#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
+. $(dirname $0)/functions.sh
assert_DIR () {
local failed=""
local quota_usrs=$QUOTA_USERS
# get_filesystem_size
- local disksz=$(lfs df | grep "filesystem summary:" | awk '{print $3}')
+ local disksz=$(lfs df $mntpt | grep "filesystem summary:" | awk '{print $3}')
local blk_soft=$((disksz + 1024))
local blk_hard=$((blk_soft + blk_soft / 20)) # Go 5% over
- local Inodes=$(lfs df -i | grep "filesystem summary:" | awk '{print $3}')
+ local Inodes=$(lfs df -i $mntpt | grep "filesystem summary:" | awk '{print $3}')
local i_soft=$Inodes
local i_hard=$((i_soft + i_soft / 20))
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# };
[ "$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 $mnt;
fi
fi"
}
local pids=$(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
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
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=$NETTYPE
Please use correct config or set mds_HOST correctly!"
fi
+
+ sanity_mount_check ||
+ error "environments are insane!"
}
check_timeout () {
lsmod | grep lnet > /dev/null || load_modules
local MSG="$*"
- # Get rif of '
+ # Get rid of '
MSG=${MSG//\'/\\\'}
MSG=${MSG//\(/\\\(}
MSG=${MSG//\)/\\\)}
umask 0022
local BEFORE=`date +%s`
- log "== test $testnum: $message ============ `date +%H:%M:%S` ($BEFORE)"
+ log "== test $testnum: $message == `date +%H:%M:%S` ($BEFORE)"
#check_mds
export TESTNAME=test_$testnum
TEST_FAILED=false
}
calc_osc_kbytes () {
+ df $MOUNT > /dev/null
$LCTL get_param -n osc.*[oO][sS][cC][-_][0-9a-f]*.$1 | calc_sum
}
mpi_run () {
local mpirun="$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
+ 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
}
git://git.whamcloud.com / fs / lustre-release.git / blobdiff