-freiburg.de
Home page: http://www.informatik.uni-freiburg.de/~danlee
- Next message: Mike Mol: "Re: Data recovery/kernel hangs"
- Previous message: Lee Sau Dan: "Re: Change Filesystem"
- In reply to: Dances With Crows: "Re: Change Filesystem"
- Next in thread: John Thompson: "Re: Change Filesystem"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Relevant Pages
|
|
