X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Ftests%2Fconf-sanity.sh;h=c306f4a7520466926c33f881685a4c5458300e62;hp=9c259301eb9295d0c05cbd6ff9fc7f7cd54bb5af;hb=86b2211e55dcc509da85b21ece8830e2a9b70db1;hpb=95edc0dd8eab493f35b18aa1d0b66f32e1bcbf2d diff --git a/lustre/tests/conf-sanity.sh b/lustre/tests/conf-sanity.sh index 9c25930..c306f4a 100644 --- a/lustre/tests/conf-sanity.sh +++ b/lustre/tests/conf-sanity.sh @@ -16,6 +16,12 @@ ONLY=${ONLY:-"$*"} ALWAYS_EXCEPT="$CONF_SANITY_EXCEPT" # UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT! +if [ "$FAILURE_MODE" = "HARD" ]; then + CONFIG_EXCEPTIONS="24a " && \ + echo "Except the tests: $CONFIG_EXCEPTIONS for FAILURE_MODE=$FAILURE_MODE, bug 23573" && \ + ALWAYS_EXCEPT="$ALWAYS_EXCEPT $CONFIG_EXCEPTIONS" +fi + SRCDIR=`dirname $0` PATH=$PWD/$SRCDIR:$SRCDIR:$SRCDIR/../utils:$PATH @@ -33,6 +39,11 @@ MDSSIZE=200000 OSTSIZE=200000 . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh} +if ! combined_mgs_mds; then + # bug number for skipped test: 23954 + ALWAYS_EXCEPT="$ALWAYS_EXCEPT 24b" +fi + # STORED_MDSSIZE is used in test_18 if [ -n "$MDSSIZE" ]; then STORED_MDSSIZE=$MDSSIZE @@ -40,10 +51,12 @@ fi init_logging +# require_dsh_mds || exit 0 require_dsh_ost || exit 0 +# +[ "$SLOW" = "no" ] && EXCEPT_SLOW="30a 31 45" -[ "$SLOW" = "no" ] && EXCEPT_SLOW="0 1 2 3 6 7 15 18 24b 25 30 31 32 33 34a 45" assert_DIR @@ -51,16 +64,25 @@ reformat() { formatall } -writeconf() { - local facet=$SINGLEMDS - local dev=${facet}_dev - shift +writeconf1() { + local facet=$1 + local dev=$2 + stop ${facet} -f rm -f ${facet}active # who knows if/where $TUNEFS is installed? Better reformat if it fails... - do_facet ${facet} "$TUNEFS --writeconf ${!dev}" || - echo "tunefs failed, reformatting instead" && reformat_and_config + do_facet ${facet} "$TUNEFS --quiet --writeconf $dev" || + { echo "tunefs failed, reformatting instead" && reformat_and_config && return 1; } + return 0 +} +writeconf() { + # we need ldiskfs + load_modules + # if writeconf fails anywhere, we reformat everything + writeconf1 mds `mdsdevname 1` || return 0 + writeconf1 ost1 `ostdevname 1` || return 0 + writeconf1 ost2 `ostdevname 2` || return 0 } gen_config() { @@ -83,7 +105,7 @@ reformat_and_config() { start_mgs () { echo "start mgs" - start mgs $MGSDEV $mgs_MOUNT_OPTS + start mgs $MGSDEV $MGS_MOUNT_OPTS } start_mds() { @@ -92,7 +114,14 @@ start_mds() { local num=$(echo $facet | tr -d "mds") local dev=$(mdsdevname $num) echo "start mds service on `facet_active_host $facet`" - start $facet ${dev} $MDS_MOUNT_OPTS || return 94 + start $facet ${dev} $MDS_MOUNT_OPTS $@ || return 94 +} + +start_mgsmds() { + if ! combined_mgs_mds ; then + start_mgs + fi + start_mds $@ } stop_mds() { @@ -101,9 +130,15 @@ stop_mds() { stop $SINGLEMDS -f || return 97 } +stop_mgs() { + echo "stop mgs service on `facet_active_host mgs`" + # These tests all use non-failover stop + stop mgs -f || return 97 +} + start_ost() { echo "start ost1 service on `facet_active_host ost1`" - start ost1 `ostdevname 1` $OST_MOUNT_OPTS || return 95 + start ost1 `ostdevname 1` $OST_MOUNT_OPTS $@ || return 95 } stop_ost() { @@ -114,7 +149,7 @@ stop_ost() { start_ost2() { echo "start ost2 service on `facet_active_host ost2`" - start ost2 `ostdevname 2` $OST_MOUNT_OPTS || return 92 + start ost2 `ostdevname 2` $OST_MOUNT_OPTS $@ || return 92 } stop_ost2() { @@ -130,12 +165,10 @@ mount_client() { } remount_client() { - local SAVEMOUNTOPT=$MOUNTOPT - MOUNTOPT="remount,$1" + local mountopt="-o remount,$1" local MOUNTPATH=$2 echo "remount '$1' lustre on ${MOUNTPATH}....." - zconf_mount `hostname` $MOUNTPATH || return 96 - MOUNTOPT=$SAVEMOUNTOPT + zconf_mount `hostname` $MOUNTPATH "$mountopt" || return 96 } umount_client() { @@ -157,6 +190,7 @@ setup() { start_mds || error "MDT start failed" start_ost || error "OST start failed" mount_client $MOUNT || error "client start failed" + client_up || error "client_up failed" } setup_noconfig() { @@ -169,10 +203,16 @@ setup_noconfig() { mount_client $MOUNT } +unload_modules_conf () { + if combined_mgs_mds || ! local_mode; then + unload_modules || return 1 + fi +} + cleanup_nocli() { - stop_mds || return 201 stop_ost || return 202 - unload_modules || return 203 + stop_mds || return 201 + unload_modules_conf || return 203 } cleanup() { @@ -316,60 +356,132 @@ test_5a() { # was test_5 } run_test 5a "force cleanup mds, then cleanup" +cleanup_5b () { + trap 0 + start_mgs +} + test_5b() { + grep " $MOUNT " /etc/mtab && \ + error false "unexpected entry in mtab before mount" && return 10 + + local rc=0 start_ost + if ! combined_mgs_mds ; then + trap cleanup_5b EXIT ERR + start_mds + stop mgs + fi + [ -d $MOUNT ] || mkdir -p $MOUNT - grep " $MOUNT " /etc/mtab && echo "test 5b: mtab before mount" && return 10 - mount_client $MOUNT && return 1 - grep " $MOUNT " /etc/mtab && echo "test 5b: mtab after failed mount" && return 11 + mount_client $MOUNT && rc=1 + grep " $MOUNT " /etc/mtab && \ + error "$MOUNT entry in mtab after failed mount" && rc=11 umount_client $MOUNT # stop_mds is a no-op here, and should not fail - cleanup_nocli || return $? - return 0 + cleanup_nocli || rc=$? + if ! combined_mgs_mds ; then + cleanup_5b + fi + return $rc } -run_test 5b "mds down, cleanup after failed mount (bug 2712) (should return errs)" +run_test 5b "Try to start a client with no MGS (should return errs)" test_5c() { + grep " $MOUNT " /etc/mtab && \ + error false "unexpected entry in mtab before mount" && return 10 + + local rc=0 start_mds start_ost [ -d $MOUNT ] || mkdir -p $MOUNT - grep " $MOUNT " /etc/mtab && echo "test 5c: mtab before mount" && return 10 local oldfs="${FSNAME}" FSNAME="wrong.${FSNAME}" mount_client $MOUNT || : FSNAME=${oldfs} - grep " $MOUNT " /etc/mtab && echo "test 5c: mtab after failed mount" && return 11 + grep " $MOUNT " /etc/mtab && \ + error "$MOUNT entry in mtab after failed mount" && rc=11 umount_client $MOUNT - cleanup_nocli || return $? + cleanup_nocli || rc=$? + return $rc } run_test 5c "cleanup after failed mount (bug 2712) (should return errs)" test_5d() { + grep " $MOUNT " /etc/mtab && \ + error false "unexpected entry in mtab before mount" && return 10 + + local rc=0 start_ost start_mds stop_ost -f - grep " $MOUNT " /etc/mtab && echo "test 5d: mtab before mount" && return 10 - mount_client $MOUNT || return 1 - cleanup || return $? - grep " $MOUNT " /etc/mtab && echo "test 5d: mtab after unmount" && return 11 - return 0 + mount_client $MOUNT || rc=1 + cleanup || rc=$? + grep " $MOUNT " /etc/mtab && \ + error "$MOUNT entry in mtab after unmount" && rc=11 + return $rc } run_test 5d "mount with ost down" test_5e() { + grep " $MOUNT " /etc/mtab && \ + error false "unexpected entry in mtab before mount" && return 10 + + local rc=0 start_mds start_ost #define OBD_FAIL_PTLRPC_DELAY_SEND 0x506 do_facet client "lctl set_param fail_loc=0x80000506" - grep " $MOUNT " /etc/mtab && echo "test 5e: mtab before mount" && return 10 mount_client $MOUNT || echo "mount failed (not fatal)" - cleanup || return $? - grep " $MOUNT " /etc/mtab && echo "test 5e: mtab after unmount" && return 11 - return 0 + cleanup || rc=$? + grep " $MOUNT " /etc/mtab && \ + error "$MOUNT entry in mtab after unmount" && rc=11 + return $rc } run_test 5e "delayed connect, don't crash (bug 10268)" +test_5f() { + if combined_mgs_mds ; then + skip "combined mgs and mds" + return 0 + fi + + grep " $MOUNT " /etc/mtab && \ + error false "unexpected entry in mtab before mount" && return 10 + + local rc=0 + start_ost + [ -d $MOUNT ] || mkdir -p $MOUNT + mount_client $MOUNT & + local pid=$! + echo client_mount pid is $pid + + sleep 5 + + if ! ps -f -p $pid >/dev/null; then + wait $pid + rc=$? + grep " $MOUNT " /etc/mtab && echo "test 5f: mtab after mount" + error "mount returns $rc, expected to hang" + rc=11 + cleanup || rc=$? + return $rc + fi + + # start mds + start_mds + + # mount should succeed after start mds + wait $pid + rc=$? + [ $rc -eq 0 ] || error "mount returned $rc" + grep " $MOUNT " /etc/mtab && echo "test 5f: mtab after mount" + cleanup || return $? + return $rc +} +run_test 5f "mds down, cleanup after failed mount (bug 2712)" + test_6() { setup manual_umount_client @@ -419,70 +531,55 @@ test_9() { fi stop_ost || return $? } - run_test 9 "test ptldebug and subsystem for mkfs" -# LOGS/PENDING do not exist anymore since CMD3 -test_16() { - local TMPMTPT="${TMP}/conf16" - local dev=${SINGLEMDS}_dev - local MDSDEV=${!dev} - if [ ! -e "$MDSDEV" ]; then - log "no $MDSDEV existing, so mount Lustre to create one" - setup - check_mount || return 41 - cleanup || return $? - fi +is_blkdev () { + local facet=$1 + local dev=$2 + local size=${3:-""} - [ -f "$MDSDEV" ] && LOOPOPT="-o loop" + local rc=0 + do_facet $facet "test -b $dev" || rc=1 + if [[ "$size" ]]; then + local in=$(do_facet $facet "dd if=$dev of=/dev/null bs=1k count=1 skip=$size 2>&1" |\ + awk '($3 == "in") { print $1 }') + [[ $in = "1+0" ]] || rc=1 + fi + return $rc +} - log "change the mode of $MDSDEV/OBJECTS to 555" - do_facet $SINGLEMDS "mkdir -p $TMPMTPT && - mount $LOOPOPT -t $FSTYPE $MDSDEV $TMPMTPT && - chmod 555 $TMPMTPT/OBJECTS && - umount $TMPMTPT" || return $? +# +# Test 16 was to "verify that lustre will correct the mode of OBJECTS". +# But with new MDS stack we don't care about the mode of local objects +# anymore, so this test is removed. See bug 22944 for more details. +# - log "mount Lustre to change the mode of OBJECTS, then umount Lustre" - setup +test_17() { + setup check_mount || return 41 cleanup || return $? - log "read the mode of OBJECTS and check if they has been changed properly" - EXPECTEDOBJECTSMODE=`do_facet $SINGLEMDS "$DEBUGFS -R 'stat OBJECTS' $MDSDEV 2> /dev/null" | grep 'Mode: ' | sed -e "s/.*Mode: *//" -e "s/ *Flags:.*//"` - - if [ "$EXPECTEDOBJECTSMODE" = "0777" ]; then - log "Success:Lustre change the mode of OBJECTS correctly" - else - error "Lustre does not change mode of OBJECTS properly" + echo "Remove mds config log" + if ! combined_mgs_mds ; then + stop mgs fi -} -run_test 16 "verify that lustre will correct the mode of OBJECTS" -test_17() { - local dev=${SINGLEMDS}_dev - local MDSDEV=${!dev} - - if [ ! -e "$MDSDEV" ]; then - echo "no $MDSDEV existing, so mount Lustre to create one" - setup - check_mount || return 41 - cleanup || return $? - fi + do_facet mgs "$DEBUGFS -w -R 'unlink CONFIGS/$FSNAME-MDT0000' $MGSDEV || return \$?" || return $? - echo "Remove mds config log" - do_facet $SINGLEMDS "$DEBUGFS -w -R 'unlink CONFIGS/$FSNAME-MDT0000' $MDSDEV || return \$?" || return $? + if ! combined_mgs_mds ; then + start_mgs + fi start_ost - start_mds && return 42 - reformat_and_config + start_mds && return 42 + reformat_and_config } run_test 17 "Verify failed mds_postsetup won't fail assertion (2936) (should return errs)" test_18() { [ "$FSTYPE" != "ldiskfs" ] && skip "not needed for FSTYPE=$FSTYPE" && return - local dev=${SINGLEMDS}_dev - local MDSDEV=${!dev} + local MDSDEV=$(mdsdevname ${SINGLEMDS//mds/}) local MIN=2000000 @@ -497,17 +594,18 @@ test_18() { log "use STORED_MDSSIZE=$STORED_MDSSIZE" # check if the block device is large enough - [ -z "$OK" -a -b $MDSDEV ] && \ - [ "$(dd if=$MDSDEV of=/dev/null bs=1k count=1 skip=$MIN 2>&1 | - awk '($3 == "in") { print $1 }')" = "1+0" ] && OK=1 && \ + [ -z "$OK" ] && $(is_blkdev $SINGLEMDS $MDSDEV $MIN) && OK=1 && myMDSSIZE=$MIN && log "use device $MDSDEV with MIN=$MIN" # check if a loopback device has enough space for fs metadata (5%) - [ -z "$OK" ] && [ -f $MDSDEV -o ! -e $MDSDEV ] && - SPACE=$(df -P $(dirname $MDSDEV) | - awk '($1 != "Filesystem") {print $4}') && - [ $SPACE -gt $((MIN / 20)) ] && OK=1 && myMDSSIZE=$MIN && \ + + if [ -z "$OK" ]; then + local SPACE=$(do_facet $SINGLEMDS "[ -f $MDSDEV -o ! -e $MDSDEV ] && df -P \\\$(dirname $MDSDEV)" | + awk '($1 != "Filesystem") {print $4}') + ! [ -z "$SPACE" ] && [ $SPACE -gt $((MIN / 20)) ] && \ + OK=1 && myMDSSIZE=$MIN && \ log "use file $MDSDEV with MIN=$MIN" + fi [ -z "$OK" ] && skip_env "$MDSDEV too small for ${MIN}kB MDS" && return @@ -515,15 +613,21 @@ test_18() { echo "mount mds with large journal..." local OLD_MDS_MKFS_OPTS=$MDS_MKFS_OPTS - MDS_MKFS_OPTS="--mgs --mdt --fsname=$FSNAME --device-size=$myMDSSIZE --param sys.timeout=$TIMEOUT $MDSOPT" + local opts="--mdt --fsname=$FSNAME --device-size=$myMDSSIZE --param sys.timeout=$TIMEOUT $MDSOPT" + + if combined_mgs_mds ; then + MDS_MKFS_OPTS="--mgs $opts" + else + MDS_MKFS_OPTS="--mgsnode=$MGSNID $opts" + fi reformat_and_config echo "mount lustre system..." - setup + setup check_mount || return 41 echo "check journal size..." - local FOUNDSIZE=`do_facet mds "$DEBUGFS -c -R 'stat <8>' $MDSDEV" | awk '/Size: / { print $NF; exit;}'` + local FOUNDSIZE=$(do_facet $SINGLEMDS "$DEBUGFS -c -R 'stat <8>' $MDSDEV" | awk '/Size: / { print $NF; exit;}') if [ $FOUNDSIZE -gt $((32 * 1024 * 1024)) ]; then log "Success: mkfs creates large journals. Size: $((FOUNDSIZE >> 20))M" else @@ -601,6 +705,29 @@ test_21c() { } run_test 21c "start mds between two osts, stop mds last" +test_21d() { + if combined_mgs_mds ; then + skip "need separate mgs device" && return 0 + fi + stopall + + reformat + + start_mgs + start_ost + start_ost2 + start_mds + wait_osc_import_state mds ost2 FULL + + stop_ost + stop_ost2 + stop_mds + stop_mgs + #writeconf to remove all ost2 traces for subsequent tests + writeconf +} +run_test 21d "start mgs then ost and then mds" + test_22() { start_mds @@ -655,10 +782,12 @@ test_23a() { # was test_23 echo mount pid is ${MOUNT_PID}, mount.lustre pid is ${MOUNT_LUSTRE_PID} ps --ppid $MOUNT_PID ps --ppid $MOUNT_LUSTRE_PID - # FIXME why o why can't I kill these? Manual "ctrl-c" works... - kill -TERM $MOUNT_LUSTRE_PID echo "waiting for mount to finish" ps -ef | grep mount + # "ctrl-c" sends SIGINT but it usually (in script) does not work on child process + # SIGTERM works but it does not spread to offspring processses + kill -s TERM $MOUNT_PID + kill -s TERM $MOUNT_LUSTRE_PID # we can not wait $MOUNT_PID because it is not a child of this shell local PID1 local PID2 @@ -712,15 +841,15 @@ cleanup_24a() { } test_24a() { - #set up fs1 - gen_config - #set up fs2 - [ -n "$ost1_HOST" ] && fs2ost_HOST=$ost1_HOST + local MDSDEV=$(mdsdevname ${SINGLEMDS//mds/}) + if [ -z "$fs2ost_DEV" -o -z "$fs2mds_DEV" ]; then - do_facet $SINGLEMDS [ -b "$MDSDEV" ] && \ + is_blkdev $SINGLEMDS $MDSDEV && \ skip_env "mixed loopback and real device not working" && return fi + [ -n "$ost1_HOST" ] && fs2ost_HOST=$ost1_HOST + local fs2mdsdev=${fs2mds_DEV:-${MDSDEV}_2} local fs2ostdev=${fs2ost_DEV:-$(ostdevname 1)_2} @@ -762,8 +891,12 @@ test_24a() { run_test 24a "Multiple MDTs on a single node" test_24b() { + local MDSDEV=$(mdsdevname ${SINGLEMDS//mds/}) + if [ -z "$fs2mds_DEV" ]; then - do_facet $SINGLEMDS [ -b "$MDSDEV" ] && \ + local dev=${SINGLEMDS}_dev + local MDSDEV=${!dev} + is_blkdev $SINGLEMDS $MDSDEV && \ skip_env "mixed loopback and real device not working" && return fi @@ -793,9 +926,9 @@ test_26() { do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000135" start_mds && echo MDS started && return 1 lctl get_param -n devices - DEVS=$(lctl get_param -n devices | wc -l) + DEVS=$(lctl get_param -n devices | egrep -v MG | wc -l) [ $DEVS -gt 0 ] && return 2 - unload_modules || return 203 + unload_modules_conf || return $? } run_test 26 "MDT startup failure cleans LOV (should return errs)" @@ -885,7 +1018,9 @@ test_29() { fi # check MDT too - local MPROC="osc.$FSNAME-OST0001-osc-[M]*.active" + local mdtosc=$(get_mdtosc_proc_path $SINGLEMDS $FSNAME-OST0001) + mdtosc=${mdtosc/-MDT*/-MDT\*} + local MPROC="osc.$mdtosc.active" local MAX=30 local WAIT=0 while [ 1 ]; do @@ -904,6 +1039,9 @@ test_29() { echo "Waiting $(($MAX - $WAIT)) secs for MDT deactivated" done + # quotacheck should not fail immediately after deactivate + [ -n "$ENABLE_QUOTA" ] && { $LFS quotacheck -ug $MOUNT || error "quotacheck has failed" ; } + # test new client starts deactivated umount_client $MOUNT || return 200 mount_client $MOUNT @@ -915,6 +1053,9 @@ test_29() { echo "New client success: got $RESULT" fi + # quotacheck should not fail after umount/mount operation + [ -n "$ENABLE_QUOTA" ] && { $LFS quotacheck -ug $MOUNT || error "quotacheck has failed" ; } + # make sure it reactivates set_and_check client "lctl get_param -n $PROC_ACT" "$PARAM" $ACTV || return 6 @@ -958,18 +1099,31 @@ run_test 30a "Big config llog and conf_param deletion" test_30b() { setup - NEW="1.2.3.4@tcp" - TEST="lctl get_param -n osc.$FSNAME-OST0000-osc-????????*/import | grep failover_nids | sed -n 's/.*\($NEW\).*/\1/p'" + # Make a fake nid. Use the OST nid, and add 20 to the least significant + # numerical part of it. Hopefully that's not already a failover address for + # the server. + OSTNID=$(do_facet ost1 "$LCTL get_param nis" | tail -1 | awk '{print $1}') + ORIGVAL=$(echo $OSTNID | egrep -oi "[0-9]*@") + NEWVAL=$((($(echo $ORIGVAL | egrep -oi "[0-9]*") + 20) % 256)) + NEW=$(echo $OSTNID | sed "s/$ORIGVAL/$NEWVAL@/") + echo "Using fake nid $NEW" + + TEST="$LCTL get_param -n osc.$FSNAME-OST0000-osc-[^M]*.import | grep failover_nids | sed -n 's/.*\($NEW\).*/\1/p'" set_and_check client "$TEST" "$FSNAME-OST0000.failover.node" $NEW || error "didn't add failover nid $NEW" + NIDS=$($LCTL get_param -n osc.$FSNAME-OST0000-osc-[^M]*.import | grep failover_nids) + echo $NIDS + NIDCOUNT=$(($(echo "$NIDS" | wc -w) - 1)) + echo "should have 2 failover nids: $NIDCOUNT" + [ $NIDCOUNT -eq 2 ] || error "Failover nid not added" do_facet mgs "$LCTL conf_param -d $FSNAME-OST0000.failover.node" || error "conf_param delete failed" umount_client $MOUNT mount_client $MOUNT || return 3 - NIDS=$($LCTL get_param -n osc.$FSNAME-OST0000-osc-????????*/import | grep failover_nids) - NIDCOUNT=$(echo "$NIDS" | wc -w) - echo "only 1 final nid should remain: $NIDS" - [ $NIDCOUNT -eq 2 ] || fail "Failover nids not removed" - pass + NIDS=$($LCTL get_param -n osc.$FSNAME-OST0000-osc-[^M]*.import | grep failover_nids) + echo $NIDS + NIDCOUNT=$(($(echo "$NIDS" | wc -w) - 1)) + echo "only 1 final nid should remain: $NIDCOUNT" + [ $NIDCOUNT -eq 1 ] || error "Failover nids not removed" cleanup } @@ -1022,7 +1176,8 @@ cleanup_32() { umount -f $MOUNT || true echo "Cleanup local mds ost1 ..." cleanup_nocli32 - unload_modules + combined_mgs_mds || start_mgs + unload_modules_conf } test_32a() { @@ -1042,6 +1197,8 @@ test_32a() { $TUNEFS $tmpdir/mds || error "tunefs failed" + combined_mgs_mds || stop mgs + # nids are wrong, so client wont work, but server should start start32 mds1 $tmpdir/mds "-o loop,exclude=lustre-OST0000" && \ trap cleanup_32 EXIT INT || return 3 @@ -1078,6 +1235,9 @@ test_32a() { # mount a second time to make sure we didnt leave upgrade flag on load_modules $TUNEFS --dryrun $tmpdir/mds || error "tunefs failed" + + combined_mgs_mds || stop mgs + start32 mds1 $tmpdir/mds "-o loop,exclude=lustre-OST0000" && \ trap cleanup_32 EXIT INT || return 12 @@ -1105,6 +1265,8 @@ test_32b() { # writeconf will cause servers to register with their current nids $TUNEFS --writeconf --fsname=$NEWNAME $tmpdir/mds || error "tunefs failed" + combined_mgs_mds || stop mgs + start32 mds1 $tmpdir/mds "-o loop" && \ trap cleanup_32 EXIT INT || return 3 @@ -1161,13 +1323,19 @@ run_test 32b "Upgrade from 1.8 with writeconf" test_33a() { # bug 12333, was test_33 local rc=0 local FSNAME2=test-123 + local MDSDEV=$(mdsdevname ${SINGLEMDS//mds/}) + [ -n "$ost1_HOST" ] && fs2ost_HOST=$ost1_HOST if [ -z "$fs2ost_DEV" -o -z "$fs2mds_DEV" ]; then - do_facet $SINGLEMDS [ -b "$MDSDEV" ] && \ + local dev=${SINGLEMDS}_dev + local MDSDEV=${!dev} + is_blkdev $SINGLEMDS $MDSDEV && \ skip_env "mixed loopback and real device not working" && return fi + combined_mgs_mds || mkfs_opts="$mkfs_opts --nomgs" + local fs2mdsdev=${fs2mds_DEV:-${MDSDEV}_2} local fs2ostdev=${fs2ost_DEV:-$(ostdevname 1)_2} add fs2mds $MDS_MKFS_OPTS --mkfsoptions='\"-J size=8\"' --fsname=${FSNAME2} --reformat $fs2mdsdev || exit 10 @@ -1296,6 +1464,8 @@ test_35a() { # bug 12459 }" $TMP/lustre-log-$TESTNAME.log` [ "$NEXTCONN" != "0" ] && log "The client didn't try to reconnect to the last active server (tried ${NEXTCONN} instead)" && return 7 cleanup + # remove nid settings + writeconf } run_test 35a "Reconnect to the last active server first" @@ -1311,9 +1481,9 @@ test_35b() { # bug 18674 log "Set up a fake failnode for the MDS" FAKENID="127.0.0.2" - local device=$(do_facet mds "$LCTL get_param -n devices" | \ + local device=$(do_facet $SINGLEMDS "$LCTL get_param -n devices" | \ awk '($3 ~ "mdt" && $4 ~ "MDT") { print $4 }' | head -1) - do_facet mds "$LCTL conf_param ${device}.failover.node=$FAKENID" || \ + do_facet $SINGLEMDS "$LCTL conf_param ${device}.failover.node=$FAKENID" || \ return 1 local at_max_saved=0 @@ -1323,19 +1493,21 @@ test_35b() { # bug 18674 at_max_set 0 mds client fi - mkdir -p $MOUNT/testdir - touch $MOUNT/testdir/test + mkdir -p $MOUNT/$tdir log "Injecting EBUSY on MDS" # Setting OBD_FAIL_MDS_RESEND=0x136 - do_facet mds "$LCTL set_param fail_loc=0x80000136" || return 2 + do_facet $SINGLEMDS "$LCTL set_param fail_loc=0x80000136" || return 2 - log "Stat on a test file" - stat $MOUNT/testdir/test + $LCTL set_param mdc.${FSNAME}*.stats=clear + + log "Creating a test file and stat it" + touch $MOUNT/$tdir/$tfile + stat $MOUNT/$tdir/$tfile log "Stop injecting EBUSY on MDS" - do_facet mds "$LCTL set_param fail_loc=0" || return 3 - rm -f $MOUNT/testdir/test + do_facet $SINGLEMDS "$LCTL set_param fail_loc=0" || return 3 + rm -f $MOUNT/$tdir/$tfile log "done" # restore adaptive timeout @@ -1343,6 +1515,8 @@ test_35b() { # bug 18674 $LCTL dk $TMP/lustre-log-$TESTNAME.log + CONNCNT=`$LCTL get_param mdc.${FSNAME}*.stats | awk '/mds_connect/{print $2}'` + # retrieve from the log if the client has ever tried to # contact the fake server after the loss of connection FAILCONN=`awk "BEGIN {ret = 0;} @@ -1362,26 +1536,36 @@ test_35b() { # bug 18674 log "ERROR: The client tried to reconnect to the failover server while the primary was busy" && \ return 5 - cleanup + # LU-290 + # When OBD_FAIL_MDS_RESEND is hit, we sleep for 2 * obd_timeout + # Reconnects are supposed to be rate limited to one every 5s + [ $CONNCNT -gt $((2 * $TIMEOUT / 5 + 1)) ] && \ + log "ERROR: Too many reconnects $CONNCNT" && \ + return 6 + + cleanup + # remove nid settings + writeconf } run_test 35b "Continue reconnection retries, if the active server is busy" test_36() { # 12743 - local rc + [ $OSTCOUNT -lt 2 ] && skip_env "skipping test for single OST" && return + + [ "$ost_HOST" = "`hostname`" -o "$ost1_HOST" = "`hostname`" ] || \ + { skip "remote OST" && return 0; } + + local rc=0 local FSNAME2=test1234 local fs3ost_HOST=$ost_HOST + local MDSDEV=$(mdsdevname ${SINGLEMDS//mds/}) [ -n "$ost1_HOST" ] && fs2ost_HOST=$ost1_HOST && fs3ost_HOST=$ost1_HOST - rc=0 if [ -z "$fs2ost_DEV" -o -z "$fs2mds_DEV" -o -z "$fs3ost_DEV" ]; then - do_facet $SINGLEMDS [ -b "$MDSDEV" ] && \ + is_blkdev $SINGLEMDS $MDSDEV && \ skip_env "mixed loopback and real device not working" && return fi - [ $OSTCOUNT -lt 2 ] && skip_env "skipping test for single OST" && return - - [ "$ost_HOST" = "`hostname`" -o "$ost1_HOST" = "`hostname`" ] || \ - { skip "remote OST" && return 0; } local fs2mdsdev=${fs2mds_DEV:-${MDSDEV}_2} local fs2ostdev=${fs2ost_DEV:-$(ostdevname 1)_2} @@ -1434,34 +1618,38 @@ test_36() { # 12743 stop fs2ost -f || return 201 stop fs2mds -f || return 202 rm -rf $MOUNT2 $fs2mdsdev $fs2ostdev $fs3ostdev - unload_modules || return 203 + unload_modules_conf || return 203 return $rc } run_test 36 "df report consistency on OSTs with different block size" test_37() { - client_only && skip "client only testing" && return 0 - LOCAL_MDSDEV="$TMP/mdt.img" - SYM_MDSDEV="$TMP/sym_mdt.img" + local mntpt=$(facet_mntpt $SINGLEMDS) + local mdsdev=$(mdsdevname ${SINGLEMDS//mds/}) + local mdsdev_sym="$TMP/sym_mdt.img" - echo "MDS : $LOCAL_MDSDEV" - echo "SYMLINK : $SYM_MDSDEV" - rm -f $LOCAL_MDSDEV + echo "MDS : $mdsdev" + echo "SYMLINK : $mdsdev_sym" + do_facet $SINGLEMDS rm -f $mdsdev_sym - touch $LOCAL_MDSDEV - mkfs.lustre --reformat --fsname=lustre --mdt --mgs --device-size=9000 $LOCAL_MDSDEV || - error "mkfs.lustre $LOCAL_MDSDEV failed" - ln -s $LOCAL_MDSDEV $SYM_MDSDEV + do_facet $SINGLEMDS ln -s $mdsdev $mdsdev_sym - echo "mount symlink device - $SYM_MDSDEV" + echo "mount symlink device - $mdsdev_sym" + + local rc=0 + mount_op=$(do_facet $SINGLEMDS mount -v -t lustre $MDS_MOUNT_OPTS $mdsdev_sym $mntpt 2>&1 ) + rc=${PIPESTATUS[0]} - mount_op=`mount -v -t lustre -o loop $SYM_MDSDEV ${MOUNT%/*}/mds 2>&1 | grep "unable to set tunable"` - umount -d ${MOUNT%/*}/mds - rm -f $LOCAL_MDSDEV $SYM_MDSDEV + echo mount_op=$mount_op - if [ -n "$mount_op" ]; then - error "**** FAIL: set tunables failed for symlink device" + do_facet $SINGLEMDS "umount -d $mntpt && rm -f $mdsdev_sym" + + if $(echo $mount_op | grep -q "unable to set tunable"); then + error "set tunables failed for symlink device" fi + + [ $rc -eq 0 ] || error "mount symlink $mdsdev_sym failed! rc=$rc" + return 0 } run_test 37 "verify set tunables works for symlink device" @@ -1482,8 +1670,7 @@ test_38() { # bug 14222 log "rename lov_objid file on MDS" rm -f $TMP/lov_objid.orig - local dev=${SINGLEMDS}_dev - local MDSDEV=${!dev} + local MDSDEV=$(mdsdevname ${SINGLEMDS//mds/}) do_facet $SINGLEMDS "$DEBUGFS -c -R \\\"dump lov_objid $TMP/lov_objid.orig\\\" $MDSDEV" do_facet $SINGLEMDS "$DEBUGFS -w -R \\\"rm lov_objid\\\" $MDSDEV" @@ -1536,16 +1723,17 @@ run_test 39 "leak_finder recognizes both LUSTRE and LNET malloc messages" test_40() { # bug 15759 start_ost #define OBD_FAIL_TGT_TOOMANY_THREADS 0x706 - do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000706" + do_facet $SINGLEMDS "$LCTL set_param fail_loc=0x80000706" start_mds cleanup } run_test 40 "race during service thread startup" -test_41() { #bug 14134 +test_41a() { #bug 14134 + echo $MDS_MOUNT_OPTS | grep "loop" && skip " loop devices does not work with nosvc option" && return + local rc - local dev=${SINGLEMDS}_dev - local MDSDEV=${!dev} + local MDSDEV=$(mdsdevname ${SINGLEMDS//mds/}) start $SINGLEMDS $MDSDEV $MDS_MOUNT_OPTS -o nosvc -n start ost1 `ostdevname 1` $OST_MOUNT_OPTS @@ -1561,11 +1749,36 @@ test_41() { #bug 14134 stop ost1 -f || return 201 stop_mds -f || return 202 stop_mds -f || return 203 - unload_modules || return 204 + unload_modules_conf || return 204 return $rc } -run_test 41 "mount mds with --nosvc and --nomgs" +run_test 41a "mount mds with --nosvc and --nomgs" + +test_41b() { + echo $MDS_MOUNT_OPTS | grep "loop" && skip " loop devices does not work with nosvc option" && return + + stopall + reformat + local MDSDEV=$(mdsdevname ${SINGLEMDS//mds/}) + + start $SINGLEMDS $MDSDEV $MDS_MOUNT_OPTS -o nosvc -n + start_ost + start $SINGLEMDS $MDSDEV $MDS_MOUNT_OPTS -o nomgs,force + mkdir -p $MOUNT + mount_client $MOUNT || return 1 + sleep 5 + + echo "blah blah" > $MOUNT/$tfile + cat $MOUNT/$tfile || return 200 + + umount_client $MOUNT + stop_ost || return 201 + stop_mds -f || return 202 + stop_mds -f || return 203 + +} +run_test 41b "mount mds with --nosvc and --nomgs on first mount" test_42() { #bug 14693 setup check_mount || return 2 @@ -1686,7 +1899,7 @@ test_44() { # 16317 check_mount || return 2 UUID=$($LCTL get_param llite.${FSNAME}*.uuid | cut -d= -f2) STATS_FOUND=no - UUIDS=$(do_facet mds "$LCTL get_param mdt.${FSNAME}*.exports.*.uuid") + UUIDS=$(do_facet $SINGLEMDS "$LCTL get_param mdt.${FSNAME}*.exports.*.uuid") for VAL in $UUIDS; do NID=$(echo $VAL | cut -d= -f1) CLUUID=$(echo $VAL | cut -d= -f2) @@ -1730,10 +1943,9 @@ cleanup_46a() { let count=count-1 done stop_mds || rc=$? - # writeconf is needed after the test, otherwise, - # we might end up with extra OSTs - writeconf || rc=$? cleanup_nocli || rc=$? + #writeconf to remove all ost2 traces for subsequent tests + writeconf return $rc } @@ -1866,7 +2078,7 @@ test_49() { # bug 17710 check_mount || return 1 echo "check ldlm_timout..." - LDLM_MDS="`do_facet mds lctl get_param -n ldlm_timeout`" + LDLM_MDS="`do_facet $SINGLEMDS lctl get_param -n ldlm_timeout`" LDLM_OST1="`do_facet ost1 lctl get_param -n ldlm_timeout`" LDLM_CLIENT="`do_facet client lctl get_param -n ldlm_timeout`" @@ -1888,7 +2100,7 @@ test_49() { # bug 17710 setup_noconfig check_mount || return 7 - LDLM_MDS="`do_facet mds lctl get_param -n ldlm_timeout`" + LDLM_MDS="`do_facet $SINGLEMDS lctl get_param -n ldlm_timeout`" LDLM_OST1="`do_facet ost1 lctl get_param -n ldlm_timeout`" LDLM_CLIENT="`do_facet client lctl get_param -n ldlm_timeout`" @@ -1974,6 +2186,8 @@ test_50c() { umount_client $MOUNT || error "Unable to unmount client" stop_ost2 || error "Unable to stop OST2" stop_mds || error "Unable to stop MDS" + #writeconf to remove all ost2 traces for subsequent tests + writeconf } run_test 50c "lazystatfs one server down ==========================" @@ -1994,6 +2208,8 @@ test_50d() { umount_client $MOUNT || error "Unable to unmount client" stop_ost2 || error "Unable to stop OST2" stop_mds || error "Unable to stop MDS" + #writeconf to remove all ost2 traces for subsequent tests + writeconf } run_test 50d "lazystatfs client/server conn race ==========================" @@ -2074,6 +2290,7 @@ test_50f() { umount_client $MOUNT || error "Unable to unmount client" stop_ost || error "Unable to stop OST1" stop_mds || error "Unable to stop MDS" + #writeconf to remove all ost2 traces for subsequent tests writeconf } run_test 50f "normal statfs one server in down ==========================" @@ -2082,6 +2299,8 @@ test_50g() { [ "$OSTCOUNT" -lt "2" ] && skip_env "$OSTCOUNT < 2, skipping" && return setup start_ost2 || error "Unable to start OST2" + wait_osc_import_state mds ost2 FULL + wait_osc_import_state client ost2 FULL local PARAM="${FSNAME}-OST0001.osc.active" @@ -2099,6 +2318,7 @@ test_50g() { stop_ost2 || error "Unable to stop OST2" stop_ost || error "Unable to stop OST1" stop_mds || error "Unable to stop MDS" + #writeconf to remove all ost2 traces for subsequent tests writeconf } run_test 50g "deactivated OST should not cause panic=====================" @@ -2121,6 +2341,8 @@ test_51() { wait $pid stop_ost2 || return 3 cleanup + #writeconf to remove all ost2 traces for subsequent tests + writeconf } run_test 51 "Verify that mdt_reint handles RMF_MDT_MD correctly when an OST is added" @@ -2189,7 +2411,9 @@ test_52() { [ $? -eq 0 ] || { error "Unable to create tdir"; return 4; } touch $TMP/modified_first [ $? -eq 0 ] || { error "Unable to create temporary file"; return 5; } - do_node $ost1node "mkdir -p $ost1tmp && touch $ost1tmp/modified_first" + local mtime=$(stat -c %Y $TMP/modified_first) + do_node $ost1node "mkdir -p $ost1tmp && touch -m -d @$mtime $ost1tmp/modified_first" + [ $? -eq 0 ] || { error "Unable to create temporary file"; return 6; } sleep 1 @@ -2293,17 +2517,17 @@ thread_sanity() { tmax=$(do_facet $facet "lctl get_param -n ${paramp}.threads_max" || echo 0) tstarted=$(do_facet $facet "lctl get_param -n ${paramp}.threads_started" || echo 0) lassert 23 "$msg (PDSH problems?)" '(($tstarted && $tmin && $tmax))' || return $? - lassert 24 "$msg" '(($tstarted >= $tmin && $tstarted <= tmax ))' || return $? + lassert 24 "$msg" '(($tstarted >= $tmin && $tstarted <= $tmax ))' || return $? - # Check that we can lower min/max - do_facet $facet "lctl set_param ${paramp}.threads_min=$((tmin - 1))" - do_facet $facet "lctl set_param ${paramp}.threads_max=$((tmax - 10))" + # Check that we can change min/max + do_facet $facet "lctl set_param ${paramp}.threads_min=$((tmin + 1))" + do_facet $facet "lctl set_param ${paramp}.threads_max=$((tmax - 1))" tmin2=$(do_facet $facet "lctl get_param -n ${paramp}.threads_min" || echo 0) tmax2=$(do_facet $facet "lctl get_param -n ${paramp}.threads_max" || echo 0) - lassert 25 "$msg" '(($tmin2 == ($tmin - 1) && $tmax2 == ($tmax -10)))' || return $? + lassert 25 "$msg" '(($tmin2 == ($tmin + 1) && $tmax2 == ($tmax -1)))' || return $? # Check that we can set min/max to the same value - do_facet $facet "lctl set_param ${paramp}.threads_min=$tmin" + tmin=$(do_facet $facet "lctl get_param -n ${paramp}.threads_min" || echo 0) do_facet $facet "lctl set_param ${paramp}.threads_max=$tmin" tmin2=$(do_facet $facet "lctl get_param -n ${paramp}.threads_min" || echo 0) tmax2=$(do_facet $facet "lctl get_param -n ${paramp}.threads_max" || echo 0) @@ -2313,7 +2537,7 @@ thread_sanity() { do_facet $facet "lctl set_param ${paramp}.threads_max=$((tmin - 1))" tmin2=$(do_facet $facet "lctl get_param -n ${paramp}.threads_min" || echo 0) tmax2=$(do_facet $facet "lctl get_param -n ${paramp}.threads_max" || echo 0) - lassert 27 "$msg" '(($tmin <= $tmax2))' || return $? + lassert 27 "$msg" '(($tmin2 <= $tmax2))' || return $? # We need to ensure that we get the module options desired; to do this # we set LOAD_MODULES_REMOTE=true and we call setmodopts below. @@ -2354,10 +2578,6 @@ test_53b() { } run_test 53b "check MDT thread count params" -if ! combined_mgs_mds ; then - stop mgs -fi - run_llverfs() { local dir=$1 @@ -2433,11 +2653,112 @@ test_56() { mount_client $MOUNT || error "Unable to mount client" echo ok $LFS osts + [ -n "$ENABLE_QUOTA" ] && { $LFS quotacheck -ug $MOUNT || error "quotacheck has failed" ; } stopall reformat } run_test 56 "check big indexes" +test_57a() { # bug 22656 + local NID=$(do_facet ost1 "$LCTL get_param nis" | tail -1 | awk '{print $1}') + writeconf + do_facet ost1 "$TUNEFS --failnode=$NID `ostdevname 1`" || error "tunefs failed" + start_mgsmds + start_ost && error "OST registration from failnode should fail" + reformat +} +run_test 57a "initial registration from failnode should fail (should return errs)" + +test_57b() { + local NID=$(do_facet ost1 "$LCTL get_param nis" | tail -1 | awk '{print $1}') + writeconf + do_facet ost1 "$TUNEFS --servicenode=$NID `ostdevname 1`" || error "tunefs failed" + start_mgsmds + start_ost || error "OST registration from servicenode should not fail" + reformat +} +run_test 57b "initial registration from servicenode should not fail" + +count_osts() { + do_facet mgs $LCTL get_param mgs.MGS.live.$FSNAME | grep OST | wc -l +} + +test_58() { # bug 22658 + [ "$FSTYPE" != "ldiskfs" ] && skip "not supported for $FSTYPE" && return + setup + mkdir -p $DIR/$tdir + createmany -o $DIR/$tdir/$tfile-%d 100 + # make sure that OSTs do not cancel llog cookies before we unmount the MDS +#define OBD_FAIL_OBD_LOG_CANCEL_NET 0x601 + do_facet mds "lctl set_param fail_loc=0x601" + unlinkmany $DIR/$tdir/$tfile-%d 100 + stop mds + local MNTDIR=$(facet_mntpt mds) + # remove all files from the OBJECTS dir + do_facet mds "mount -t ldiskfs $MDSDEV $MNTDIR" + do_facet mds "find $MNTDIR/OBJECTS -type f -delete" + do_facet mds "umount $MNTDIR" + # restart MDS with missing llog files + start_mds + do_facet mds "lctl set_param fail_loc=0" + reformat +} +run_test 58 "missing llog files must not prevent MDT from mounting" + +test_59() { + start_mgsmds >> /dev/null + local C1=$(count_osts) + if [ $C1 -eq 0 ]; then + start_ost >> /dev/null + C1=$(count_osts) + fi + stopall + echo "original ost count: $C1 (expect > 0)" + [ $C1 -gt 0 ] || error "No OSTs in $FSNAME log" + start_mgsmds -o writeconf >> /dev/null || error "MDT start failed" + local C2=$(count_osts) + echo "after mdt writeconf count: $C2 (expect 0)" + [ $C2 -gt 0 ] && error "MDT writeconf should erase OST logs" + echo "OST start without writeconf should fail:" + start_ost >> /dev/null && error "OST start without writeconf didn't fail" + echo "OST start with writeconf should succeed:" + start_ost -o writeconf >> /dev/null || error "OST1 start failed" + local C3=$(count_osts) + echo "after ost writeconf count: $C3 (expect 1)" + [ $C3 -eq 1 ] || error "new OST writeconf should add:" + start_ost2 -o writeconf >> /dev/null || error "OST2 start failed" + local C4=$(count_osts) + echo "after ost2 writeconf count: $C4 (expect 2)" + [ $C4 -eq 2 ] || error "OST2 writeconf should add log" + stop_ost2 >> /dev/null + cleanup_nocli >> /dev/null + #writeconf to remove all ost2 traces for subsequent tests + writeconf +} +run_test 59 "writeconf mount option" + +test_60() { # LU-471 + add mds1 $MDS_MKFS_OPTS --mkfsoptions='\" -E stride=64 -O ^uninit_bg\"' --reformat $(mdsdevname 1) + + dump=$(do_facet $SINGLEMDS dumpe2fs $(mdsdevname 1)) + rc=${PIPESTATUS[0]} + [ $rc -eq 0 ] || error "dumpe2fs $(mdsdevname 1) failed" + + # MDT default has dirdata feature + echo $dump | grep dirdata > /dev/null || error "dirdata is not set" + # we disable uninit_bg feature + echo $dump | grep uninit_bg > /dev/null && error "uninit_bg is set" + # we set stride extended options + echo $dump | grep stride > /dev/null || error "stride is not set" + reformat +} +run_test 60 "check mkfs.lustre --mkfsoptions -E -O options setting" + +if ! combined_mgs_mds ; then + stop mgs +fi + cleanup_gss -equals_msg `basename $0`: test complete -[ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG && grep -q FAIL $TESTSUITELOG && exit 1 || true + +complete $(basename $0) $SECONDS +exit_status