On Thu, 2008-07-17 at 10:11 +1000, Tony Sceats wrote:
> >
> >
> >
> > Ah, now this all depends.  If /var/run was a normal disk filesystem
> > under Ubuntu you would be correct.  It isn't, though:
> >
> > ] mount | grep /var/run
> > varrun on /var/run type tmpfs (rw,noexec,nosuid,nodev,mode=0755)
> >
> > Note 'type tmpfs' there?  tmpfs is a swap backed ramdisk, essentially,
> > and like /var/lock is put there by Ubuntu to ensure that these
> > directories are cleaned at reboot without having to fuss about deleting
> > files or read-only filesystems.
> 
> 
> hey, that's pretty cool, can't say I'm familiar with tmpfs  - does it
> pre-allocate the space or grow to the specified size as it needs to? I ask
> because clearly 2Gb is an enormous amount of RAM/SWAP to spend on /var/run
> just so you don't have to clean it at boot time (which can be done
> trivially)
> 
> > imho, 2Gb is either way too big or too small depending on what the
> computer
> > does.. try 'du -hs /var/run/*' to see what else is happening there - for
> me
> > it's usually less than 1Mb, but sometimes spools (eg, mail spools) will be
> > put there which will increase it a lot,
> 
> The mail spool lives under /var, but not under /var/run.  The FHS, in
> > fact, carefully defines the purpose of /var/run -- and that does not
> > include being scratch space for large processes.
> 
> 
> FHS aside (because I know you're right about it), I know I've seen spools
> there, there used to pretty much always be a /var/run/spool directory, and I
> know this has caused me problems, many moons ago now though, that's for
> sure, probably on RH6 or Debian Potato
> 
> 
> Y'all should file a bug report indicating that Mondo should use /tmp or
> > /var/tmp to build big scratch files.
> 
> 
> Indeed, this would be very annoying, space aside - I'm not actually familiar
> with this app, but given the in-built inpermanence of /var/run, this seems
> like a very quick way to lose hours of hard work!


Now please excuse my ignorance, but that makes it sound as if
Ubuntu's /var/run is actually held in RAM, and not on disk at all?

My system is a clean install of Hardy server and the install didn't ask
for a size for /var/run so it must have been automated. There is 4G of
Ram, so if the memory is allocated as needed, then 2G makes some sense
if /var/run is used for performance reasons. 

The problem only arises with apps like mondo that assume /var/run will
be large. I haven't seen the problem before. I used mondo on Feisty and
Sarge and the question never arose.

Given that there is an option in mondoarchive to specify another path,
should I report this as a bug? It kept me busy for several hours before
the penny dropped as to why I was getting disk full errors.

David.

-- 
SLUG - Sydney Linux User's Group Mailing List - http://slug.org.au/
Subscription info and FAQs: http://slug.org.au/faq/mailinglists.html

Reply via email to