1
0
mirror of https://github.com/samba-team/samba.git synced 2025-01-10 01:18:15 +03:00

ctdb-scripts: Rewrite statd-callout to avoid 10 minute lag

This is naive and assumes no performance problems when updating
persistent DBs.  It also does no error handling.

Signed-off-by: Martin Schwenke <martin@meltin.net>
Pair-programmed-with: Amitay Isaacs <amitay@gmail.com>
Reviewed-by: Michael Adam <obnox@samba.org>
This commit is contained in:
Martin Schwenke 2013-11-08 16:41:11 +11:00 committed by Michael Adam
parent 8dc416c069
commit 1dcf01f4a6
3 changed files with 95 additions and 112 deletions

View File

@ -95,10 +95,6 @@ case "$1" in
nfs_check_rpc_services
nfs_check_thread_count
# Every 10 minutes, update the statd state database for which
# clients need notifications
nfs_statd_update 600
;;
*)

View File

@ -12,103 +12,69 @@
export CTDB_BASE=$(cd -P $(dirname "$0") ; echo "$PWD")
. $CTDB_BASE/functions
# Overwrite this so we get some logging
die ()
{
script_log "statd-callout" "$@"
exit 1
}
loadconfig ctdb
loadconfig nfs
[ -z $NFS_HOSTNAME ] && {
echo NFS_HOSTNAME is not configured. statd-callout failed.
exit 0
}
[ -n "$NFS_HOSTNAME" ] || \
die "NFS_HOSTNAME is not configured. statd-callout failed"
# A handy newline
nl="
"
case "$1" in
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
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
touch $CTDB_VARDIR/state/statd/ip/$IP/$2
done
add-client)
# statd does not tell us to which IP the client connected so
# we must add it to all the IPs that we serve
cip="$2"
pnn=$(ctdb xpnn | sed -e 's/.*://')
date=$(date '+%s')
ctdb ip -Y |
tail -n +2 | {
# This all needs to be in the end of the pipe so it
# doesn't get lost
items=""
while IFS=":" read x sip node x ; do
[ "$node" = "$pnn" ] || continue # not us
key="statd-state@${sip}@${cip}"
item="\"${key}\" \"${date}\""
items="${items}${items:+${nl}}${item}"
done
if ! echo "$items" | ctdb ptrans "ctdb.tdb" ; then
die "Failed to add clients"
fi
}
;;
del-client)
# 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
rm -f $CTDB_VARDIR/state/statd/ip/$IP/$2
done
# statd does not tell us from which IP the client disconnected
# so we must add it to all the IPs that we serve
cip="$2"
pnn=$(ctdb xpnn | sed -e 's/.*://')
ctdb ip -Y |
tail -n +2 | {
# This all needs to be in the end of the pipe so it
# doesn't get lost
items=""
while IFS=":" read x sip node x ; do
[ "$node" = "$pnn" ] || continue # not us
key="statd-state@${sip}@${cip}"
item="\"${key}\" \"\""
items="${items}${items:+${nl}}${item}"
done
if ! echo "$items" | ctdb ptrans "ctdb.tdb" ; then
die "Failed to delete clients"
fi
}
;;
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)
notify)
# we must restart the lockmanager (on all nodes) so that we get
# a clusterwide grace period (so other clients dont take out
# conflicting locks through other nodes before all locks have been
@ -131,8 +97,7 @@ case "$1" in
# We use epoch and hope the nodes are close enough in clock.
# Even numbers mean service is shut down, odd numbers mean
# service is started.
STATE=$(( $(date '+%s') / 2 * 2))
state_even=$(( $(date '+%s') / 2 * 2))
# we must also let some time pass between stopping and restarting the
# lockmanager since othervise there is a window where the lockmanager
@ -174,22 +139,44 @@ case "$1" in
# probability that the client will accept the statd notify packet and
# not just ignore it.
# 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 $LINE | cut -f2 -d:`
pnn=$(ctdb xpnn | sed -e 's/.*://')
ls $CTDB_VARDIR/state/statd/ip/$IP | while read CLIENT; do
rm $CTDB_VARDIR/state/statd/ip/$IP/$CLIENT
smnotify --client=$CLIENT --ip=$IP --server=$ip --stateval=$STATE
smnotify --client=$CLIENT --ip=$IP --server=$NFS_HOSTNAME --stateval=$STATE
STATE=$(($STATE + 1))
smnotify --client=$CLIENT --ip=$IP --server=$ip --stateval=$STATE
smnotify --client=$CLIENT --ip=$IP --server=$NFS_HOSTNAME --stateval=$STATE
done
# Construct a sed expression to take catdb output and produce pairs of:
# server-IP client-IP
# but only for the server-IPs that are hosted on this node.
sed_expr=$(ctdb ip | tail -n +2 |
awk -v pnn=$pnn 'pnn == $2 { printf "s/^key.*=.*statd-state@\\(%s\\)@\\([^\"]*\\).*/\\1 \\2/p\n", gensub(/\./, "\\\\.", "g", $1) }')
statd_state=$(ctdb catdb ctdb.tdb | sed -n "$sed_expr" | sort)
# The following is dangerous if this script times out before
# all of the smnotify commands are run. Revert to individual
# pdelete commands for now and consider optimising smnotify to
# read all the data from stdin and then run it in the
# background.
#
# Delete all the items from the TDB
#if ! echo "$statd_state" | \
# awk '{ printf "\"statd-state@%s@%s\" \"\"\n", $1, $2 }') | \
# ctdb ptrans ctdb.tdb ; then
# die "Yikes!"
#fi
prev=""
echo "$statd_state" |
while read sip cip ; do
# Delete the entry from the DB
ctdb pdelete ctdb.tdb "statd-state@${sip}@${cip}"
# Reset stateval for each serverip
[ "$sip" = "$prev" ] || stateval="$state_even"
# Send notifies for server shutdown
smnotify --client=$cip --ip=$sip --server=$sip --stateval=$stateval
smnotify --client=$cip --ip=$sip --server=$NFS_HOSTNAME --stateval=$stateval
# Send notifies for server startup
stateval=$(($stateval + 1))
smnotify --client=$cip --ip=$sip --server=$sip --stateval=$stateval
smnotify --client=$cip --ip=$sip --server=$NFS_HOSTNAME --stateval=$stateval
done
;;
esac

View File

@ -4337,13 +4337,13 @@ static int control_ptrans(struct ctdb_context *ctdb,
FILE *file;
int ret;
if (argc != 2) {
if (argc < 1) {
talloc_free(tmp_ctx);
usage();
}
file = stdin;
if (strcmp(argv[1], "-") != 0) {
if (argc == 2) {
file = fopen(argv[1], "r");
if (file == NULL) {
DEBUG(DEBUG_ERR,("Unable to open file for reading '%s'\n", argv[1]));