ONLY=${ONLY:-"$*"}
# These tests don't apply to mountconf
-MOUNTCONFSKIP="9 10 11 12 13 13b 14 15 18"
+MOUNTCONFSKIP="9 10 11 12 13 13b 14 15"
# bug number for skipped test:
ALWAYS_EXCEPT=" $CONF_SANITY_EXCEPT $MOUNTCONFSKIP 16 23"
PATH=$PWD/$SRCDIR:$SRCDIR:$SRCDIR/../utils:$PATH
PTLDEBUG=${PTLDEBUG:--1}
+SAVE_PWD=$PWD
LUSTRE=${LUSTRE:-`dirname $0`/..}
RLUSTRE=${RLUSTRE:-$LUSTRE}
MOUNTLUSTRE=${MOUNTLUSTRE:-/sbin/mount.lustre}
echo "save $MOUNTLUSTRE to $MOUNTLUSTRE.sav"
mv $MOUNTLUSTRE $MOUNTLUSTRE.sav && trap cleanup_15 EXIT INT
if [ -f $MOUNTLUSTRE ]; then
- echo "$MOUNTLUSTRE cannot be moved, skipping test"
+ skip "$MOUNTLUSTRE cannot be moved, skipping test"
return 0
fi
fi
test_18() {
[ -f $MDSDEV ] && echo "remove $MDSDEV" && rm -f $MDSDEV
echo "mount mds with large journal..."
- OLDMDSSIZE=$MDSSIZE
- MDSSIZE=2000000
- #FIXME have to change MDS_MKFS_OPTS
- gen_config
+ local myMDSSIZE=2000000
+ OLD_MDS_MKFS_OPTS=$MDS_MKFS_OPTS
+
+ MDS_MKFS_OPTS="--mgs --mdt --fsname=$FSNAME --device-size=$myMDSSIZE --param sys.timeout=$TIMEOUT $MDSOPT"
+ gen_config
echo "mount lustre system..."
setup
check_mount || return 41
echo "check journal size..."
FOUNDSIZE=`do_facet mds "debugfs -c -R 'stat <8>' $MDSDEV" | awk '/Size: / { print $NF; exit;}'`
if [ $FOUNDSIZE -gt $((32 * 1024 * 1024)) ]; then
- log "Success: mkfs creates large journals"
+ log "Success: mkfs creates large journals. Size: $((FOUNDSIZE >> 20))M"
else
error "expected journal size > 32M, found $((FOUNDSIZE >> 20))M"
fi
cleanup || return $?
- MDSSIZE=$OLDMDSSIZE
+ MDS_MKFS_OPTS=$OLD_MDS_MKFS_OPTS
gen_config
}
run_test 18 "check mkfs creates large journals"
test_26() {
load_modules
# we need modules before mount for sysctl, so make sure...
- [ -z "$(lsmod | grep lustre)" ] && modprobe lustre
+ do_facet mds "lsmod | grep -q lustre || modprobe lustre"
#define OBD_FAIL_MDS_FS_SETUP 0x135
do_facet mds "sysctl -w lustre.fail_loc=0x80000135"
start_mds && echo MDS started && return 1
run_test 26 "MDT startup failure cleans LOV (should return errs)"
set_and_check() {
- local TEST=$1
- local PARAM=$2
- local ORIG=$($TEST)
- if [ $# -gt 2 ]; then
- local FINAL=$3
+ local myfacet=$1
+ local TEST=$2
+ local PARAM=$3
+ local ORIG=$(do_facet $myfacet "$TEST")
+ if [ $# -gt 3 ]; then
+ local FINAL=$4
else
local -i FINAL
FINAL=$(($ORIG + 5))
fi
echo "Setting $PARAM from $ORIG to $FINAL"
- $LCTL conf_param $PARAM=$FINAL
+ do_facet mds "$LCTL conf_param $PARAM=$FINAL" || error conf_param failed
local RESULT
- local MAX=20
+ local MAX=90
local WAIT=0
while [ 1 ]; do
sleep 5
- RESULT=$($TEST)
+ RESULT=$(do_facet $myfacet "$TEST")
if [ $RESULT -eq $FINAL ]; then
echo "Updated config after $WAIT sec (got $RESULT)"
break
start_mds || return 2
echo "Requeue thread should have started: "
ps -e | grep ll_cfg_requeue
- set_and_check "cat $LPROC/obdfilter/$FSNAME-OST0000/client_cache_seconds" "$FSNAME-OST0000.ost.client_cache_seconds" || return 3
+ set_and_check ost1 "cat $LPROC/obdfilter/$FSNAME-OST0000/client_cache_seconds" "$FSNAME-OST0000.ost.client_cache_seconds" || return 3
cleanup_nocli
}
run_test 27a "Reacquire MGS lock if OST started first"
test_27b() {
setup
facet_failover mds
- set_and_check "cat $LPROC/mdt/$FSNAME-MDT0000/identity_acquire_expire" "$FSNAME-MDT0000.mdt.identity_acquire_expire" || return 3
- set_and_check "cat $LPROC/mdc/$FSNAME-MDT0000-mdc-*/max_rpcs_in_flight" "$FSNAME-MDT0000.mdc.max_rpcs_in_flight" || return 4
+ set_and_check mds "cat $LPROC/mdt/$FSNAME-MDT0000/identity_acquire_expire" "$FSNAME-MDT0000.mdt.identity_acquire_expire" || return 3
+ set_and_check client "cat $LPROC/mdc/$FSNAME-MDT0000-mdc-*/max_rpcs_in_flight" "$FSNAME-MDT0000.mdc.max_rpcs_in_flight" || return 4
cleanup
}
run_test 27b "Reacquire MGS lock after failover"
ORIG=$($TEST)
declare -i FINAL
FINAL=$(($ORIG + 10))
- set_and_check "$TEST" "$FSNAME.llite.max_read_ahead_whole_mb" || return 3
- set_and_check "$TEST" "$FSNAME.llite.max_read_ahead_whole_mb" || return 3
+ set_and_check client "$TEST" "$FSNAME.llite.max_read_ahead_whole_mb" || return 3
+ set_and_check client "$TEST" "$FSNAME.llite.max_read_ahead_whole_mb" || return 3
umount_client $MOUNT || return 200
mount_client $MOUNT
RESULT=$($TEST)
run_test 28 "permanent parameter setting"
test_29() {
- [ "$OSTCOUNT" -lt "2" ] && echo "skipping deactivate test" && return
+ [ "$OSTCOUNT" -lt "2" ] && skip "$OSTCOUNT < 2, skipping" && return
setup > /dev/null 2>&1
start_ost2
sleep 10
fi
ACTV=$(cat $PROC_ACT)
DEAC=$((1 - $ACTV))
- set_and_check "cat $PROC_ACT" "$PARAM" $DEAC || return 2
+ set_and_check client "cat $PROC_ACT" "$PARAM" $DEAC || return 2
# also check ost_server_uuid status
RESULT=$(grep DEACTIV $PROC_UUID)
if [ -z "$RESULT" ]; then
# check MDT too
local MPROC="$LPROC/osc/$FSNAME-OST0001-osc-[M]*/active"
- if [ -r $MPROC ]; then
- RESULT=$(cat $MPROC)
- if [ $RESULT -ne $DEAC ]; then
- echo "MDT not deactivated: $(cat $MPROC)"
+ local MAX=30
+ local WAIT=0
+ while [ 1 ]; do
+ sleep 5
+ RESULT=`do_facet mds " [ -r $MPROC ] && cat $MPROC"`
+ [ ${PIPESTATUS[0]} = 0 ] || error "Can't read $MPROC"
+ if [ $RESULT -eq $DEAC ]; then
+ echo "MDT deactivated also after $WAIT sec (got $RESULT)"
+ break
+ fi
+ WAIT=$((WAIT + 5))
+ if [ $WAIT -eq $MAX ]; then
+ echo "MDT not deactivated: wanted $DEAC got $RESULT"
return 4
fi
- echo "MDT deactivated also"
- fi
+ echo "Waiting $(($MAX - $WAIT)) secs for MDT deactivated"
+ done
# test new client starts deactivated
umount_client $MOUNT || return 200
fi
# make sure it reactivates
- set_and_check "cat $PROC_ACT" "$PARAM" $ACTV || return 6
+ set_and_check client "cat $PROC_ACT" "$PARAM" $ACTV || return 6
umount_client $MOUNT
stop_ost2
TEST="cat $LPROC/llite/$FSNAME-*/max_read_ahead_whole_mb"
ORIG=$($TEST)
for i in $(seq 1 20); do
- set_and_check "$TEST" "$FSNAME.llite.max_read_ahead_whole_mb" $i || return 3
+ set_and_check client "$TEST" "$FSNAME.llite.max_read_ahead_whole_mb" $i || return 3
done
# make sure client restart still works
umount_client $MOUNT
mount_client $MOUNT || return 4
[ "$($TEST)" -ne "$i" ] && return 5
- set_and_check "$TEST" "$FSNAME.llite.max_read_ahead_whole_mb" $ORIG || return 6
+ set_and_check client "$TEST" "$FSNAME.llite.max_read_ahead_whole_mb" $ORIG || return 6
cleanup
}
run_test 30 "Big config llog"
run_test 31 "Connect to non-existent node (shouldn't crash)"
test_32a() {
- [ -z "$TUNEFS" ] && echo "No tunefs" && return
- [ ! -r disk1_4.zip ] && echo "Cant find disk1_4.zip, skipping" && return
- unzip -o -j -d $TMP/$tdir disk1_4.zip || { echo "Cant unzip disk1_4, skipping" && return ; }
+ # XXX - make this run on client-only systems with real hardware on
+ # the OST and MDT
+ # there appears to be a lot of assumption here about loopback
+ # devices
+ # or maybe this test is just totally useless on a client-only system
+ [ -z "$TUNEFS" ] && skip "No tunefs" && return
+ local DISK1_4=$LUSTRE/tests/disk1_4.zip
+ [ ! -r $DISK1_4 ] && skip "Cant find $DISK1_4, skipping" && return
+ unzip -o -j -d $TMP/$tdir $DISK1_4 || { skip "Cant unzip $DISK1_4, skipping" && return ; }
load_modules
sysctl lnet.debug=$PTLDEBUG
MOUNTOPT="exclude=lustre-OST0000"
mount_client $MOUNT
MOUNTOPT=$OLDMOUNTOPT
- set_and_check "cat $LPROC/mdc/*/max_rpcs_in_flight" "lustre-MDT0000.mdc.max_rpcs_in_flight" || return 11
+ set_and_check client "cat $LPROC/mdc/*/max_rpcs_in_flight" "lustre-MDT0000.mdc.max_rpcs_in_flight" || return 11
zconf_umount `hostname` $MOUNT -f
cleanup_nocli
run_test 32a "Upgrade from 1.4 (not live)"
test_32b() {
- [ -z "$TUNEFS" ] && echo "No tunefs" && return
- [ ! -r disk1_4.zip ] && echo "Cant find disk1_4.zip, skipping" && return
- unzip -o -j -d $TMP/$tdir disk1_4.zip || { echo "Cant unzip disk1_4, skipping" && return ; }
+ # XXX - make this run on client-only systems with real hardware on
+ # the OST and MDT
+ # there appears to be a lot of assumption here about loopback
+ # devices
+ # or maybe this test is just totally useless on a client-only system
+ [ -z "$TUNEFS" ] && skip "No tunefs" && return
+ local DISK1_4=$LUSTRE/tests/disk1_4.zip
+ [ ! -r $DISK1_4 ] && skip "Cant find $DISK1_4, skipping" && return
+ unzip -o -j -d $TMP/$tdir $DISK1_4 || { skip "Cant unzip $DISK1_4, skipping" && return ; }
load_modules
sysctl lnet.debug=$PTLDEBUG
# a fully-functioning client
echo "Check client and old fs contents"
mount_client $MOUNT
- set_and_check "cat $LPROC/mdc/*/max_rpcs_in_flight" "lustre-MDT0000.mdc.max_rpcs_in_flight" || return 11
+ set_and_check client "cat $LPROC/mdc/*/max_rpcs_in_flight" "${NEWNAME}-MDT0000.mdc.max_rpcs_in_flight" || return 11
[ "$(cksum $MOUNT/passwd | cut -d' ' -f 1,2)" == "2479747619 779" ] || return 12
echo "ok."
}
run_test 33 "Mount ost with a large index number"
+test_35() { # bug 12459
+ setup
+
+ DBG_SAVE="`sysctl -n lnet.debug`"
+ sysctl -w lnet.debug="ha"
+
+ log "Set up a fake failnode for the MDS"
+ FAKENID="127.0.0.2"
+ do_facet mds $LCTL conf_param ${FSNAME}-MDT0000.failover.node=$FAKENID || return 4
+
+ log "Wait for RECONNECT_INTERVAL seconds (10s)"
+ sleep 10
+
+ MSG="conf-sanity.sh test_33 `date +%F%kh%Mm%Ss`"
+ $LCTL clear
+ log "$MSG"
+ log "Stopping the MDT:"
+ stop_mds || return 5
+
+ df $MOUNT > /dev/null 2>&1 &
+ DFPID=$!
+ log "Restarting the MDT:"
+ start_mds || return 6
+ log "Wait for df ($DFPID) ... "
+ wait $DFPID
+ log "done"
+ sysctl -w lnet.debug="$DBG_SAVE"
+
+ # retrieve from the log the first server that the client tried to
+ # contact after the connection loss
+ $LCTL dk $TMP/lustre-log-$TESTNAME.log
+ NEXTCONN=`awk "/${MSG}/ {start = 1;}
+ /import_select_connection.*${FSNAME}-MDT0000-mdc.* using connection/ {
+ if (start) {
+ if (\\\$NF ~ /$FAKENID/)
+ print \\\$NF;
+ else
+ print 0;
+ exit;
+ }
+ }" $TMP/lustre-log-$TESTNAME.log`
+ [ "$NEXTCONN" != "0" ] && log "The client didn't try to reconnect to the last active server (tried ${NEXTCONN} instead)" && return 7
+ cleanup
+}
+run_test 35 "Reconnect to the last active server first"
+
+test_36() { # 12743
+ local rc
+ local FSNAME2=test1234
+ local fs2mds_HOST=$mds_HOST
+ local fs2ost_HOST=$ost_HOST
+ local fs3ost_HOST=$ost_HOST
+ rc=0
+
+ [ -z "$fs2ost_DEV" -o -z "$fs2mds_DEV" ] && [ -b "$MDSDEV" ] && \
+ log "mixed loopback and real device not working" && return
+
+ [ $OSTCOUNT -lt 2 ] && skip "skipping test for single OST" && return
+
+ local fs2mdsdev=${fs2mds_DEV:-${MDSDEV}_2}
+ local fs2ostdev=${fs2ost_DEV:-$(ostdevname 1)_2}
+ local fs3ostdev=${fs3ost_DEV:-$(ostdevname 2)_2}
+ add fs2mds $MDS_MKFS_OPTS --fsname=${FSNAME2} --reformat $fs2mdsdev || exit 10
+ add fs2ost $OST_MKFS_OPTS --mkfsoptions='-b1024' --fsname=${FSNAME2} --mgsnode=`hostname`@tcp --reformat $fs2ostdev || exit 10
+ add fs3ost $OST_MKFS_OPTS --mkfsoptions='-b4096' --fsname=${FSNAME2} --mgsnode=`hostname`@tcp --reformat $fs3ostdev || exit 10
+
+ start fs2mds $fs2mdsdev $MDS_MOUNT_OPTS
+ start fs2ost $fs2ostdev $OST_MOUNT_OPTS
+ start fs3ost $fs3ostdev $OST_MOUNT_OPTS
+ mkdir -p $MOUNT2
+ mount -t lustre $MGSNID:/${FSNAME2} $MOUNT2 || return 1
+
+ dd if=/dev/zero of=$MOUNT2/$tfile bs=1M count=7 || return 2
+ [ $(grep -c obdfilter $LPROC/devices) -eq 0 ] &&
+ skip "skipping test for remote OST" && return
+ BKTOTAL=`awk 'BEGIN{total=0}; {total+=$1}; END{print total}' \
+ $LPROC/obdfilter/*/kbytestotal`
+ BKFREE=`awk 'BEGIN{free=0}; {free+=$1}; END{print free}' \
+ $LPROC/obdfilter/*/kbytesfree`
+ BKAVAIL=`awk 'BEGIN{avail=0}; {avail+=$1}; END{print avail}' \
+ $LPROC/obdfilter/*/kbytesavail`
+ STRING=`df -P $MOUNT2 | tail -n 1 | awk '{print $2","$3","$4}'`
+ DFTOTAL=`echo $STRING | cut -d, -f1`
+ DFUSED=`echo $STRING | cut -d, -f2`
+ DFAVAIL=`echo $STRING | cut -d, -f3`
+ DFFREE=$(($DFTOTAL - $DFUSED))
+
+ ALLOWANCE=$((64 * $OSTCOUNT))
+
+ if [ $DFTOTAL -lt $(($BKTOTAL - $ALLOWANCE)) ] ||
+ [ $DFTOTAL -gt $(($BKTOTAL + $ALLOWANCE)) ] ; then
+ echo "**** FAIL: df total($DFTOTAL) mismatch OST total($BKTOTAL)"
+ rc=1
+ fi
+ if [ $DFFREE -lt $(($BKFREE - $ALLOWANCE)) ] ||
+ [ $DFFREE -gt $(($BKFREE + $ALLOWANCE)) ] ; then
+ echo "**** FAIL: df free($DFFREE) mismatch OST free($BKFREE)"
+ rc=2
+ fi
+ if [ $DFAVAIL -lt $(($BKAVAIL - $ALLOWANCE)) ] ||
+ [ $DFAVAIL -gt $(($BKAVAIL + $ALLOWANCE)) ] ; then
+ echo "**** FAIL: df avail($DFAVAIL) mismatch OST avail($BKAVAIL)"
+ rc=3
+ fi
+
+ umount -d $MOUNT2
+ stop fs3ost -f || return 200
+ stop fs2ost -f || return 201
+ stop fs2mds -f || return 202
+ rm -rf $MOUNT2 $fs2mdsdev $fs2ostdev $fs3ostdev
+ unload_modules || return 203
+ return $rc
+}
+run_test 36 "df report consistency on OSTs with different block size"
+
umount_client $MOUNT
cleanup_nocli
cleanup_krb5_env
-equals_msg "Done"
-echo "$0: completed"
+equals_msg `basename $0`: test complete
+[ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG || true
git://git.whamcloud.com / fs / lustre-release.git / blobdiff