# /etc/sysconfig/nfs:
# STATD_HOSTNAME="myhostname -H /etc/ctdb/statd-callout"
-. /etc/ctdb/functions
-loadconfig nfs
+[ -z "$CTDB_BASE" ] && {
+ export CTDB_BASE="/etc/ctdb"
+}
-[ -z "$STATD_SHARED_DIRECTORY" ] && {
- echo STATD_SHARED_DIRECTORY not configured. statd-callout failed.
- exit 0
+[ -z "$CTDB_VARDIR" ] && {
+ export CTDB_VARDIR="/var/ctdb"
}
-[ -d $STATD_SHARED_DIRECTORY ] || exit 0
+. $CTDB_BASE/functions
+loadconfig ctdb
+loadconfig nfs
[ -z $NFS_HOSTNAME ] && {
echo NFS_HOSTNAME is not configured. statd-callout failed.
add-client)
# the callout does not tell us to which ip the client connected
# so we must add it to all the ips that we serve
- for f in `/bin/ls /etc/ctdb/state/statd/ip/*`; do
- ip=`/bin/basename $f`
- [ -d $STATD_SHARED_DIRECTORY/$ip ] || /bin/mkdir $STATD_SHARED_DIRECTORY/$ip
- /bin/touch $STATD_SHARED_DIRECTORY/$ip/$2
+ PNN=`ctdb xpnn | sed -e "s/.*://"`
+ ctdb ip -Y | while read LINE; do
+ NODE=`echo $LINE | cut -f3 -d:`
+ [ "$NODE" = "$PNN" ] || {
+ # not us
+ continue
+ }
+ IP=`echo $LINE | cut -f2 -d:`
+ /bin/mkdir -p $CTDB_VARDIR/state/statd/ip/$IP
+ touch $CTDB_VARDIR/state/statd/ip/$IP/$2
done
;;
del-client)
- # the callout does not tell us to which ip the client connected
- # so we must add it to all the ips that we serve
- for f in `/bin/ls /etc/ctdb/state/statd/ip/*`; do
- ip=`/bin/basename $f`
- /bin/rm -f $STATD_SHARED_DIRECTORY/$ip/$2
+ # the callout does not tell us to which ip the client disconnected
+ # so we must remove it from all the ips that we serve
+ PNN=`ctdb xpnn | sed -e "s/.*://"`
+ ctdb ip -Y | while read LINE; do
+ NODE=`echo $LINE | cut -f3 -d:`
+ [ "$NODE" = "$PNN" ] || {
+ # not us
+ continue
+ }
+ IP=`echo $LINE | cut -f2 -d:`
+ mkdir -p $CTDB_VARDIR/state/statd/ip/$IP
+ /bin/rm -f $CTDB_VARDIR/state/statd/ip/$IP/$2
+ done
+ ;;
+ updatelocal)
+ # For all IPs we serve, collect info and push to the config database
+ PNN=`ctdb xpnn | sed -e "s/.*://"`
+ ctdb ip -Y | tail -n +2 | while read LINE; do
+ NODE=`echo $LINE | cut -f3 -d:`
+ [ "$NODE" = "$PNN" ] || {
+ continue
+ }
+ IP=`echo $LINE | cut -f2 -d:`
+
+ mkdir -p $CTDB_VARDIR/state/statd/ip/$IP
+
+ rm -f $CTDB_VARDIR/state/statd/ip/$IP.tar
+ tar cfP $CTDB_VARDIR/state/statd/ip/$IP.tar $CTDB_VARDIR/state/statd/ip/$IP
+
+ rm -f $CTDB_VARDIR/state/statd/ip/$IP.rec
+ ctdb pfetch ctdb.tdb statd-state:$IP $CTDB_VARDIR/state/statd/ip/$IP.rec 2>/dev/null
+ [ "$?" = "0" ] || {
+ # something went wrong, try storing this data
+ echo No record. Store STATD state data for $IP
+ ctdb pstore ctdb.tdb statd-state:$IP $CTDB_VARDIR/state/statd/ip/$IP.tar 2>/dev/null
+ continue
+ }
+
+ cmp $CTDB_VARDIR/state/statd/ip/$IP.tar $CTDB_VARDIR/state/statd/ip/$IP.rec >/dev/null 2>/dev/null
+ [ "$?" = "0" ] || {
+ # something went wrong, try storing this data
+ echo Updated record. Store STATD state data for $IP
+ ctdb pstore ctdb.tdb statd-state:$IP $CTDB_VARDIR/state/statd/ip/$IP.tar 2>/dev/null
+ continue
+ }
done
;;
+
+ updateremote)
+ # For all IPs we dont serve, pull the state from the database
+ PNN=`ctdb xpnn | sed -e "s/.*://"`
+ ctdb ip -Y | tail -n +2 | while read LINE; do
+ NODE=`echo $LINE | cut -f3 -d:`
+ [ "$NODE" = "$PNN" ] && {
+ continue
+ }
+ IP=`echo $LINE | cut -f2 -d:`
+
+ mkdir -p $CTDB_VARDIR/state/statd/ip/$IP
+
+ rm -f $CTDB_VARDIR/state/statd/ip/$IP.rec
+ ctdb pfetch ctdb.tdb statd-state:$IP $CTDB_VARDIR/state/statd/ip/$IP.rec 2>/dev/null
+ [ "$?" = "0" ] || {
+ continue
+ }
+
+ rm -f $CTDB_VARDIR/state/statd/ip/$IP/*
+ tar xfP $CTDB_VARDIR/state/statd/ip/$IP.rec
+ done
+ ;;
+
notify)
# we must restart the lockmanager (on all nodes) so that we get
# a clusterwide grace period (so other clients dont take out
# we need these settings to make sure that no tcp connections survive
# across a very fast failover/failback
- echo 0 > /proc/sys/net/ipv4/tcp_max_tw_buckets
- echo 0 > /proc/sys/net/ipv4/tcp_fin_timeout
- echo 0 > /proc/sys/net/ipv4/tcp_max_orphans
+ #echo 10 > /proc/sys/net/ipv4/tcp_fin_timeout
+ #echo 0 > /proc/sys/net/ipv4/tcp_max_tw_buckets
+ #echo 0 > /proc/sys/net/ipv4/tcp_max_orphans
- # rebuild the state directory for the local statd to use the correct
- # state value and to initally send notifications to all clients
+ # Delete the notification list for statd, we dont want it to
+ # ping any clients
rm -f /var/lib/nfs/statd/sm/*
rm -f /var/lib/nfs/statd/sm.bak/*
- cat $STATD_SHARED_DIRECTORY/state >/var/lib/nfs/statd/state
-
# we must keep a monotonically increasing state variable for the entire
# cluster so state always increases when ip addresses fail from one
# node to another
- [ ! -f $STATD_SHARED_DIRECTORY/state ] && {
- echo 1 | awk '{printf("%c%c%c%c", $0, $0/256, $0/256/256, $0/256/256/256);}' >$STATD_SHARED_DIRECTORY/state
- }
- # read current state
- STATE=`od -t d4 $STATD_SHARED_DIRECTORY/state | head -1 | sed -e "s/^[0-9]*[^0-9]*//"`
- # write current state+2 back to the state file
- # the /2 *2 are to ensure that state is odd. state must be odd.
- STATE=`expr $STATE "/" 2 "*" 2 "+" 3`
- echo $STATE | awk '{printf("%c%c%c%c", $0, $0/256, $0/256/256, $0/256/256/256);}' >$STATD_SHARED_DIRECTORY/state
-
+ # We use epoch and hope the nodes are close enough in clock.
+ # Even numbers mean service is shut down, odd numbers mean
+ # service is started.
+ STATE=`date +"%s"`
+ STATE=`expr "$STATE" "/" "2"`
# we must also let some time pass between stopping and restarting the
# will respond "strangely" immediately after restarting it, which
# causes clients to fail to reclaim the locks.
#
- service nfslock stop > /dev/null 2>&1
+ startstop_nfslock stop > /dev/null 2>&1
sleep 2
- # copy all monitored clients on this node to the local lockmanager
- for f in `/bin/ls /etc/ctdb/state/statd/ip/* 2>/dev/null`; do
- ip=`/bin/basename $f`
- [ -d $STATD_SHARED_DIRECTORY/$ip ] && [ -x /usr/bin/smnotify ] && {
- for g in `/bin/ls $STATD_SHARED_DIRECTORY/$ip/* 2>/dev/null`; do
- client=`/bin/basename $g`
- touch /var/lib/nfs/statd/sm/$client
- done
- }
- done
-
# now start lockmanager again with the new state directory.
- service nfslock start > /dev/null 2>&1
+ startstop_nfslock start > /dev/null 2>&1
# we now need to send out additional statd notifications to ensure
# that clients understand that the lockmanager has restarted.
# Both 2a and 2b are commonly used in lockmanagers since they maximize
# probability that the client will accept the statd notify packet and
# not just ignore it.
- for f in `/bin/ls /etc/ctdb/state/statd/ip/* 2>/dev/null`; do
- ip=`/bin/basename $f`
- [ -d $STATD_SHARED_DIRECTORY/$ip ] && [ -x /usr/bin/smnotify ] && {
- for g in `/bin/ls $STATD_SHARED_DIRECTORY/$ip/* 2>/dev/null`; do
- client=`/bin/basename $g`
-# /bin/rm -f $g
- # send out notifications from the "correct" address
- # (the same addresse as where the lock was taken out
- # on) some clients require that the source address
- # matches where the lock was taken out.
- # also send it both as a name that the client
- # hopefully can resolve into the server ip and
- # and also by specifying the raw ip address as name.
- /usr/bin/smnotify --client=$client --ip=$ip --server=$ip --stateval=$STATE
- /usr/bin/smnotify --client=$client --ip=$ip --server=$NFS_HOSTNAME --stateval=$STATE
+ # For all IPs we serve, collect info and push to the config database
+ PNN=`ctdb xpnn | sed -e "s/.*://"`
+ ctdb ip -Y | tail -n +2 | while read LINE; do
+ NODE=`echo $LINE | cut -f3 -d:`
+ [ "$NODE" = "$PNN" ] || {
+ continue
+ }
+ IP=`echo $LINE | cut -f2 -d:`
+
+ ls $CTDB_VARDIR/state/statd/ip/$IP | while read CLIENT; do
+ rm $CTDB_VARDIR/state/statd/ip/$IP/$CLIENT
+ /usr/bin/smnotify --client=$CLIENT --ip=$IP --server=$ip --stateval=$STATE
+ /usr/bin/smnotify --client=$CLIENT --ip=$IP --server=$NFS_HOSTNAME --stateval=$STATE
+ STATE=`expr "$STATE" "+" "1"`
+ /usr/bin/smnotify --client=$CLIENT --ip=$IP --server=$ip --stateval=$STATE
+ /usr/bin/smnotify --client=$CLIENT --ip=$IP --server=$NFS_HOSTNAME --stateval=$STATE
done
- }
done
;;
esac
git.samba.org / sahlberg / ctdb.git / blobdiff