if ! echo $PATH | grep -q $LUSTRE/tests; then
export PATH=$LUSTRE/tests:$PATH
fi
+ if ! echo $PATH | grep -q $LUSTRE/../lustre-iokit/sgpdd-survey; then
+ export PATH=$PATH:$LUSTRE/../lustre-iokit/sgpdd-survey
+ fi
export LST=${LST:-"$LUSTRE/../lnet/utils/lst"}
[ ! -f "$LST" ] && export LST=$(which lst)
- export SGPDDSURVEY=${SGPDDSURVEY:-$(which sgpdd-survey)}
+ export SGPDDSURVEY=${SGPDDSURVEY:-"$LUSTRE/../lustre-iokit/sgpdd-survey/sgpdd-survey")}
+ [ ! -f "$SGPDDSURVEY" ] && export SGPDDSURVEY=$(which sgpdd-survey)
+ # Ubuntu, at least, has a truncate command in /usr/bin
+ # so fully path our truncate command.
+ export TRUNCATE=${TRUNCATE:-$LUSTRE/tests/truncate}
export MDSRATE=${MDSRATE:-"$LUSTRE/tests/mpi/mdsrate"}
[ ! -f "$MDSRATE" ] && export MDSRATE=$(which mdsrate 2> /dev/null)
if ! echo $PATH | grep -q $LUSTRE/tests/racer; then
export RPWD=${RPWD:-$PWD}
export I_MOUNTED=${I_MOUNTED:-"no"}
if [ ! -f /lib/modules/$(uname -r)/kernel/fs/lustre/mds.ko -a \
+ ! -f /lib/modules/$(uname -r)/updates/kernel/fs/lustre/mds.ko -a \
! -f `dirname $0`/../mds/mds.ko ]; then
export CLIENTMODSONLY=yes
fi
+ export SHUTDOWN_ATTEMPTS=${SHUTDOWN_ATTEMPTS:-3}
+
# command line
while getopts "rvwf:" opt $*; do
load_module mgc/mgc
if ! client_only; then
grep -q crc16 /proc/kallsyms || { modprobe crc16 2>/dev/null || true; }
- grep -q jbd /proc/kallsyms || { modprobe jbd 2>/dev/null || true; }
+ grep -q -w jbd /proc/kallsyms || { modprobe jbd 2>/dev/null || true; }
+ grep -q -w jbd2 /proc/kallsyms || { modprobe jbd2 2>/dev/null || true; }
[ "$FSTYPE" = "ldiskfs" ] && load_module ../ldiskfs/ldiskfs/ldiskfs
load_module mgs/mgs
load_module mds/mds
}
# Facet functions
+mount_facets () {
+ local facets=${1:-$(get_facets)}
+ local facet
+
+ for facet in ${facets//,/ }; do
+ mount_facet $facet || error "Restart of $facet failed!"
+ done
+}
+
mount_facet() {
local facet=$1
shift
local dev=$(facet_active $facet)_dev
local opt=${facet}_opt
- echo "Starting ${facet}: ${!opt} $@ ${!dev} ${MOUNT%/*}/${facet}"
- do_facet ${facet} mount -t lustre ${!opt} $@ ${!dev} ${MOUNT%/*}/${facet}
+ local mntpt=$(facet_mntpt $facet)
+
+ echo "Starting ${facet}: ${!opt} $@ ${!dev} $mntpt"
+ do_facet ${facet} "mkdir -p $mntpt; mount -t lustre ${!opt} $@ ${!dev} $mntpt"
RC=${PIPESTATUS[0]}
if [ $RC -ne 0 ]; then
- echo "mount -t lustre $@ ${!dev} ${MOUNT%/*}/${facet}"
+ echo "mount -t lustre $@ ${!dev} $mntpt"
echo "Start of ${!dev} on ${facet} failed ${RC}"
else
do_facet ${facet} "lctl set_param debug=\\\"$PTLDEBUG\\\"; \
eval export ${facet}failover_dev=$device
fi
- do_facet ${facet} mkdir -p ${MOUNT%/*}/${facet}
+ local mntpt=$(facet_mntpt $facet)
+ do_facet ${facet} mkdir -p $mntpt
+ eval export ${facet}_MOUNT=$mntpt
mount_facet ${facet}
RC=$?
return $RC
local running
local facet=$1
shift
- HOST=`facet_active_host $facet`
+ local HOST=`facet_active_host $facet`
[ -z $HOST ] && echo stop: no host for $facet && return 0
- running=$(do_facet ${facet} "grep -c ${MOUNT%/*}/${facet}' ' /proc/mounts") || true
+ local mntpt=$(facet_mntpt $facet)
+ running=$(do_facet ${facet} "grep -c $mntpt' ' /proc/mounts") || true
if [ ${running} -ne 0 ]; then
- echo "Stopping ${MOUNT%/*}/${facet} (opts:$@)"
- do_facet ${facet} umount -d $@ ${MOUNT%/*}/${facet}
+ echo "Stopping $mntpt (opts:$@)"
+ do_facet ${facet} umount -d $@ $mntpt
fi
# umount should block, but we should wait for unrelated obd's
# FIXME: modify get_facets to display all facets wo params
local facets="$(get_facets OST),$(get_facets MDS),mgs"
local node
- local mnt
+ local mntpt
local facet
for facet in ${facets//,/ }; do
node=$(facet_host ${facet})
- mnt=${MOUNT%/*}/${facet}
- sanity_mount_check_nodes $node $mnt ||
+ mntpt=$(facet_mntpt $facet)
+ sanity_mount_check_nodes $node $mntpt ||
{ error "server $node environments are insane!"; return 1; }
done
}
shutdown_node_hard () {
local host=$1
- local attempts=3
+ local attempts=$SHUTDOWN_ATTEMPTS
for i in $(seq $attempts) ; do
shutdown_node $host
sleep 1
- ping -w 3 -c 1 $host > /dev/null 2>&1 || return 0
+ wait_for_function --quiet "! ping -w 3 -c 1 $host" 5 1 && return 0
echo "waiting for $host to fail attempts=$attempts"
[ $i -lt $attempts ] || \
{ echo "$host still pingable after power down! attempts=$attempts" && return 1; }
echo $(comma_list $affected)
}
+facet_up () {
+ local facet=$1
+ local host=${2:-$(facet_host $facet)}
+
+ local label=$(convert_facet2label $facet)
+ do_node $host lctl dl | awk '{print $4}' | grep -q $label
+}
+
+facets_up_on_host () {
+ local host=$1
+ local facets=$(facets_on_host $host)
+ local affected_up
+
+ for facet in ${facets//,/ }; do
+ if $(facet_up $facet $host); then
+ affected_up="$affected_up $facet"
+ fi
+ done
+
+ echo $(comma_list $affected_up)
+}
+
shutdown_facet() {
local facet=$1
fi
}
-# recovery-scale functions
+facets_hosts () {
+ local facets=$1
+ local hosts
+
+ for facet in ${facets//,/ }; do
+ hosts=$(expand_list $hosts $(facet_host $facet) )
+ done
+
+ echo $hosts
+}
+
+_check_progs_installed () {
+ local progs=$@
+ local rc=0
+
+ for prog in $progs; do
+ if ! [ "$(which $prog)" -o "${!prog}" ]; then
+ echo $prog missing on $(hostname)
+ rc=1
+ fi
+ done
+ return $rc
+}
+
check_progs_installed () {
- local clients=$1
+ local nodes=$1
shift
- local progs=$@
- do_nodes $clients "PATH=:$PATH; status=true;
-for prog in $progs; do
- if ! [ \\\"\\\$(which \\\$prog)\\\" -o \\\"\\\${!prog}\\\" ]; then
- echo \\\$prog missing on \\\$(hostname);
- status=false;
- fi
-done;
-eval \\\$status"
+ do_rpc_nodes $nodes _check_progs_installed $@
}
+# recovery-scale functions
client_var_name() {
echo __$(echo $1 | tr '-' 'X')
}
rm $MOUNT/recon
}
+affected_facets () {
+ local facet=$1
+
+ local host=$(facet_active_host $facet)
+ local affected=$facet
+
+ if [ "$FAILURE_MODE" = HARD ]; then
+ affected=$(facets_up_on_host $host)
+ fi
+ echo $affected
+}
+
facet_failover() {
local facet=$1
local sleep_time=$2
echo "Failing $facet on node $host"
- local affected=$facet
-
- if [ "$FAILURE_MODE" = HARD ]; then
- affected=$(facets_on_host $host)
- fi
+ local affected=$(affected_facets $facet)
shutdown_facet $facet
fi
# FIXME; has to be changed to mount all facets concurrently
affected=$(exclude_items_from_list $affected mgs)
- for facet in ${affected//,/ }; do
- mount_facet $facet || error "Restart of $facet on node $host failed!"
- done
+ mount_facets $affected
}
obd_name() {
echo -n $DEVPTR
}
+facet_mntpt () {
+ local facet=$1
+ local var=${facet}_MOUNT
+ eval mntpt=${!var:-${MOUNT%/*}/$facet}
+
+ echo -n $mntpt
+}
+
########
## MountConf setup
else
eval export ${facet}failover_dev=$device
fi
+
+ # get mount point of already mounted device
+ # is facet_dev is already mounted then use the real
+ # mount point of this facet; otherwise use $(facet_mntpt $facet)
+ # i.e. ${facet}_MOUNT if specified by user or default
+ local mntpt=$(do_facet ${facet} cat /proc/mounts | \
+ awk '"'${!dev}'" == $1 && $3 == "lustre" { print $2 }')
+ if [ -z $mntpt ]; then
+ mntpt=$(facet_mntpt $facet)
+ fi
+ eval export ${facet}_MOUNT=$mntpt
}
init_facets_vars () {
#######
# General functions
-check_network() {
- local NETWORK=0
- local WAIT=0
- local MAX=$2
- while [ $NETWORK -eq 0 ]; do
- if ping -c 1 -w 3 $1 > /dev/null; then
- NETWORK=1
- else
- WAIT=$((WAIT + 5))
- echo "waiting for $1, $((MAX - WAIT)) secs left"
- sleep 5
- fi
- if [ $WAIT -gt $MAX ]; then
- echo "Network not available"
- exit 1
- fi
+wait_for_function () {
+ local quiet=""
+
+ # suppress fn both stderr and stdout
+ if [ "$1" = "--quiet" ]; then
+ shift
+ quiet=" > /dev/null 2>&1"
+
+ fi
+
+ local fn=$1
+ local max=${2:-900}
+ local sleep=${3:-5}
+
+ local wait=0
+
+ while true; do
+
+ eval $fn $quiet && return 0
+
+ wait=$((wait + sleep))
+ [ $wait -lt $max ] || return 1
+ echo waiting $fn, $((max - wait)) secs left ...
+ sleep $sleep
done
}
-check_port() {
- while( !($DSH2 $1 "netstat -tna | grep -q $2") ) ; do
- sleep 9
- done
+
+check_network() {
+ local host=$1
+ local max=$2
+ local sleep=${3:-5}
+
+ echo `date +"%H:%M:%S (%s)"` waiting for $host network $max secs ...
+ if ! wait_for_function --quiet "ping -c 1 -w 3 $host" $max $sleep ; then
+ echo "Network not available!"
+ exit 1
+ fi
+
+ echo `date +"%H:%M:%S (%s)"` network interface is UP
}
no_dsh() {
(cd `dirname $1`; echo $PWD/`basename $1`)
}
-sync_clients() {
- [ -d $DIR1 ] && cd $DIR1 && sync; sleep 1; sync
- [ -d $DIR2 ] && cd $DIR2 && sync; sleep 1; sync
- cd $SAVE_PWD
-}
check_grant() {
export base=`basetest $1`
[ ${!testname}x == x ] && return 0
echo -n "checking grant......"
- cd $SAVE_PWD
- # write some data to sync client lost_grant
- rm -f $DIR1/${tfile}_check_grant_* 2>&1
- for i in `seq $OSTCOUNT`; do
- $LFS setstripe $DIR1/${tfile}_check_grant_$i -i $(($i -1)) -c 1
- dd if=/dev/zero of=$DIR1/${tfile}_check_grant_$i bs=4k \
- count=1 > /dev/null 2>&1
- done
- # sync all the data and make sure no pending data on server
- sync_clients
-
- #get client grant and server grant
- client_grant=0
- for d in `lctl get_param -n osc.*.cur_grant_bytes`; do
- client_grant=$((client_grant + $d))
- done
- server_grant=0
- for d in `lctl get_param -n obdfilter.*.tot_granted`; do
- server_grant=$((server_grant + $d))
- done
- # cleanup the check_grant file
- for i in `seq $OSTCOUNT`; do
- rm $DIR1/${tfile}_check_grant_$i
- done
+ local clients=$CLIENTS
+ [ -z $clients ] && clients=$(hostname)
+
+ # sync all the data and make sure no pending data on server
+ do_nodes $clients sync
+
+ # get client grant
+ client_grant=`do_nodes $clients \
+ "$LCTL get_param -n osc.${FSNAME}-*.cur_*grant_bytes" | \
+ awk '{total += $1} END{print total}'`
+
+ # get server grant
+ server_grant=`do_nodes $(comma_list $(osts_nodes)) \
+ "$LCTL get_param -n obdfilter.${FSNAME}-OST*.tot_granted" | \
+ awk '{total += $1} END{print total}'`
+
+ # check whether client grant == server grant
+ if [ $client_grant -ne $server_grant ]; then
+ echo "failed: client:${client_grant} server: ${server_grant}."
+ do_nodes $(comma_list $(osts_nodes)) \
+ "$LCTL get_param obdfilter.${FSNAME}-OST*.tot*"
+ do_nodes $clients "$LCTL get_param osc.${FSNAME}-*.cur_*_bytes"
+ return 1
+ else
+ echo "pass: client:${client_grant} server: ${server_grant}"
+ fi
- #check whether client grant == server grant
- if [ $client_grant != $server_grant ]; then
- echo "failed: client:${client_grant} server: ${server_grant}"
- return 1
- else
- echo "pass"
- fi
}
########################
echo $ost
}
+ostuuid_from_index()
+{
+ $LFS osts $2 | awk '/^'$1'/ { print $2 }'
+}
+
remote_node () {
local node=$1
[ "$node" != "$(hostname)" ]
shift
# Add paths to lustre tests for 32 and 64 bit systems.
- local RPATH="$RLUSTRE/tests:/usr/lib/lustre/tests:/usr/lib64/lustre/tests:$PATH"
- do_nodesv $list "PATH=$RPATH sh rpc.sh $@ "
+ local RPATH="PATH=$RLUSTRE/tests:/usr/lib/lustre/tests:/usr/lib64/lustre/tests:$PATH"
+ do_nodesv $list "${RPATH} NAME=${NAME} sh rpc.sh $@ "
}
wait_clients_import_state () {
local list=$1
local ts=$(date +%s)
-
- # bug 20237, comment 11
- # It would also be useful to provide the option
- # of writing the file to an NFS directory so it doesn't need to be copied.
- local tmp=$TMP
local docp=true
[ -f $LOGDIR/shared ] && docp=false
local res=0
for ((i=0;i<20;i++)); do
- echo -n "checking..."
+ echo -n "checking $dir..."
res=$(do_check_flavor $dir $flavor)
- if [ $res -eq $expect ]; then
- echo "found $res $flavor connections of $dir, OK"
- return 0
- else
- echo "found $res $flavor connections of $dir, not ready ($expect)"
- return 0
- sleep 4
- fi
+ echo "found $res/$expect $flavor connections"
+ [ $res -eq $expect ] && return 0
+ sleep 4
done
echo "Error checking $flavor of $dir: expect $expect, actual $res"
for rule in `do_facet mgs lctl get_param -n $proc 2>/dev/null | grep ".srpc.flavor."`; do
echo "remove rule: $rule"
spec=`echo $rule | awk -F = '{print $1}'`
- do_facet mgs "$LCTL conf_param $spec="
+ do_facet mgs "$LCTL conf_param -d $spec"
done
fi
cat ${rslt}.detail
}
+# returns the canonical name for an ldiskfs device
+ldiskfs_canon() {
+ local dev="$1"
+ local facet="$2"
+
+ do_facet $facet "dv=\\\$(lctl get_param -n $dev);
+if foo=\\\$(lvdisplay -c \\\$dv 2>/dev/null); then
+ echo dm-\\\${foo##*:};
+else
+ echo \\\$(basename \\\$dv);
+fi;"
+}