Purging files from slave nodes

[Donald Becker]

On Wed, 27 Mar 2002, Jakob Østergaard wrote:

> On Mon, Mar 25, 2002 at 12:22:39PM -0700, Arthur H. Edwards wrote:
> > Is there a command to purge files (I know bpsh -a rm /tmp/*.* doesn't work)
> > to remove files from slave nodes? Other than reformatting, can one clean up?
> Please note that rm /tmp/*.*  will only delete files containing a period in
> their names...

More subtlety, doing
   bpsh -a rm /tmp/*.*
will do filename globbing on the originating machine (the master).  Thus
it will only clear filenames that exist both on the master and on the slave.

> Also, tmpwatch does this job in most distributions already, fully
> automated. It will delete any file in /tmp that hasn't been accessed
> for a number of days (typically 30 I think, but easily adjustable).

The tmpwatch program is reasonable for cleaning up "user droppings" in a
workstation environment, but it isn't as useful on a cluster with
per-node /tmp directories.  Most programs usually clean up their /tmp files.
Those that don't tend to spew forth too many files to handle with a 30
day delay.

Note that in the default Scyld system the slaves run no extraneous
daemons such as 'cron' or 'at'.  Thus 'tmpwatch' (if activated) is run
only on the master.

-- 
Donald Becker				becker at scyld.com
Scyld Computing Corporation		http://www.scyld.com
410 Severn Ave. Suite 210		Second Generation Beowulf Clusters
Annapolis MD 21403			410-990-9993