export TESTSUITE=`basename $0 .sh`
export TEST_FAILED=false
+ export MKE2FS=${MKE2FS:-mke2fs}
+ export DEBUGFS=${DEBUGFS:-debugfs}
+ export TUNE2FS=${TUNE2FS:-tune2fs}
+ export E2LABEL=${E2LABEL:-e2label}
+ export DUMPE2FS=${DUMPE2FS:-dumpe2fs}
+ export E2FSCK=${E2FSCK:-e2fsck}
+
#[ -d /r ] && export ROOT=${ROOT:-/r}
export TMP=${TMP:-$ROOT/tmp}
export TESTSUITELOG=${TMP}/${TESTSUITE}.log
fi
export MDSRATE=${MDSRATE:-"$LUSTRE/tests/mdsrate"}
[ ! -f "$MDSRATE" ] && export MDSRATE=$(which mdsrate 2> /dev/null)
- if ! echo $PATH | grep -q $LUSTRE/test/racer; then
+ if ! echo $PATH | grep -q $LUSTRE/tests/racer; then
export PATH=$PATH:$LUSTRE/tests/racer
fi
export LCTL=${LCTL:-"$LUSTRE/utils/lctl"}
export TUNEFS=${TUNEFS:-"$LUSTRE/utils/tunefs.lustre"}
[ ! -f "$TUNEFS" ] && export TUNEFS=$(which tunefs.lustre)
export CHECKSTAT="${CHECKSTAT:-"checkstat -v"} "
- export FSYTPE=${FSTYPE:-"ldiskfs"}
+ export FSTYPE=${FSTYPE:-"ldiskfs"}
export NAME=${NAME:-local}
export LGSSD=${LGSSD:-"$LUSTRE/utils/gss/lgssd"}
[ "$GSS_PIPEFS" = "true" ] && [ ! -f "$LGSSD" ] && \
load_module mgc/mgc
if [ -z "$CLIENTONLY" ] && [ -z "$CLIENTMODSONLY" ]; then
grep -q crc16 /proc/kallsyms || { modprobe crc16 2>/dev/null || true; }
+ grep -q jbd /proc/kallsyms || { modprobe jbd 2>/dev/null || true; }
[ "$FSTYPE" = "ldiskfs" ] && load_module ../ldiskfs/ldiskfs/ldiskfs
load_module mgs/mgs
load_module mds/mds
init_gss() {
if $GSS; then
start_gss_daemons
+
+ if [ -n "$LGSS_KEYRING_DEBUG" ]; then
+ echo $LGSS_KEYRING_DEBUG > /proc/fs/lustre/sptlrpc/gss/lgss_keyring/debug_level
+ fi
fi
}
lctl set_param debug_mb=${DEBUG_SIZE}; \
sync"
- label=$(do_facet ${facet} "e2label ${!dev}")
+ label=$(do_facet ${facet} "$E2LABEL ${!dev}")
[ -z "$label" ] && echo no label for ${!dev} && exit 1
eval export ${facet}_svc=${label}
echo Started ${label}
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
+ if [ -n \\\"\\\$pids\\\" ]; then
kill -9 \\\$pids
fi
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 $facet
fi
}
start_client_load() {
- local list=(${1//,/ })
- local nodenum=$2
-
- local numloads=${#CLIENT_LOADS[@]}
- local testnum=$((nodenum % numloads))
+ local client=$1
+ local var=${client}_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_${CLIENT_LOADS[testnum]}.sh" &
+ run_${!var}.sh" &
CLIENT_LOAD_PIDS="$CLIENT_LOAD_PIDS $!"
- log "Started client load: ${CLIENT_LOADS[testnum]} on ${list[nodenum]}"
+ log "Started client load: ${!var} on $client"
- eval export ${list[nodenum]}_load=${CLIENT_LOADS[testnum]}
return 0
}
start_client_loads () {
local 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))
+ eval export ${clients[nodenum]}_load=${CLIENT_LOADS[testnum]}
+ start_client_load ${clients[nodenum]}
done
}
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
return 0
}
+wait_update () {
+ local node=$1
+ local TEST=$2
+ local FINAL=$3
+ local MAX=${4:-90}
+
+ local RESULT
+ local WAIT=0
+ local sleep=5
+ while [ $WAIT -lt $MAX ]; do
+ RESULT=$(do_node $node "$TEST")
+ if [ "$RESULT" == "$FINAL" ]; then
+ echo "Updated after $WAIT sec: wanted '$FINAL' got '$RESULT'"
+ return 0
+ fi
+ echo "Waiting $((MAX - WAIT)) secs for update"
+ WAIT=$((WAIT + sleep))
+ sleep $sleep
+ done
+ echo "Update not seen after $MAX sec: wanted '$FINAL' got '$RESULT'"
+ return 3
+}
+
+wait_update_facet () {
+ local facet=$1
+ wait_update $(facet_host $facet) $@
+}
+
wait_delete_completed () {
local TOTALPREV=`lctl get_param -n osc.*.kbytesavail | \
awk 'BEGIN{total=0}; {total+=$1}; END{print total}'`
}
wait_for_host() {
- 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() {
- facet=$1
- HOST=`facet_active_host $facet`
- wait_for_host $HOST
+ local facet=$1
+ local host=`facet_active_host $facet`
+ wait_for_host $host
}
-wait_mds_recovery_done () {
- local timeout=`do_facet $SINGLEMDS lctl get_param -n timeout`
-#define OBD_RECOVERY_TIMEOUT (obd_timeout * 5 / 2)
-# as we are in process of changing obd_timeout in different ways
-# let's set MAX longer than that
- MAX=$(( timeout * 4 ))
- WAIT=0
+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
}
}
facet_failover() {
- facet=$1
- sleep_time=$2
+ local facet=$1
+ local sleep_time=$2
echo "Failing $facet on node `facet_active_host $facet`"
shutdown_facet $facet
[ -n "$sleep_time" ] && sleep $sleep_time
DFPID=$!
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"
zconf_mount `hostname` $1 || error "mount failed"
}
-set_obd_timeout() {
- local facet=$1
- local timeout=$2
-
- do_facet $facet lsmod | grep -q obdclass || \
- do_facet $facet "modprobe obdclass"
-
- do_facet $facet "lctl set_param timeout=$timeout"
-}
-
writeconf_facet () {
local facet=$1
local dev=$2
writeconf_all
for num in `seq $MDSCOUNT`; do
DEVNAME=$(mdsdevname $num)
- set_obd_timeout mds$num $TIMEOUT
start mds$num $DEVNAME $MDS_MOUNT_OPTS
# We started mds, now we should set failover variables properly.
done
for num in `seq $OSTCOUNT`; do
DEVNAME=$(ostdevname $num)
- set_obd_timeout ost$num $TIMEOUT
start ost$num $DEVNAME $OST_MOUNT_OPTS
# We started ost$num, now we should set ost${num}failover variable properly.
[ -n "$CLIENTS" ] && zconf_mount_clients $CLIENTS $MOUNT2
fi
- init_versions_vars
+ init_param_vars
# by remounting mdt before ost, initial connect from mdt to ost might
# timeout because ost is not ready yet. wait some time to its fully
eval export ${facet}_opt=\"$@\"
local dev=${facet}_dev
- local label=$(do_facet ${facet} "e2label ${!dev}")
+ local label=$(do_facet ${facet} "$E2LABEL ${!dev}")
[ -z "$label" ] && echo no label for ${!dev} && exit 1
eval export ${facet}_svc=${label}
done
}
-init_versions_vars () {
+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=$(lctl get_param version | cut -d. -f 1,2)
+
+ TIMEOUT=$(do_facet $SINGLEMDS "lctl get_param -n timeout")
+ log "Using TIMEOUT=$TIMEOUT"
}
check_config () {
fi
}
+check_timeout () {
+ local mdstimeout=$(do_facet $SINGLEMDS "lctl get_param -n timeout")
+ local cltimeout=$(lctl get_param -n timeout)
+ if [ $mdstimeout -ne $TIMEOUT ] || [ $mdstimeout -ne $cltimeout ]; then
+ error "timeouts are wrong! mds: $mdstimeout, client: $cltimeout, TIMEOUT=$TIMEOUT"
+ return 1
+ fi
+}
+
check_and_setup_lustre() {
local MOUNTED=$(mounted_lustre_filesystems)
if [ -z "$MOUNTED" ] || ! $(echo $MOUNTED | grep -w -q $MOUNT); then
else
check_config $MOUNT
init_facets_vars
- init_versions_vars
+ init_param_vars
fi
if [ "$ONLY" == "setup" ]; then
exit 0
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 $1`)
}
+get_facets () {
+ local name=$(echo $1 | tr "[:upper:]" "[:lower:]")
+ local type=$(echo $1 | tr "[:lower:]" "[:upper:]")
+
+ local list=""
+ local count=${type}COUNT
+ for ((i=1; i<=${!count}; i++)) do
+ list="$list ${name}$i"
+ done
+ echo $(comma_list $list)
+}
+
##################################
# Adaptive Timeouts funcs
local SAVE_UMASK=`umask`
umask 0022
- BEFORE=`date +%s`
+ local BEFORE=`date +%s`
log "== test $testnum: $message ============ `date +%H:%M:%S` ($BEFORE)"
#check_mds
export TESTNAME=test_$testnum
remote_ost && [ "$PDSH" = "no_dsh" -o -z "$PDSH" -o -z "$ost_HOST" ]
}
+remote_mgs_nodsh()
+{
+ local MGS
+ MGS=$(facet_host mgs)
+ remote_node $MGS && [ "$PDSH" = "no_dsh" -o -z "$PDSH" -o -z "$ost_HOST" ]
+}
+
mdts_nodes () {
local MDSNODES
local NODES_sort
local nodes=($rnodes)
local num=${#nodes[@]}
- local i=$((RANDOM * num / 65536))
+ local i=$((RANDOM * num * 2 / 65536))
echo ${nodes[i]}
}
return 0
}
-check_rate() {
- local OP=$1
- local TARGET_RATE=$2
- local NUM_CLIENTS=$3
- local LOG=$4
-
- local RATE=$(awk '/^Rate: [0-9\.]+ '"${OP}"'s\/sec/ { print $2}' ${LOG})
-
- # We need to use bc since the rate is a floating point number
- local RES=$(echo "${RATE} < ${TARGET_RATE}" | bc -l )
- if [ "${RES}" = 0 ]; then
- echo "Success: ${RATE} ${OP}s/sec met target rate" \
- "${TARGET_RATE} ${OP}s/sec for ${NUM_CLIENTS} client(s)."
- return 0
- else
- echo "Failure: ${RATE} ${OP}s/sec did not meet target rate" \
- "${TARGET_RATE} ${OP}s/sec for ${NUM_CLIENTS} client(s)."
- return 1
- fi
+inodes_available () {
+ local IFree=$($LFS df -i $MOUNT | grep ^$FSNAME | awk '{print $4}' | sort -un | head -1) || return 1
+ echo $IFree
}
# reset llite stat counters
# $1 node
# $2 file
+# $3 $RUNAS
get_stripe_info() {
local tmp_file
stripe_index=0
tmp_file=$(mktemp)
- do_facet $1 lfs getstripe -v $2 > $tmp_file
+ do_facet $1 $3 lfs getstripe -v $2 > $tmp_file
stripe_size=`awk '$1 ~ /size/ {print $2}' $tmp_file`
stripe_count=`awk '$1 ~ /count/ {print $2}' $tmp_file`
# CMD: determine mds index where directory inode presents
get_mds_dir () {
local dir=$1
- local file=$dir/$tfile
+ local file=$dir/f0.get_mds_dir_tmpfile
rm -f $file
local iused=$(lfs df -i $dir | grep MDT | awk '{print $3}')
for ((i=0; i<${#newused[@]}; i++)); do
if [ ${oldused[$i]} -lt ${newused[$i]} ]; then
echo $(( i + 1 ))
- rm -f $dir/$tfile
+ rm -f $file
return 0
fi
done
eval $command
}
+mdsrate_cleanup () {
+ mpi_run -np $1 -machinefile $2 ${MDSRATE} --unlink --nfiles $3 --dir $4 --filefmt $5
+}
+
git://git.whamcloud.com / fs / lustre-release.git / blobdiff