Unionfs and nullfs question

[Marc G. Fournier]

On Fri, 22 Oct 2004, Vulpes Velox wrote:

> I am guessing the answer, given the big warning in the man for both is 
> that this is going to be a no that both are dangerous to the data and 
> luck is mainly involved in not having the data screwed over, but just 
> wanna check :)

I run over 200 VPSs over 4 machines with all application data (installed 
ports) mounted through unionfs to reduce disk space usage ... every once 
in a blue moon, I'll get a crash resulting from a bug in the unionfs code, 
but it isn't as bad as it was, say, a year ago ... but I am running 
production servers with it.

There are a few things you can't do right now ... for instance, I don't 
have /var union mounted, as FIFO's/sockets tend to consistently blow it up 
... but, my more loaded server looks like:

# df -t union | wc -l
       73
# uptime
11:41AM  up 47 days, 22:25, 1 user, load averages: 12.12, 20.67, 22.46

There is an annoying 'bug' in fsck that Don Lewis has been working on 
correcting that is very exasperated by unionfs ... namely how the list of 
inodes to check is generated.  If you, for instance, mount a blank file 
systems over top of /usr/ports, and then do a find of /usr/ports, the 
blank file system will fill up with a bunch of directories to 'mirror' 
ports ... the files don't come through, only the directories.  On a crash, 
the OS leaves behind a bunch of ZERO LENGTH DIRECTORIES ... I've had fsck 
run for 12-14hrs after one of these, its that messy :(  Don has been 
working on a patch to handle the ZLDs better, but it hasn't been committed 
to -stable yet, pending more testing ... I'm running it live here, but 
*knock on wood* haven't had a crash since putting it into place ...



----
Marc G. Fournier           Hub.Org Networking Services (http://www.hub.org)
Email: scrappy at hub.org           Yahoo!: yscrappy              ICQ: 7615664