I have two public load balanced webservers and one private database server.
The database server runs nfsd and is nfs mounted by the two webservers.
I am adding a backup database server so that when the master database fails,
the backup takes over. No problems there.
The part I am having difficulty with is the nfs mounts. It is no secret that
when the nfsd goes down, it becomes a headache to umount the directories that
relay on this.
I have tried everything that I have been able to find in my searching. My
most successful attempt was like this:
- the two databases use carp and ifstated, when the master database fails and
the backup database takesover, it triggers a script in ifstated.
- The script will ssh into each webserver, umount the nfs mounts then mount
them again with the backup database as the target.
When the initial mount used "mount -t nfs ..." the umount would just hang.
When the initial mount used "mount_nfs -is ..." the umount would work only
sometimes, with no obvious pattern of when it would work or not.
I will consider any suggestion, or alternative. Is OpenAFS an option?
I am just looking for a way to have failover for network file storage.
Thanks,
- Tom
-------------------------------------------------------
# cat /etc/ifstated.conf
init-state auto
carp_up = "carp0.link.up && carp1.link.up"
carp_down = "!carp0.link.up || !carp1.link.up"
state auto {
if $carp_up
set-state master
if $carp_down
set-state backup
}
# in this state we are the master mysql server, and the nfs mount
state master {
init {
# start nfs posse
#run "/usr/sbin/portmap"
#run "echo -n>/var/db/mountdtab"
#run "/sbin/mountd"
#run "/sbin/nfsd -tun 4"
#run "/bin/sleep 20"
# go through each client using our nfsd and umount them
# from their current nfs mounts and remount them using us
run "/etc/ifstated_scripts/remount_nfs.sh"
# let things settle
run "/bin/sleep 20"
}
if $carp_down
set-state backup
}
# in this state we are just waiting to take over
state backup {
init {
# stop nfs posse (in reverse order)
#run "ps -acx | grep nfsd | awk '{print $1}' | xargs kill "
#run "kill -TERM `cat /var/run/mountd.pid`"
#run "ps -acx | grep portmap | awk '{print $1}' | xargs kill
"
# let things settle
run "/bin/sleep 20"
}
if $carp_up
set-state master
}
--------------------------------------------------------
# cat /etc/ifstated_scripts/remount_nfs.sh
#!/bin/sh
# when using master nfs server, ip=192.168.20.32
# when using backup nfs server, ip=192.168.20.31
# webserver (nfs-client) ips = 192.168.20.12, 192.168.20.11
# the same test directory exists on all nfs-server and nfs-client servers
# /backup/nfs_test_mount
NFS_SERVER_IP=192.168.20.32
SSH=/usr/bin/ssh
MOUNT=/sbin/mount_nfs
UMOUNT=/sbin/umount
SLEEP=/bin/sleep
ECHO=/bin/echo
# list all directories that haves to me umount'd
NFS_MOUNTS="/backup/nfs_test_mount"
for machine in 192.168.20.12 192.168.20.11
do
$ECHO "Logging into $machine ... ";
for directory in $NFS_MOUNTS
do
NFS_UNMOUNT="$UMOUNT -f $directory"
NFS_MOUNT="$MOUNT -si $NFS_SERVER_IP:$directory $directory"
$ECHO "- unmounting $directory"
$SSH [EMAIL PROTECTED] "$NFS_UNMOUNT"
$ECHO "- sleeping zZzZ"
$SLEEP 20
$ECHO "- mounting $directory"
$SSH [EMAIL PROTECTED] "$NFS_MOUNT"
done
done
_________________________________________________________________
Spell a grand slam in this game where word skill meets World Series. Get in
the game.
http://club.live.com/word_slugger.aspx?icid=word_slugger_wlhm_admod_april08
