#export PDSH="pdsh -S -Rssh -w"
-# eg, assert_env LUSTRE MDSNODES OSTNODES CLIENTS
-assert_env() {
- local failed=""
- for name in $@; do
- if [ -z "${!name}" ]; then
- echo "$0: $name must be set"
- failed=1
- fi
- done
- [ $failed ] && exit 1 || true
-}
+# function used by scripts run on remote nodes
+LUSTRE=${LUSTRE:-$(cd $(dirname $0)/..; echo $PWD)}
+. $LUSTRE/tests/functions.sh
assert_DIR () {
local failed=""
[ "$GSS_PIPEFS" = "true" ] && [ ! -f "$LGSSD" ] && \
export LGSSD=$(which lgssd)
export LSVCGSSD=${LSVCGSSD:-"$LUSTRE/utils/gss/lsvcgssd"}
- [ ! -f "$LSVCGSSD" ] && export LSVCGSSD=$(which lsvcgssd)
+ [ ! -f "$LSVCGSSD" ] && export LSVCGSSD=$(which lsvcgssd 2> /dev/null)
export KRB5DIR=${KRB5DIR:-"/usr/kerberos"}
export DIR2
export SAVE_PWD=${SAVE_PWD:-$LUSTRE/tests}
echo Loading modules from $LUSTRE
load_module ../libcfs/libcfs/libcfs
- [ "$PTLDEBUG" ] && lctl set_param debug=$PTLDEBUG
- [ "$SUBSYSTEM" ] && lctl set_param subsystem_debug=${SUBSYSTEM# }
+ [ "$PTLDEBUG" ] && lctl set_param debug="$PTLDEBUG"
+ [ "$SUBSYSTEM" ] && lctl set_param subsystem_debug="${SUBSYSTEM# }"
local MODPROBECONF=
[ -f /etc/modprobe.conf ] && MODPROBECONF=/etc/modprobe.conf
[ ! "$MODPROBECONF" -a -d /etc/modprobe.d ] && MODPROBECONF=/etc/modprobe.d/Lustre
mount_facet() {
local facet=$1
shift
- local dev=${facet}_dev
+ 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}
echo "mount -t lustre $@ ${!dev} ${MOUNT%/*}/${facet}"
echo "Start of ${!dev} on ${facet} failed ${RC}"
else
- do_facet ${facet} "lctl set_param debug=$PTLDEBUG; \
- lctl set_param subsystem_debug=${SUBSYSTEM# }; \
+ do_facet ${facet} "lctl set_param debug=\\\"$PTLDEBUG\\\"; \
+ lctl set_param subsystem_debug=\\\"${SUBSYSTEM# }\\\"; \
lctl set_param debug_mb=${DEBUG_SIZE}; \
sync"
shift
eval export ${facet}_dev=${device}
eval export ${facet}_opt=\"$@\"
+
+ local varname=${facet}failover_dev
+ if [ -n "${!varname}" ] ; then
+ eval export ${facet}failover_dev=${!varname}
+ else
+ eval export ${facet}failover_dev=$device
+ fi
+
do_facet ${facet} mkdir -p ${MOUNT%/*}/${facet}
mount_facet ${facet}
RC=$?
do_node $client mkdir -p $mnt
do_node $client mount -t lustre $OPTIONS $device $mnt || return 1
- do_node $client "lctl set_param debug=$PTLDEBUG;
- lctl set_param subsystem_debug=${SUBSYSTEM# };
+ do_node $client "lctl set_param debug=\\\"$PTLDEBUG\\\";
+ lctl set_param subsystem_debug=\\\"${SUBSYSTEM# }\\\";
lctl set_param debug_mb=${DEBUG_SIZE}"
return 0
fi
}
+# nodes is comma list
+sanity_mount_check_nodes () {
+ local nodes=$1
+ shift
+ local mnts="$@"
+ local mnt
+
+ # FIXME: assume that all cluster nodes run the same os
+ [ "$(uname)" = Linux ] || return 0
+
+ local rc=0
+ for mnt in $mnts ; do
+ do_nodes $nodes "set -x; running=\\\$(grep -c $mnt' ' /proc/mounts);
+mpts=\\\$(mount | grep -w -c $mnt);
+if [ \\\$running -ne \\\$mpts ]; then
+ echo \\\$(hostname) env are INSANE!;
+ exit 1;
+fi"
+ [ $? -eq 0 ] || rc=1
+ done
+ return $rc
+}
+
+sanity_mount_check_servers () {
+ echo Checking servers environments
+
+ # FIXME: modify get_facets to display all facets wo params
+ local facets="$(get_facets OST),$(get_facets MDS)"
+ local node
+ local mnt
+ local facet
+ for facet in ${facets//,/ }; do
+ node=$(facet_host ${facet})
+ mnt=${MOUNT%/*}/${facet}
+ sanity_mount_check_nodes $node $mnt ||
+ { error "server $node environments are insane!"; return 1; }
+ done
+}
+
+sanity_mount_check_clients () {
+ local clients=${1:-$CLIENTS}
+ local mntpt=${2:-$MOUNT}
+ local mntpt2=${3:-$MOUNT2}
+
+ [ -z $clients ] && clients=$(hostname)
+ echo Checking clients $clients environments
+
+ sanity_mount_check_nodes $clients $mntpt $mntpt2 ||
+ error "clients environments are insane!"
+}
+
+sanity_mount_check () {
+ sanity_mount_check_servers || return 1
+ sanity_mount_check_clients || return 2
+}
+
# mount clients if not mouted
zconf_mount_clients() {
local OPTIONS
fi
echo "Starting client $clients: $OPTIONS $device $mnt"
- do_nodes $clients "mount | grep $mnt || { mkdir -p $mnt && mount -t lustre $OPTIONS $device $mnt || false; }"
+
+ do_nodes $clients "set -x;
+running=\\\$(mount | grep -c $mnt' ');
+rc=0;
+if [ \\\$running -eq 0 ] ; then
+ mkdir -p $mnt;
+ mount -t lustre $OPTIONS $device $mnt;
+ rc=$?;
+fi;
+exit $rc"
echo "Started clients $clients: "
- do_nodes $clients "mount | grep $mnt"
+ do_nodes $clients "mount | grep -w $mnt"
- do_nodes $clients "lctl set_param debug=$PTLDEBUG;
- lctl set_param subsystem_debug=${SUBSYSTEM# };
+ do_nodes $clients "lctl set_param debug=\\\"$PTLDEBUG\\\";
+ lctl set_param subsystem_debug=\\\"${SUBSYSTEM# }\\\";
lctl set_param debug_mb=${DEBUG_SIZE};"
return 0
[ "$3" ] && force=-f
echo "Stopping clients: $clients $mnt (opts:$force)"
- do_nodes $clients "set -x; running=\\\$(grep -c $mnt' ' /proc/mounts)
+ do_nodes $clients "set -x; running=\\\$(grep -c $mnt' ' /proc/mounts);
if [ \\\$running -ne 0 ] ; then
-echo Stopping client \\\$(hostname) client $mnt opts:$force
-lsof -t $mnt || need_kill=no
+echo Stopping client \\\$(hostname) client $mnt opts:$force;
+lsof -t $mnt || need_kill=no;
if [ "x$force" != "x" -a "x\\\$need_kill" != "xno" ]; then
pids=\\\$(lsof -t $mnt | sort -u);
if [ -n \\\"\\\$pids\\\" ]; then
- kill -9 \\\$pids
+ kill -9 \\\$pids;
fi
-fi
-busy=\\\$(umount $force $mnt 2>&1 | grep -c "busy")
+fi;
+busy=\\\$(umount $force $mnt 2>&1 | grep -c "busy");
if [ \\\$busy -ne 0 ] ; then
- echo "$mnt is still busy, wait one second" && sleep 1
- umount $force $mnt
+ echo "$mnt is still busy, wait one second" && sleep 1;
+ umount $force $mnt;
fi
fi"
}
eval \\\$status"
}
+client_var_name() {
+ echo __$(echo $1 | tr '-' 'X')
+}
+
start_client_load() {
local client=$1
- local var=${client}_load
+ local load=$2
+ local var=$(client_var_name $client)_load
+ eval export ${var}=$load
do_node $client "PATH=$PATH MOUNT=$MOUNT ERRORS_OK=$ERRORS_OK \
BREAK_ON_ERROR=$BREAK_ON_ERROR \
END_RUN_FILE=$END_RUN_FILE \
LOAD_PID_FILE=$LOAD_PID_FILE \
TESTSUITELOG=$TESTSUITELOG \
- run_${!var}.sh" &
+ run_${load}.sh" &
CLIENT_LOAD_PIDS="$CLIENT_LOAD_PIDS $!"
- log "Started client load: ${!var} on $client"
+ log "Started client load: ${load} on $client"
return 0
}
for ((nodenum=0; nodenum < ${#clients[@]}; nodenum++ )); do
testnum=$((nodenum % numloads))
- eval export ${clients[nodenum]}_load=${CLIENT_LOADS[testnum]}
- start_client_load ${clients[nodenum]}
+ start_client_load ${clients[nodenum]} ${CLIENT_LOADS[testnum]}
done
}
# only for remote client
check_client_load () {
local client=$1
- local var=${client}_load
-
+ local var=$(client_var_name $client)_load
local TESTLOAD=run_${!var}.sh
ps auxww | grep -v grep | grep $client | grep -q "$TESTLOAD" || return 1
local RESULT
local WAIT=0
local sleep=5
- while [ $WAIT -lt $MAX ]; do
+ while [ true ]; do
RESULT=$(do_node $node "$TEST")
if [ "$RESULT" == "$FINAL" ]; then
echo "Updated after $WAIT sec: wanted '$FINAL' got '$RESULT'"
return 0
fi
+ [ $WAIT -ge $MAX ] && break
echo "Waiting $((MAX - WAIT)) secs for update"
WAIT=$((WAIT + sleep))
sleep $sleep
wait_update_facet () {
local facet=$1
- wait_update $(facet_host $facet) $@
+ wait_update $(facet_active_host $facet) "$@"
}
wait_delete_completed () {
local pids=$(ps uax | grep "$PDSH.*$prog.*$MOUNT" | grep -v grep | awk '{print $2}')
[ -z "$pids" ] && return 0
echo "$PDSH processes still exists after $WAIT seconds. Still running: $pids"
+ # FIXME: not portable
for pid in $pids; do
cat /proc/${pid}/status || true
cat /proc/${pid}/wchan || true
do_facet $facet sync
df $MOUNT
local svc=${facet}_svc
- do_facet $facet $LCTL --device %${!svc} readonly
do_facet $facet $LCTL --device %${!svc} notransno
+ do_facet $facet $LCTL --device %${!svc} readonly
do_facet $facet $LCTL mark "$facet REPLAY BARRIER on ${!svc}"
$LCTL mark "local REPLAY BARRIER on ${!svc}"
}
do_facet $facet sync
local svc=${facet}_svc
echo Replay barrier on ${!svc}
- do_facet $facet $LCTL --device %${!svc} readonly
do_facet $facet $LCTL --device %${!svc} notransno
+ do_facet $facet $LCTL --device %${!svc} readonly
do_facet $facet $LCTL mark "$facet REPLAY BARRIER on ${!svc}"
$LCTL mark "local REPLAY BARRIER on ${!svc}"
}
local facet=$1 echo running=${running}
local svc=${facet}_svc
echo Replay barrier on ${!svc}
- do_facet $facet $LCTL --device %${!svc} readonly
do_facet $facet $LCTL --device %${!svc} notransno
+ do_facet $facet $LCTL --device %${!svc} readonly
do_facet $facet $LCTL mark "$facet REPLAY BARRIER on ${!svc}"
$LCTL mark "local REPLAY BARRIER on ${!svc}"
}
change_active() {
local facet=$1
- failover=${facet}failover
+ local failover=${facet}failover
host=`facet_host $failover`
[ -z "$host" ] && return
- curactive=`facet_active $facet`
+ local curactive=`facet_active $facet`
if [ -z "${curactive}" -o "$curactive" == "$failover" ] ; then
eval export ${facet}active=$facet
else
eval export ${facet}active=$failover
fi
# save the active host for this facet
- activevar=${facet}active
+ local activevar=${facet}active
echo "$activevar=${!activevar}" > $TMP/$activevar
}
}
do_facet() {
- facet=$1
+ local facet=$1
shift
- HOST=`facet_active_host $facet`
+ local HOST=`facet_active_host $facet`
[ -z $HOST ] && echo No host defined for facet ${facet} && exit 1
do_node $HOST "$@"
}
fail mds1
fi
- # assume client mount is local
- grep " $MOUNT " /proc/mounts && zconf_umount $HOSTNAME $MOUNT $*
- grep " $MOUNT2 " /proc/mounts && zconf_umount $HOSTNAME $MOUNT2 $*
+ local clients=$CLIENTS
+ [ -z $clients ] && clients=$(hostname)
- if [ -n "$CLIENTS" ]; then
- zconf_umount_clients $CLIENTS $MOUNT "$*" || true
- [ -n "$MOUNT2" ] && zconf_umount_clients $CLIENTS $MOUNT2 "$*" || true
- fi
+ zconf_umount_clients $clients $MOUNT "$*" || true
+ [ -n "$MOUNT2" ] && zconf_umount_clients $clients $MOUNT2 "$*" || true
[ "$CLIENTONLY" ] && return
# The add fn does rm ${facet}active file, this would be enough
}
setupall() {
+ sanity_mount_check ||
+ error "environments are insane!"
+
load_modules
init_gss
if [ -z "$CLIENTONLY" ]; then
if [ -z "${!varname}" ]; then
eval $varname=$(facet_host $facet)
fi
+
+ # ${facet}failover_dev is set in cfg file
+ varname=${facet}failover_dev
+ if [ -n "${!varname}" ] ; then
+ eval export ${facet}failover_dev=${!varname}
+ else
+ eval export ${facet}failover_dev=$device
+ fi
}
init_facets_vars () {
local DEVNAME
- for num in `seq $MDSCOUNT`; do
- DEVNAME=`mdsdevname $num`
- init_facet_vars mds$num $DEVNAME $MDS_MOUNT_OPTS
- done
+ if ! remote_mds_nodsh; then
+ for num in `seq $MDSCOUNT`; do
+ DEVNAME=`mdsdevname $num`
+ init_facet_vars mds$num $DEVNAME $MDS_MOUNT_OPTS
+ done
+ fi
+
+ remote_ost_nodsh && return
for num in `seq $OSTCOUNT`; do
DEVNAME=`ostdevname $num`
}
init_param_vars () {
- export MDSVER=$(do_facet $SINGLEMDS "lctl get_param version" | cut -d. -f1,2)
- export OSTVER=$(do_facet ost1 "lctl get_param version" | cut -d. -f1,2)
- export CLIVER=$(lctl get_param version | cut -d. -f 1,2)
+ if ! remote_ost_nodsh && ! remote_mds_nodsh; then
+ export MDSVER=$(do_facet $SINGLEMDS "lctl get_param version" | cut -d. -f1,2)
+ export OSTVER=$(do_facet ost1 "lctl get_param version" | cut -d. -f1,2)
+ export CLIVER=$(lctl get_param version | cut -d. -f 1,2)
+ fi
+
+ remote_mds_nodsh ||
+ TIMEOUT=$(do_facet $SINGLEMDS "lctl get_param -n timeout")
- TIMEOUT=$(do_facet $SINGLEMDS "lctl get_param -n timeout")
log "Using TIMEOUT=$TIMEOUT"
if [ "$ENABLE_QUOTA" ]; then
setup_quota $MOUNT || return 2
fi
-
}
check_config () {
error "Bad config file: lustre is mounted with mgs $mgshost, but mgs_HOST=$mgs_HOST, NETTYPE=$NETTYPE
Please use correct config or set mds_HOST correctly!"
fi
+
+ sanity_mount_check ||
+ error "environments are insane!"
}
check_timeout () {
ERRLOG=$TMP/lustre_${TESTSUITE}_${TESTNAME}.$(date +%s)
echo "Dumping lctl log to $ERRLOG"
# We need to dump the logs on all nodes
- local NODES=$(nodes_list)
+ local NODES=${NODES:-$(nodes_list)}
for NODE in $NODES; do
do_node $NODE $LCTL dk $ERRLOG
done
}
skip () {
- log " SKIP: ${TESTSUITE} ${TESTNAME} $@"
- [ "$TESTSUITELOG" ] && \
- echo "${TESTSUITE}: SKIP: $TESTNAME $@" >> $TESTSUITELOG || true
+ echo
+ log " SKIP: ${TESTSUITE} ${TESTNAME} $@"
+ [ "$TESTSUITELOG" ] && \
+ echo "${TESTSUITE}: SKIP: $TESTNAME $@" >> $TESTSUITELOG || true
}
build_test_filter() {
lsmod | grep lnet > /dev/null || load_modules
local MSG="$*"
- # Get rif of '
+ # Get rid of '
MSG=${MSG//\'/\\\'}
MSG=${MSG//\(/\\\(}
MSG=${MSG//\)/\\\)}
local myNODES=$(nodes_list)
local NODE
+ echo -n "Resetting fail_loc on all nodes..."
for NODE in $myNODES; do
- do_node $NODE "lctl set_param fail_loc=0 2>/dev/null || true"
+ do_node $NODE "lctl set_param -n fail_loc=0 2>/dev/null || true"
done
+ echo done.
}
run_one() {
umask 0022
local BEFORE=`date +%s`
- log "== test $testnum: $message ============ `date +%H:%M:%S` ($BEFORE)"
+ echo
+ log "== test $testnum: $message == `date +%H:%M:%S` ($BEFORE)"
#check_mds
export TESTNAME=test_$testnum
TEST_FAILED=false
done
}
-check_catastrophe () {
+check_catastrophe() {
local rnodes=${1:-$(comma_list $(remote_nodes_list))}
+ local C=$CATASTROPHE
+ [ -f $C ] && [ $(cat $C) -ne 0 ] && return 1
- [ -f $CATASTROPHE ] && [ $(cat $CATASTROPHE) -ne 0 ] && return 1
if [ $rnodes ]; then
- do_nodes $rnodes "set -x; [ -f $CATASTROPHE ] && { [ \`cat $CATASTROPHE\` -eq 0 ] || false; } || true"
+ do_nodes $rnodes "rc=\\\$([ -f $C ] && echo \\\$(< $C) || echo 0);
+if [ \\\$rc -ne 0 ]; then echo \\\$(hostname): \\\$rc; fi
+exit \\\$rc;"
fi
}
local file=$dir/f0.get_mds_dir_tmpfile
rm -f $file
+ sleep 1
local iused=$(lfs df -i $dir | grep MDT | awk '{print $3}')
local oldused=($iused)
mpi_run () {
local mpirun="$MPIRUN $MPIRUN_OPTIONS"
local command="$mpirun $@"
+ local mpilog=$TMP/mpi.log
+ local rc
if [ "$MPI_USER" != root -a $mpirun ]; then
echo "+ chmod 0777 $MOUNT"
ls -ald $MOUNT
echo "+ $command"
- eval $command
+ eval $command 2>&1 > $mpilog || true
+
+ rc=${PIPESTATUS[0]}
+ if [ $rc -eq 0 ] && grep -q "p4_error: : [^0]" $mpilog ; then
+ rc=1
+ fi
+ cat $mpilog
+ return $rc
}
mdsrate_cleanup () {
mpi_run -np $1 -machinefile $2 ${MDSRATE} --unlink --nfiles $3 --dir $4 --filefmt $5 $6
}
+delayed_recovery_enabled () {
+ local var=${SINGLEMDS}_svc
+ do_facet $SINGLEMDS lctl get_param -n mdd.${!var}.stale_export_age > /dev/null 2>&1
+}
+
+########################
+convert_facet2name() {
+ case "$1" in
+ "ost" ) echo "OST0000" ;;
+ "ost1") echo "OST0000" ;;
+ "ost2") echo "OST0001" ;;
+ "ost3") echo "OST0002" ;;
+ "ost4") echo "OST0003" ;;
+ "ost5") echo "OST0004" ;;
+ *) error "unknown facet!" ;;
+ esac
+}
+
+get_clientosc_proc_path() {
+ local ost=$1
+
+ echo "{$1}-osc-*"
+}
+
+get_lustre_version () {
+ local node=${1:-"mds"}
+ do_facet $node $LCTL get_param -n version | awk '/^lustre:/ {print $2}'
+}
+
+get_mds_version_major () {
+ local version=$(get_lustre_version mds)
+ echo $version | awk -F. '{print $1}'
+}
+
+get_mds_version_minor () {
+ local version=$(get_lustre_version mds)
+ echo $version | awk -F. '{print $2}'
+}
+
+get_mdtosc_proc_path() {
+ local ost=$1
+ local major=$(get_mds_version_major)
+ local minor=$(get_mds_version_minor)
+ if [ $major -le 1 -a $minor -le 8 ] ; then
+ echo "${ost}-osc"
+ else
+ echo "${ost}-osc-MDT0000"
+ fi
+}
+
+get_osc_import_name() {
+ local node=$1
+ local ost=$2
+ local name=$(convert_facet2name $ost)
+
+ if [ "$node" == "mds" ]; then
+ get_mdtosc_proc_path $name
+ return 0
+ fi
+
+ get_clientosc_proc_path $name
+ return 0
+}
+
+wait_osc_import_state() {
+ local node=$1
+ local ost_facet=$2
+ local expected=$3
+ local ost=$(get_osc_import_name $node $ost_facet)
+ local CONN_PROC
+ local CONN_STATE
+ local i=0
+
+ CONN_PROC="osc.${FSNAME}-${ost}.ost_server_uuid"
+ CONN_STATE=$(do_facet $node lctl get_param -n $CONN_PROC | cut -f2)
+ while [ "${CONN_STATE}" != "${expected}" ]; do
+ # for disconn we can check after proc entry is removed
+ [ "x${CONN_STATE}" == "x" -a "${expected}" == "DISCONN" ] && return 0
+ # disconnect rpc should be wait not more obd_timeout
+ [ $i -ge $(($TIMEOUT * 3 / 2)) ] && \
+ error "can't put import for ${ost}(${ost_facet}) into ${expected} state" && return 1
+ sleep 1
+ CONN_STATE=$(do_facet $node lctl get_param -n $CONN_PROC | cut -f2)
+ i=$(($i + 1))
+ done
+
+ log "${ost_facet} now in ${CONN_STATE} state"
+ return 0
+}