All responses agreed on the cause; Thanks to:
etxstfr@helix.ericsson.se (Stefan Frick)
Chris Lopez <clopez>
dvsc.a@minster.york.ac.uk
b28384@achilles.ctd.anl.gov (Ray Carlson)
hamilton@kickapoo.cs.iastate.edu (Jon Hamilton)
rjd4@cus.cam.ac.uk (Bob Dowling)
I will merely reproduce Bob Dowlings response below as it gives a
comprehensive summary and solution for the problem:
Do you have a CD inserted? If you do then this is bug 1133445, whose solution
(from Sun) I post below. Sun have taken option 3 of the set below and used it
in Solaris 2.3 which perhaps your experimental system should be running now.
--------
Bob Dowling: UNIX Support,
University of Cambridge Computing Service,
rjd4@ucs.cam.ac.uk New Museums Site, Pembroke Street,
+44 223 334728 Cambridge, UK. CB2 3QG.
--- Response from Sun when I asked. Pick one of the three. ---
1. umount the cdrom every night (not really suitable!)
2. Change the find command to be as follows:
find / -name .nfs\* -mtime +7 -exec rm -f {} \; -o -fstype nfs -prune -fstype hsfs -prune
3. Follow the directions in SRDB 5853 (included below)
_______________________________________________________________________________
SRDB ID : 5853
SYNOPSIS : root's find in crontab hangs system overnight
DETAIL DESCRIPTION :
The following line appears in /var/spool/cron/crontabs/root:
15 3 * * * find / -name .nfs\* -mtime +7 -exec rm -f {} \; -o -fstype nfs -prun
e
This runs a find command daily at 3:15 am that looks for .nfs* files
and removes them if they're more than a week old. These files are created
by client renames if a remove is attempted on an open file.
If an hsfs cdrom is mounted on the system, this job will cause the machine
to freeze soon after it starts at 3:15 AM. The result will be a hung
machine when people attempt to login.
SOLUTION SUMMARY :
This find is unnecessarily wasteful of system resources for several reasons:
o It does not distinguish between servers and non-servers. The find
is wasted on non-servers.
o It searches filesystems that are not exported.
o The "-fstype nfs" primitive forces the find command to use the statvfs()
call on every file. Normally just lstat is called.
o The -prune facility does not work for autofs mounts. The find touches
direct autofs mountpoints and forces their mounts.
o It is run daily, even though it removes only those .nfsxxx files that
are a week old.
This process could be vastly improved with the addition of a simple
wrapper script around the find (e.g., /usr/lib/fs/nfs/nfs_find.sh)
#!/bin/sh
# Check shared NFS filesystems for .nfs* files that
# are more than a week old.
#
# These files are created by NFS clients when an open file
# is removed. To preserve some semblance of Unix semantics
# the client renames the file to a unique name so that the
# file appears to have been removed from the directory, but
# is still usable by the process that has the file open.
if [ ! -f /etc/dfs/sharetab ]; then exit ; fi
for dir in `awk '$3 == "nfs" {print $1}' /etc/dfs/sharetab`
do
find $dir -name .nfs\* -mtime +7 -mount -exec rm -f {} \;
done
This script invokes the find command only on exported filesystems.
Machines that do not export filesystems (non-servers) will not run
the find command at all.
Without the -fstype primitive the find runs 25% faster. The modified
find also does not force direct autofs mounts to be mounted.
To install:
become root
save the script as /usr/lib/fs/nfs/nfs_find.sh
chmod +x /usr/lib/fs/nfs/nfs_find.sh
EDITOR=/usr/ucb/vi;export EDITOR
crontab -e
Find the line in the crontab file that is running find at 3:15 am.
Change it to:
15 3 * * * /usr/lib/fs/nfs/nfs_find.sh >/dev/null 2>&1
This should do it!
Script and solution taken from Bugid 1113177.
PRODUCT : System_Crash
SUNOS RELEASE : 2.2
UNBUNDLED RELEASE : n/a
HARDWARE RELEASE : Sun4
ISO-9001 STATUS : Uncontrolled
_______________________________________________________________________________
This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:08:29 CDT