#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=""
if ! echo $PATH | grep -q $LUSTRE/tests; then
export PATH=$PATH:$LUSTRE/tests
fi
- export MDSRATE=${MDSRATE:-"$LUSTRE/tests/mdsrate"}
+ export MDSRATE=${MDSRATE:-"$LUSTRE/tests/mpi/mdsrate"}
[ ! -f "$MDSRATE" ] && export MDSRATE=$(which mdsrate 2> /dev/null)
if ! echo $PATH | grep -q $LUSTRE/tests/racer; then
export PATH=$PATH:$LUSTRE/tests/racer
fi
+ if ! echo $PATH | grep -q $LUSTRE/tests/mpi; then
+ export PATH=$PATH:$LUSTRE/tests/mpi
+ fi
export LCTL=${LCTL:-"$LUSTRE/utils/lctl"}
[ ! -f "$LCTL" ] && export LCTL=$(which lctl)
export LFS=${LFS:-"$LUSTRE/utils/lfs"}
[ "$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 ${facet}
RC=$?
wait_exit_ST ${facet}
}
+# save quota version (both administrative and operational quotas)
+# add an additional parameter if mountpoint is ever different from $MOUNT
+quota_save_version() {
+ local fsname=${2:-$FSNAME}
+ local spec=$1
+ local ver=$(tr -c -d "123" <<< $spec)
+ local type=$(tr -c -d "ug" <<< $spec)
+
+ [ -n "$ver" -a "$ver" != "3" ] && error "wrong quota version specifier"
+
+ $LFS quotaoff -ug $MOUNT # just in case
+ [ -n "$type" ] && { $LFS quotacheck -$type $MOUNT || error "quotacheck has failed"; }
+
+ do_facet mgs "lctl conf_param ${fsname}-MDT*.mdd.quota_type=$spec"
+ local varsvc
+ local osts=$(get_facets OST)
+ for ost in ${osts//,/ }; do
+ varsvc=${ost}_svc
+ do_facet mgs "lctl conf_param ${!varsvc}.ost.quota_type=$spec"
+ done
+}
+
+# client could mount several lustre
+quota_type () {
+ local fsname=${1:-$FSNAME}
+ local rc=0
+ do_facet mgs lctl get_param mdd.${fsname}-MDT*.quota_type || rc=$?
+ do_nodes $(comma_list $(osts_nodes)) \
+ lctl get_param obdfilter.${fsname}-OST*.quota_type || rc=$?
+ return $rc
+}
+
+restore_quota_type () {
+ local mntpt=${1:-$MOUNT}
+ local quota_type=$(quota_type $FSNAME | grep MDT | cut -d "=" -f2)
+ if [ ! "$old_QUOTA_TYPE" ] || [ "$quota_type" = "$old_QUOTA_TYPE" ]; then
+ return
+ fi
+ quota_save_version $old_QUOTA_TYPE
+}
+
+setup_quota(){
+ local mntpt=$1
+
+ # We need:
+ # 1. run quotacheck only if quota is off
+ # 2. save the original quota_type params, restore them after testing
+
+ # Suppose that quota type the same on mds and ost
+ local quota_type=$(quota_type | grep MDT | cut -d "=" -f2)
+ [ ${PIPESTATUS[0]} -eq 0 ] || error "quota_type failed!"
+ echo "[HOST:$HOSTNAME] [old_quota_type:$quota_type] [new_quota_type:$QUOTA_TYPE]"
+ if [ "$quota_type" != "$QUOTA_TYPE" ]; then
+ export old_QUOTA_TYPE=$quota_type
+ quota_save_version $QUOTA_TYPE
+ fi
+
+ local quota_usrs=$QUOTA_USERS
+
+ # get_filesystem_size
+ 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 $mntpt | grep "filesystem summary:" | awk '{print $3}')
+ local i_soft=$Inodes
+ local i_hard=$((i_soft + i_soft / 20))
+
+ echo "Total disk size: $disksz block-softlimit: $blk_soft block-hardlimit:
+ $blk_hard inode-softlimit: $i_soft inode-hardlimit: $i_hard"
+
+ local cmd
+ for usr in $quota_usrs; do
+ echo "Setting up quota on $HOSTNAME:$mntpt for $usr..."
+ for type in u g; do
+ cmd="$LFS setquota -$type $usr -b $blk_soft -B $blk_hard -i $i_soft -I $i_hard $mntpt"
+ echo "+ $cmd"
+ eval $cmd || error "$cmd FAILED!"
+ done
+ # display the quota status
+ echo "Quota settings for $usr : "
+ $LFS quota -v -u $usr $mntpt || true
+ done
+}
+
zconf_mount() {
local OPTIONS
local client=$1
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 "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 $rc
+}
+
+sanity_mount_check_servers () {
+ [ "$CLIENTONLY" ] &&
+ { echo "CLIENTONLY mode, skip mount_check_servers"; return 0; } || true
+ 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 "
+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 "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
+ if [ -n \\\"\\\$pids\\\" ]; then
+ 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"
}
+shudown_node_hard () {
+ local host=$1
+ local attempts=3
+
+ for i in $(seq $attempts) ; do
+ $POWER_DOWN $host
+ sleep 1
+ ping -w 3 -c 1 $host > /dev/null 2>&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; }
+ done
+}
+
+shutdown_client() {
+ local client=$1
+ local mnt=${2:-$MOUNT}
+ local attempts=3
+
+ if [ "$FAILURE_MODE" = HARD ]; then
+ shudown_node_hard $client
+ else
+ zconf_umount_clients $client $mnt -f
+ fi
+}
+
shutdown_facet() {
local facet=$1
if [ "$FAILURE_MODE" = HARD ]; then
- $POWER_DOWN `facet_active_host $facet`
- sleep 2
+ shudown_node_hard $(facet_active_host $facet)
elif [ "$FAILURE_MODE" = SOFT ]; then
stop $facet
fi
shift
local progs=$@
- do_nodes $clients "set -x ; PATH=:$PATH status=true; for prog in $progs; do
- which \\\$prog || { echo \\\$prog missing on \\\$(hostname) && status=false; }
- done;
- eval \\\$status"
+ do_nodes $clients "PATH=:$PATH; status=true;
+for prog in $progs; do
+ if ! [ \\\"\\\$(which \\\$prog)\\\" -o \\\"\\\${!prog}\\\" ]; then
+ echo \\\$prog missing on \\\$(hostname);
+ status=false;
+ fi
+done;
+eval \\\$status"
}
-start_client_load() {
- local list=(${1//,/ })
- local nodenum=$2
+client_var_name() {
+ echo __$(echo $1 | tr '-' 'X')
+}
- local numloads=${#CLIENT_LOADS[@]}
- local testnum=$((nodenum % numloads))
+start_client_load() {
+ local client=$1
+ local load=$2
+ local var=$(client_var_name $client)_load
+ eval export ${var}=$load
- do_node ${list[nodenum]} "PATH=$PATH MOUNT=$MOUNT ERRORS_OK=$ERRORS_OK \
+ 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_${CLIENT_LOADS[testnum]}.sh" &
+ run_${load}.sh" &
CLIENT_LOAD_PIDS="$CLIENT_LOAD_PIDS $!"
- log "Started client load: ${CLIENT_LOADS[testnum]} on ${list[nodenum]}"
+ log "Started client load: ${load} on $client"
- eval export ${list[nodenum]}_load=${CLIENT_LOADS[testnum]}
return 0
}
start_client_loads () {
- local clients=(${1//,/ })
+ local -a clients=(${1//,/ })
+ local numloads=${#CLIENT_LOADS[@]}
+ local testnum
- for ((num=0; num < ${#clients[@]}; num++ )); do
- start_client_load $1 $num
+ for ((nodenum=0; nodenum < ${#clients[@]}; nodenum++ )); do
+ testnum=$((nodenum % numloads))
+ 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
-
- check_catastrophe $client || return 2
-
- # see if the load is still on the client
+
+ # bug 18914: try to connect several times not only when
+ # check ps, but while check_catastrophe also
local tries=3
local RC=254
while [ $RC = 254 -a $tries -gt 0 ]; do
let tries=$tries-1
# assume success
RC=0
+ if ! check_catastrophe $client; then
+ RC=${PIPESTATUS[0]}
+ if [ $RC -eq 254 ]; then
+ # FIXME: not sure how long we shuold sleep here
+ sleep 10
+ continue
+ fi
+ echo "check catastrophe failed: RC=$RC "
+ return $RC
+ fi
+ done
+ # We can continue try to connect if RC=254
+ # Just print the warning about this
+ if [ $RC = 254 ]; then
+ echo "got a return status of $RC from do_node while checking catastrophe on $client"
+ fi
+
+ # see if the load is still on the client
+ tries=3
+ RC=254
+ while [ $RC = 254 -a $tries -gt 0 ]; do
+ let tries=$tries-1
+ # assume success
+ RC=0
if ! do_node $client "ps auxwww | grep -v grep | grep -q $TESTLOAD"; then
RC=${PIPESTATUS[0]}
sleep 30
fi
done
if [ $RC = 254 ]; then
- echo "got a return status of $RC from do_node while checking (i.e. with 'ps') the client load on the remote system"
+ echo "got a return status of $RC from do_node while checking (catastrophe and 'ps') the client load on $client"
# see if we can diagnose a bit why this is
fi
for client in $clients; do
check_client_load $client
- rc=$?
+ rc=${PIPESTATUS[0]}
if [ "$rc" != 0 ]; then
log "Client load failed on node $client, rc=$rc"
return $rc
fi
done
}
+
+restart_client_loads () {
+ local clients=${1//,/ }
+ local expectedfail=${2:-""}
+ local client=
+ local rc=0
+
+ for client in $clients; do
+ check_client_load $client
+ rc=${PIPESTATUS[0]}
+ if [ "$rc" != 0 -a "$expectedfail" ]; then
+ start_client_load $client
+ echo "Restarted client load: on $client. Checking ..."
+ check_client_load $client
+ rc=${PIPESTATUS[0]}
+ if [ "$rc" != 0 ]; then
+ log "Client load failed to restart on node $client, rc=$rc"
+ # failure one client load means test fail
+ # we do not need to check other
+ return $rc
+ fi
+ else
+ return $rc
+ fi
+ done
+}
# End recovery-scale functions
# verify that lustre actually cleaned up properly
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 () {
}
wait_for_host() {
- local HOST=$1
- check_network "$HOST" 900
- while ! do_node $HOST "ls -d $LUSTRE " > /dev/null; do sleep 5; done
+ local host=$1
+ check_network "$host" 900
+ while ! do_node $host "ls -d $LUSTRE " > /dev/null; do sleep 5; done
}
wait_for() {
local facet=$1
- local HOST=`facet_active_host $facet`
- wait_for_host $HOST
+ local host=`facet_active_host $facet`
+ wait_for_host $host
}
-wait_mds_recovery_done () {
- local timeout=`do_facet $SINGLEMDS lctl get_param -n timeout`
-#define OBD_RECOVERY_TIMEOUT (obd_timeout * 5 / 2)
-# as we are in process of changing obd_timeout in different ways
-# let's set MAX longer than that
- local MAX=$(( timeout * 4 ))
+wait_recovery_complete () {
+ local facet=$1
+
+ # Use default policy if $2 is not passed by caller.
+ #define OBD_RECOVERY_TIMEOUT (obd_timeout * 5 / 2)
+ # as we are in process of changing obd_timeout in different ways
+ # let's set MAX longer than that
+ local MAX=${2:-$(( TIMEOUT * 4 ))}
+
+ local var_svc=${facet}_svc
+ local procfile="*.${!var_svc}.recovery_status"
local WAIT=0
+ local STATUS=
+
while [ $WAIT -lt $MAX ]; do
- STATUS=`do_facet $SINGLEMDS "lctl get_param -n mdt.*-MDT0000.recovery_status | grep status"`
- echo $STATUS | grep COMPLETE && return 0
+ STATUS=$(do_facet $facet lctl get_param -n $procfile | grep status)
+ [[ $STATUS = "status: COMPLETE" ]] && return 0
sleep 5
WAIT=$((WAIT + 5))
- echo "Waiting $(($MAX - $WAIT)) secs for MDS recovery done"
+ echo "Waiting $((MAX - WAIT)) secs for $facet recovery done. $STATUS"
done
- echo "MDS recovery not done in $MAX sec"
+ echo "$facet recovery not done in $MAX sec. $STATUS"
return 1
}
local pids=$(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
shutdown_facet $facet
[ -n "$sleep_time" ] && sleep $sleep_time
reboot_facet $facet
- client_df &
- DFPID=$!
- echo "df pid is $DFPID"
change_active $facet
- TO=`facet_active_host $facet`
+ local TO=`facet_active_host $facet`
echo "Failover $facet to $TO"
wait_for $facet
mount_facet $facet || error "Restart of $facet failed"
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}"
}
fail() {
facet_failover $* || error "failover: $?"
- df $MOUNT || error "post-failover df: $?"
+ client_df || error "post-failover df: $?"
}
fail_nodf() {
stop $facet
change_active $facet
mount_facet $facet -o abort_recovery
- df $MOUNT || echo "first df failed: $?"
+ client_df || echo "first df failed: $?"
sleep 1
- df $MOUNT || error "post-failover df: $?"
+ client_df || error "post-failover df: $?"
}
do_lmc() {
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_node() {
- HOST=$1
+ local verbose=false
+ # do not stripe off hostname if verbose, bug 19215
+ if [ x$1 = x--verbose ]; then
+ shift
+ verbose=true
+ fi
+
+ local HOST=$1
shift
local myPDSH=$PDSH
if [ "$HOST" = "$HOSTNAME" ]; then
[ -n "$($myPDSH $HOST cat $command_status)" ] && return 1 || true
return 0
fi
- $myPDSH $HOST "(PATH=\$PATH:$RLUSTRE/utils:$RLUSTRE/tests:/sbin:/usr/sbin; cd $RPWD; sh -c \"$@\")" | sed "s/^${HOST}: //"
+
+ if $verbose ; then
+ $myPDSH $HOST "(PATH=\$PATH:$RLUSTRE/utils:$RLUSTRE/tests:/sbin:/usr/sbin; cd $RPWD; sh -c \"$@\")"
+ else
+ $myPDSH $HOST "(PATH=\$PATH:$RLUSTRE/utils:$RLUSTRE/tests:/sbin:/usr/sbin; cd $RPWD; sh -c \"$@\")" | sed "s/^${HOST}: //"
+ fi
return ${PIPESTATUS[0]}
}
}
do_nodes() {
+ local verbose=false
+ # do not stripe off hostname if verbose, bug 19215
+ if [ x$1 = x--verbose ]; then
+ shift
+ verbose=true
+ fi
+
local rnodes=$1
shift
if $(single_local_node $rnodes); then
+ $verbose && echo -n $rnodes:' ' || true
do_node $rnodes $@
return $?
fi
$myPDSH $rnodes $LCTL mark "$@" > /dev/null 2>&1 || :
fi
- $myPDSH $rnodes "(PATH=\$PATH:$RLUSTRE/utils:$RLUSTRE/tests:/sbin:/usr/sbin; cd $RPWD; sh -c \"$@\")" | sed -re "s/\w+:\s//g"
+ if $verbose ; then
+ $myPDSH $rnodes "(PATH=\$PATH:$RLUSTRE/utils:$RLUSTRE/tests:/sbin:/usr/sbin; cd $RPWD; sh -c \"$@\")"
+ else
+ $myPDSH $rnodes "(PATH=\$PATH:$RLUSTRE/utils:$RLUSTRE/tests:/sbin:/usr/sbin; cd $RPWD; sh -c \"$@\")" | sed -re "s/\w+:\s//g"
+ fi
return ${PIPESTATUS[0]}
}
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
}
cleanupall() {
+ nfs_client_mode && return
+
stopall $*
unload_modules
cleanup_gss
}
setupall() {
+ nfs_client_mode && return
+
+ sanity_mount_check ||
+ error "environments are insane!"
+
load_modules
init_gss
if [ -z "$CLIENTONLY" ]; then
}
init_facet_vars () {
+ [ "$CLIENTONLY" ] && return 0
local facet=$1
shift
local device=$1
if [ -z "${!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`
done
}
+mds_sanity_check () {
+ local timeout=$1
+ local period=0
+
+ while [ $period -lt $timeout ]; do
+ count=$(do_facet $SINGLEMDS "lctl dl | grep 'osc.*mdtlov_UUID' | grep ' IN ' 2>/dev/null | wc -l")
+ if [ $count -eq 0 ]; then
+ break
+ fi
+
+ echo "There are $count OST are inactive, wait $period seconds, and try again"
+ sleep 3
+ period=$((period+3))
+ done
+
+ [ $period -lt $timeout ] || log "$count OST are inactive after $timeout seconds, give up"
+}
+
+som_check() {
+ SOM_ENABLED=$(do_facet $SINGLEMDS "$LCTL get_param mdt.*.som" | awk -F= ' {print $2}' | head -n 1)
+ echo $SOM_ENABLED
+}
+
init_param_vars () {
- export MDSVER=$(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"
+
+ mds_sanity_check $TIMEOUT
+
+ if [ x"$(som_check)" = x"enabled" ]; then
+ ENABLE_QUOTA=""
+ fi
+ if [ "$ENABLE_QUOTA" ]; then
+ setup_quota $MOUNT || return 2
+ fi
+}
+
+nfs_client_mode () {
+ if [ "$NFSCLIENT" ]; then
+ echo "NFSCLIENT mode: setup, cleanup, check config skipped"
+ local clients=$CLIENTS
+ [ -z $clients ] && clients=$(hostname)
+
+ # FIXME: remove hostname when 19215 fixed
+ do_nodes $clients "echo \\\$(hostname); grep ' '$MOUNT' ' /proc/mounts"
+ declare -a nfsexport=(`grep ' '$MOUNT' ' /proc/mounts | awk '{print $1}' | awk -F: '{print $1 " " $2}'`)
+ do_nodes ${nfsexport[0]} "echo \\\$(hostname); df -T ${nfsexport[1]}"
+ return
+ fi
+ return 1
}
check_config () {
+ nfs_client_mode && return
+
local mntpt=$1
+
+ local mounted=$(mount | grep " $mntpt ")
+ if [ "$CLIENTONLY" ]; then
+ # bug 18021
+ # CLIENTONLY should not depend on *_HOST settings
+ local mgc=$($LCTL device_list | awk '/MGC/ {print $4}')
+ # in theory someone could create a new,
+ # client-only config file that assumed lustre was already
+ # configured and didn't set the MGSNID. If MGSNID is not set,
+ # then we should use the mgs nid currently being used
+ # as the default value. bug 18021
+ [[ x$MGSNID = x ]] &&
+ MGSNID=${mgc//MGC/}
+
+ if [[ x$mgc != xMGC$MGSNID ]]; then
+ if [ "$mgs_HOST" ]; then
+ local mgc_ip=$(ping -q -c1 -w1 $mgs_HOST | grep PING | awk '{print $3}' | sed -e "s/(//g" -e "s/)//g")
+ [[ x$mgc = xMGC$mgc_ip@$NETTYPE ]] ||
+ error_exit "MGSNID=$MGSNID, mounted: $mounted, MGC : $mgc"
+ fi
+ fi
+ return 0
+ fi
+
local myMGS_host=$mgs_HOST
if [ "$NETTYPE" = "ptl" ]; then
myMGS_host=$(h2ptl $mgs_HOST | sed -e s/@ptl//)
mgshost=$(echo $mgshost | awk -F: '{print $1}')
if [ "$mgshost" != "$myMGS_host" ]; then
- FAIL_ON_ERROR=true \
- error "Bad config file: lustre is mounted with mgs $mgshost, but mgs_HOST=$mgs_HOST, NETTYPE=$NETTYPE
+ error_exit "Bad config file: lustre is mounted with mgs $mgshost, but mgs_HOST=$mgs_HOST, NETTYPE=$NETTYPE
Please use correct config or set mds_HOST correctly!"
fi
+
+ sanity_mount_check ||
+ error "environments are insane!"
}
check_timeout () {
}
check_and_setup_lustre() {
+ nfs_client_mode && return
+
local MOUNTED=$(mounted_lustre_filesystems)
if [ -z "$MOUNTED" ] || ! $(echo $MOUNTED | grep -w -q $MOUNT); then
[ "$REFORMAT" ] && formatall
check_config $MOUNT
init_facets_vars
init_param_vars
+
+ do_nodes $(comma_list $(nodes_list)) "lctl set_param debug=\\\"$PTLDEBUG\\\";
+ lctl set_param subsystem_debug=\\\"${SUBSYSTEM# }\\\";
+ lctl set_param debug_mb=${DEBUG_SIZE};
+ sync"
fi
if [ "$ONLY" == "setup" ]; then
exit 0
check_and_cleanup_lustre() {
if [ "`mount | grep $MOUNT`" ]; then
[ -n "$DIR" ] && rm -rf $DIR/[Rdfs][0-9]*
+ [ "$ENABLE_QUOTA" ] && restore_quota_type || true
fi
if [ "$I_MOUNTED" = "yes" ]; then
cleanupall -f || error "cleanup failed"
echo "$*" | tr -s " " "\n" | sort -b -u | tr "\n" " " | sed 's/ \([^$]\)/,\1/g'
}
-# list is comma separated list
-exclude_item_from_list () {
+# list, excluded are the comma separated lists
+exclude_items_from_list () {
local list=$1
local excluded=$2
+ local item
list=${list//,/ }
- list=$(echo " $list " | sed -re "s/\s+$excluded\s+/ /g")
+ for item in ${excluded//,/ }; do
+ list=$(echo " $list " | sed -re "s/\s+$item\s+/ /g")
+ done
echo $(comma_list $list)
}
+# list, expand are the comma separated lists
+expand_list () {
+ local list=${1//,/ }
+ local expand=${2//,/ }
+ local expanded=
+
+ expanded=$(for i in $list $expand; do echo $i; done | sort -u)
+ echo $(comma_list $expanded)
+}
+
absolute_path() {
(cd `dirname $1`; echo $PWD/`basename $1`)
}
-##################################
-# Adaptive Timeouts funcs
+get_facets () {
+ local name=$(echo $1 | tr "[:upper:]" "[:lower:]")
+ local type=$(echo $1 | tr "[:lower:]" "[:upper:]")
-at_is_valid() {
- if [ -z "$AT_MAX_PATH" ]; then
- AT_MAX_PATH=$(do_facet $SINGLEMDS "find /sys/ -name at_max")
- [ -z "$AT_MAX_PATH" ] && echo "missing /sys/.../at_max " && return 1
- fi
- return 0
+ local list=""
+ local count=${type}COUNT
+ for ((i=1; i<=${!count}; i++)) do
+ list="$list ${name}$i"
+ done
+ echo $(comma_list $list)
}
-at_is_enabled() {
- at_is_valid || error "invalid call"
+##################################
+# Adaptive Timeouts funcs
+at_is_enabled() {
# only check mds, we assume at_max is the same on all nodes
- local at_max=$(do_facet $SINGLEMDS "cat $AT_MAX_PATH")
+ local at_max=$(do_facet $SINGLEMDS "lctl get_param -n at_max")
if [ $at_max -eq 0 ]; then
return 1
else
at_max_get() {
local facet=$1
- at_is_valid || error "invalid call"
-
# suppose that all ost-s has the same at_max set
if [ $facet == "ost" ]; then
- do_facet ost1 "cat $AT_MAX_PATH"
+ do_facet ost1 "lctl get_param -n at_max"
else
- do_facet $facet "cat $AT_MAX_PATH"
+ do_facet $facet "lctl get_param -n at_max"
fi
}
local at_max=$1
shift
- at_is_valid || error "invalid call"
-
local facet
for facet in $@; do
if [ $facet == "ost" ]; then
for i in `seq $OSTCOUNT`; do
- do_facet ost$i "echo $at_max > $AT_MAX_PATH"
+ do_facet ost$i "lctl set_param at_max=$at_max"
+
done
elif [ $facet == "mds" ]; then
for i in `seq $MDSCOUNT`; do
- do_facet mds$i "echo $at_max > $AT_MAX_PATH"
+ do_facet mds$i "lctl set_param at_max=$at_max"
done
else
- do_facet $facet "echo $at_max > $AT_MAX_PATH"
+ do_facet $facet "lctl set_param at_max=$at_max"
fi
done
}
}
set_nodes_failloc () {
- local nodes=$1
- local node
-
- for node in $nodes ; do
- do_node $node lctl set_param fail_loc=$2
- done
+ do_nodes $(comma_list $1) lctl set_param fail_loc=$2
}
cancel_lru_locks() {
ERRLOG=$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)
- for NODE in $NODES; do
- do_node $NODE $LCTL dk $ERRLOG
- done
+ do_nodes $(comma_list $(nodes_list)) $NODE $LCTL dk $ERRLOG
debugrestore
[ "$TESTSUITELOG" ] && echo "$0: ${TYPE}: $TESTNAME $@" >> $TESTSUITELOG
TEST_FAILED=true
}
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//\)/\\\)}
MSG=${MSG//\>/\\\>}
MSG=${MSG//\</\\\<}
MSG=${MSG//\//\\\/}
- local NODES=$(nodes_list)
- for NODE in $NODES; do
- do_node $NODE $LCTL mark "$MSG" 2> /dev/null || true
- done
+ do_nodes $(comma_list $(nodes_list)) $LCTL mark "$MSG" 2> /dev/null || true
}
trace() {
}
reset_fail_loc () {
- local myNODES=$(nodes_list)
- local NODE
-
- for NODE in $myNODES; do
- do_node $NODE "lctl set_param fail_loc=0 2>/dev/null || true"
- done
+ echo -n "Resetting fail_loc on all nodes..."
+ do_nodes $(comma_list $(nodes_list)) "lctl set_param -n fail_loc=0 2>/dev/null || true"
+ 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_$testnum
TEST_FAILED=false
remote_mds_nodsh()
{
+ [ "$CLIENTONLY" ] && return 0 || true
remote_mds && [ "$PDSH" = "no_dsh" -o -z "$PDSH" -o -z "$mds_HOST" ]
}
remote_ost_nodsh()
{
+ [ "$CLIENTONLY" ] && return 0 || true
remote_ost && [ "$PDSH" = "no_dsh" -o -z "$PDSH" -o -z "$ost_HOST" ]
}
rnodes=${rnodes//,/ }
- local nodes=($rnodes)
+ local -a nodes=($rnodes)
local num=${#nodes[@]}
- local i=$((RANDOM * num / 65536))
+ local i=$((RANDOM * num * 2 / 65536))
echo ${nodes[i]}
}
return 0
}
+do_and_time () {
+ local cmd=$1
+ local rc
+
+ SECONDS=0
+ eval '$cmd'
+
+ [ ${PIPESTATUS[0]} -eq 0 ] || rc=1
+
+ echo $SECONDS
+ return $rc
+}
+
inodes_available () {
local IFree=$($LFS df -i $MOUNT | grep ^$FSNAME | awk '{print $4}' | sort -un | head -1) || return 1
echo $IFree
}
calc_osc_kbytes () {
+ df $MOUNT > /dev/null
$LCTL get_param -n osc.*[oO][sS][cC][-_][0-9a-f]*.$1 | calc_sum
}
# generate a stream of formatted strings (<node> <param name>=<param value>)
save_lustre_params() {
local s
- do_node $1 "lctl get_param $2" | while read s; do echo "$1 $s"; done
+ do_nodes --verbose $1 "lctl get_param $2 | while read s; do echo \\\$s; done"
}
# restore lustre parameters from input stream, produces by save_lustre_params
local name
local val
while IFS=" =" read node name val; do
- do_node $node "lctl set_param -n $name $val"
+ do_node ${node//:/} "lctl set_param -n $name $val"
done
}
-check_catastrophe () {
+check_catastrophe() {
local rnodes=${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
}
stripe_size=`awk '$1 ~ /size/ {print $2}' $tmp_file`
stripe_count=`awk '$1 ~ /count/ {print $2}' $tmp_file`
- stripe_index=`awk '/obdidx/ {start = 1; getline; print $1; exit}' $tmp_file`
+ stripe_index=`awk '$1 ~ /stripe_offset/ {print $2}' $tmp_file`
rm -f $tmp_file
}
local file=$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)
+ local -a oldused=($iused)
touch $file
sleep 1
iused=$(lfs df -i $dir | grep MDT | awk '{print $3}')
- local newused=($iused)
+ local -a newused=($iused)
local num=0
for ((i=0; i<${#newused[@]}; i++)); do
error "mdt-s : inodes count OLD ${oldused[@]} NEW ${newused[@]}"
}
-mpi_run () {
- local mpirun="$MPIRUN $MPIRUN_OPTIONS"
- local command="$mpirun $@"
+mdsrate_cleanup () {
+ mpi_run -np $1 -machinefile $2 ${MDSRATE} --unlink --nfiles $3 --dir $4 --filefmt $5 $6
+}
- if [ "$MPI_USER" != root -a $mpirun ]; then
- echo "+ chmod 0777 $MOUNT"
- chmod 0777 $MOUNT
- command="su $MPI_USER sh -c \"$command \""
- fi
+delayed_recovery_enabled () {
+ local var=${SINGLEMDS}_svc
+ do_facet $SINGLEMDS lctl get_param -n mdd.${!var}.stale_export_age > /dev/null 2>&1
+}
- ls -ald $MOUNT
- echo "+ $command"
- eval $command
+########################
+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
}
-mdsrate_cleanup () {
- mpi_run -np $1 -machinefile $2 ${MDSRATE} --unlink --nfiles $3 --dir $4 --filefmt $5
+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 $CONN_PROC 2>/dev/null | cut -f2)
+ while [ "${CONN_STATE}" != "${expected}" ]; do
+ if [ "${expected}" == "DISCONN" ]; then
+ # for disconn we can check after proc entry is removed
+ [ "x${CONN_STATE}" == "x" ] && return 0
+ # with AT we can have connect request timeout ~ reconnect timeout
+ # and test can't see real disconnect
+ [ "${CONN_STATE}" == "CONNECTING" ] && return 0
+ fi
+ # 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 $CONN_PROC 2>/dev/null | cut -f2)
+ i=$(($i + 1))
+ done
+
+ log "${ost_facet} now in ${CONN_STATE} state"
+ return 0
+}
git://git.whamcloud.com / fs / lustre-release.git / blobdiff