[Dirvish] server hard locks running dirvish-expire (where to start??)

Jason Boxman jasonb at edseek.com
Wed Apr 12 23:22:59 EDT 2006


On Wednesday 12 April 2006 23:14, foner-dirvish at media.mit.edu wrote:
<snip>
> Note in particular [2] below, which talks about what happens if you
> put a few RSFS images -as files- into another RSFS and then try to
> fsck the result---you can kiss everything involved goodbye.  You
> haven't done that here (I presume), but consider just how much disk
> traversal RSFS is going through in order to reconstruct that enormous
> B-tree.
>
> And [1] points out that, as far as I'm concerned (and I obviously
> differ w/Hans on this), RSFS is broken by design, because any
> unexpected shutdown will randomly swap data between all open files,
> while leaving "correct" metadata---so your fsck is instant, but your
> data is hopelessly scrambled anyway in a way you can't automatically
> detect (short of checksumming every file against a known-good
> backup of checksums, followed by restoring backups for every mismatch).

The whole discussion also indicates that essentially all Linux journaled 
filesystems, by default, only journal metadata.  ext3 gives you the option of 
journaling data and metadata at the expense of speed.

That said, I still won't use reiserfs as I don't really like userspace fsck 
tools that crash.  That's just bad.

-- 

Jason Boxman
http://edseek.com/ - Linux and FOSS stuff



More information about the Dirvish mailing list