On Wed, Mar 12, 2008 at 12:27 AM, Paul Armstrong <opensolaris at otoh.org> 
wrote:
>  These should also be strongly considered:
>  /var/crash
>  /var/core (a common place to put core files using coreadm)

Using the name "core" as a directory name makes it so that anything
with a cwd of /var (or /var/share) cannot do per-process core dumps.

I currently configure systems so that they *try* to core dump to
/var/cores/core.<some complex string>.  Under most circumstances
/var/cores does not exist - else it would be easy for users to
accidentally fill up the file system in such a way that only the
system administrator can fix it.  This has the nice side effect that I
get a syslog message saying that the system could not write a core
file to /var/cores - in other words I get a message indicating that
programs are crashing.  This shortens the time to answer things like
"which system administrator keeps killing processes on my server?"

I think, and Sun's best practices agree, that having such a directory
for catching all core dumps is problematic because of the space
consumption issue.  I am happy to have people do it if it brings more
value than harm to them, but I don' t think it is right as a default.

I agree that /var/crash belongs there.  And several other things
probably do too.  See
http://mail.opensolaris.org/pipermail/caiman-discuss/2007-August/000737.html
for a somewhat comprehensive list of /var directories that are not
controlled by the package database and as such should be considered as
to whether you really want a copy that is private to a boot
environment.

-- 
Mike Gerdts
http://mgerdts.blogspot.com/

Reply via email to