X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Ftests%2Fsanity-hsm.sh;h=0ee67b103990e87f6b8eafec33406a983b066a4d;hp=0e4735f9b6e8967379e31e2cc627cff584209461;hb=f625f670afbe954030ff81f0f8522137d6cdd335;hpb=f0f0a4951493859d213ae41b7818722e7c37e58a diff --git a/lustre/tests/sanity-hsm.sh b/lustre/tests/sanity-hsm.sh index 0e4735f..0ee67b1 100755 --- a/lustre/tests/sanity-hsm.sh +++ b/lustre/tests/sanity-hsm.sh @@ -11,12 +11,8 @@ SRCDIR=$(dirname $0) export PATH=$PWD/$SRCDIR:$SRCDIR:$PWD/$SRCDIR/utils:$PATH:/sbin:/usr/sbin ONLY=${ONLY:-"$*"} -# bug number for skipped test: 3815 +# bug number for skipped test: LU-3815 ALWAYS_EXCEPT="$SANITY_HSM_EXCEPT 34 35 36" -# bug number for skipped test:4178 4176 -ALWAYS_EXCEPT="$ALWAYS_EXCEPT 200 221 223b 31a" -# bug number for skipped test:LU-3852 -ALWAYS_EXCEPT="$ALWAYS_EXCEPT 251" # UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT! LUSTRE=${LUSTRE:-$(cd $(dirname $0)/..; echo $PWD)} @@ -28,7 +24,6 @@ init_logging MULTIOP=${MULTIOP:-multiop} OPENFILE=${OPENFILE:-openfile} -MCREATE=${MCREATE:-mcreate} MOUNT_2=${MOUNT_2:-"yes"} FAIL_ON_ERROR=false @@ -105,6 +100,9 @@ init_agt_vars() { # archive is purged at copytool setup HSM_ARCHIVE_PURGE=true + + # Don't allow copytool error upon start/setup + HSMTOOL_NOERROR=false } # Get the backend root path for the given agent facet. @@ -212,7 +210,7 @@ copytool_setup() { if [[ -z "$arc_id" ]] && do_facet $facet "pkill -CONT -x $HSMTOOL_BASE"; then - echo "Wakeup copytool $facet on $agent" + echo "Only wakeup running copytool $facet on $agent" return 0 fi @@ -239,8 +237,13 @@ copytool_setup() { [[ -z "$TESTNAME" ]] || prefix=$prefix.$TESTNAME local copytool_log=$prefix.copytool${arc_id}_log.$agent.log - do_facet $facet "$cmd < /dev/null > $copytool_log 2>&1" || - error "start copytool $facet on $agent failed" + do_facet $facet "$cmd < /dev/null > $copytool_log 2>&1" + if [[ $? != 0 ]]; then + [[ $HSMTOOL_NOERROR == true ]] || + error "start copytool $facet on $agent failed" + echo "start copytool $facet on $agent failed" + fi + trap cleanup EXIT } @@ -257,15 +260,32 @@ get_copytool_event_log() { copytool_cleanup() { trap - EXIT - local agents=${1:-$(facet_active_host $SINGLEAGT)} + local facet=$SINGLEAGT + local agents=${1:-$(facet_active_host $facet)} local mdtno local idx local oldstate local mdt_hsmctrl + local hsm_root=$(copytool_device $facet) + local end_wait=$(( SECONDS + TIMEOUT )) do_nodesv $agents "pkill -INT -x $HSMTOOL_BASE" || return 0 - sleep 1 - echo "Copytool is stopped on $agents" + + while (( SECONDS < end_wait )); do + sleep 2 + do_nodesv $agents "pgrep -x $HSMTOOL_BASE" + if [ $? -ne 0 ]; then + echo "Copytool is stopped on $agents" + break + fi + echo "Copytool still running on $agents" + done + if do_nodesv $agents "pgrep -x $HSMTOOL_BASE"; then + error "Copytool failed to stop in ${TIMEOUT}s ..." + else + echo "Copytool has stopped in " \ + "$((TIMEOUT - (end_wait - SECONDS)))s." + fi # clean all CDTs orphans requests from previous tests # that would otherwise need to timeout to clear. @@ -286,6 +306,9 @@ copytool_cleanup() { "$oldstate" 20 || error "mds${mdtno} cdt state is not $oldstate" done + if do_facet $facet "df $hsm_root" >/dev/null 2>&1 ; then + do_facet $facet "rm -rf $hsm_root/*" + fi } copytool_suspend() { @@ -313,7 +336,7 @@ make_archive() { local file=$HSM_ARCHIVE/$1 do_facet $SINGLEAGT mkdir -p $(dirname $file) do_facet $SINGLEAGT dd if=/dev/urandom of=$file count=32 bs=1000000 || - error "cannot create $file" + file_creation_failure dd $file $? } copy2archive() { @@ -488,17 +511,19 @@ needclients() { path2fid() { $LFS path2fid $1 | tr -d '[]' + return ${PIPESTATUS[0]} } get_hsm_flags() { local f=$1 local u=$2 + local st if [[ $u == "user" ]]; then - local st=$($RUNAS $LFS hsm_state $f) + st=$($RUNAS $LFS hsm_state $f) else - local st=$($LFS hsm_state $f) u=root + st=$($LFS hsm_state $f) fi [[ $? == 0 ]] || error "$LFS hsm_state $f failed (run as $u)" @@ -509,7 +534,8 @@ get_hsm_flags() { get_hsm_archive_id() { local f=$1 - local st=$($LFS hsm_state $f) + local st + st=$($LFS hsm_state $f) [[ $? == 0 ]] || error "$LFS hsm_state $f failed" local ar=$(echo $st | grep "archive_id" | cut -f5 -d" " | @@ -533,6 +559,15 @@ check_hsm_flags_user() { [[ $st == $fl ]] || error "hsm flags on $f are $st != $fl" } +file_creation_failure() { + local cmd=$1 + local f=$2 + local err=$3 + + df $MOUNT $MOUNT2 >&2 + error "cannot create $f with $cmd, status=$err" +} + copy_file() { local f= @@ -546,20 +581,22 @@ copy_file() { f=${f/$DIR/$DIR2} fi rm -f $f - cp $1 $f || error "cannot copy $1 to $f" + cp $1 $f || file_creation_failure cp $f $? + path2fid $f || error "cannot get fid on $f" } make_small() { local file2=${1/$DIR/$DIR2} dd if=/dev/urandom of=$file2 count=2 bs=1M conv=fsync || - error "cannot create $file2" + file_creation_failure dd $file2 $? + path2fid $1 || error "cannot get fid on $1" } make_small_sync() { dd if=/dev/urandom of=$1 count=1 bs=1M conv=sync || - error "cannot create $1" + file_creation_failure dd $1 $? path2fid $1 || error "cannot get fid on $1" } @@ -569,14 +606,27 @@ cleanup_large_files() { [ $ratio -gt 50 ] && find $MOUNT -size +10M -exec rm -f {} \; } +check_enough_free_space() { + local nb=$1 + local unit=$2 + local need=$((nb * unit /1024)) + local free=$(df -kP $MOUNT | tail -1 | awk '{print $4}') + (( $need >= $free )) && return 1 + return 0 +} + make_large_for_striping() { local file2=${1/$DIR/$DIR2} local sz=$($LCTL get_param -n lov.*-clilov-*.stripesize | head -n1) cleanup_large_files + check_enough_free_space 5 $sz + [ $? != 0 ] && return $? + dd if=/dev/urandom of=$file2 count=5 bs=$sz conv=fsync || - error "cannot create $file2" + file_creation_failure dd $file2 $? + path2fid $1 || error "cannot get fid on $1" } @@ -585,12 +635,16 @@ make_large_for_progress() { cleanup_large_files + check_enough_free_space 39 1000000 + [ $? != 0 ] && return $? + # big file is large enough, so copy time is > 30s # so copytool make 1 progress # size is not a multiple of 1M to avoid stripe # aligment dd if=/dev/urandom of=$file2 count=39 bs=1000000 conv=fsync || - error "cannot create $file2" + file_creation_failure dd $file2 $? + path2fid $1 || error "cannot get fid on $1" } @@ -599,12 +653,15 @@ make_large_for_progress_aligned() { cleanup_large_files + check_enough_free_space 33 1048576 + [ $? != 0 ] && return $? + # big file is large enough, so copy time is > 30s # so copytool make 1 progress # size is a multiple of 1M to have stripe # aligment dd if=/dev/urandom of=$file2 count=33 bs=1M conv=fsync || - error "cannot create $file2" + file_creation_failure dd $file2 $? path2fid $1 || error "cannot get fid on $1" } @@ -613,9 +670,12 @@ make_large_for_cancel() { cleanup_large_files + check_enough_free_space 103 1048576 + [ $? != 0 ] && return $? + # Copy timeout is 100s. 105MB => 105s dd if=/dev/urandom of=$file2 count=103 bs=1M conv=fsync || - error "cannot create $file2" + file_creation_failure dd $file2 $? path2fid $1 || error "cannot get fid on $1" } @@ -636,7 +696,7 @@ wait_request_state() { local cmd="$LCTL get_param -n ${MDT_PREFIX}${mdtidx}.hsm.actions" cmd+=" | awk '/'$fid'.*action='$request'/ {print \\\$13}' | cut -f2 -d=" - wait_result $mds "$cmd" $state 100 || + wait_result $mds "$cmd" $state 200 || error "request on $fid is not $state on $mds" } @@ -800,7 +860,7 @@ test_3() { error "user could not change hsm flags" dd if=/etc/passwd of=$f.append bs=1 count=3\ conv=notrunc oflag=append status=noxfer || - error "could not append to test file" + file_creation_failure dd $f.append $? check_hsm_flags $f.append "0x00000003" # Modify a file sets it dirty @@ -809,7 +869,7 @@ test_3() { error "user could not change hsm flags" dd if=/dev/zero of=$f.modify bs=1 count=3\ conv=notrunc status=noxfer || - error "could not modify test file" + file_creation_failure dd $f.modify $? check_hsm_flags $f.modify "0x00000003" # Open O_TRUNC sets dirty @@ -870,7 +930,7 @@ test_9() { copytool_cleanup } -run_test 9 "Use of explict archive number, with dedicated copytool" +run_test 9 "Use of explicit archive number, with dedicated copytool" test_9a() { needclients 3 || return 0 @@ -983,7 +1043,7 @@ test_10d() { } run_test 10d "Archive a file on the default archive id" -test_11() { +test_11a() { mkdir -p $DIR/$tdir copy2archive /etc/hosts $tdir/$tfile local f=$DIR/$tdir/$tfile @@ -1007,7 +1067,31 @@ test_11() { local AFILE=$(do_facet $SINGLEAGT ls $HSM_ARCHIVE'/*/*/*/*/*/*/'$fid) || error "fid $fid not in archive $HSM_ARCHIVE" } -run_test 11 "Import a file" +run_test 11a "Import a file" + +test_11b() { + # test needs a running copytool + copytool_setup + + mkdir -p $DIR/$tdir + local f=$DIR/$tdir/$tfile + local fid=$(copy_file /etc/hosts $f) + $LFS hsm_archive -a $HSM_ARCHIVE_NUMBER $f || + error "hsm_archive failed" + wait_request_state $fid ARCHIVE SUCCEED + + local FILE_HASH=$(md5sum $f) + rm -f $f + + import_file $fid $f + + echo "$FILE_HASH" | md5sum -c + + [[ $? -eq 0 ]] || error "Restored file differs" + + copytool_cleanup +} +run_test 11b "Import a deleted file using its FID" test_12a() { # test needs a running copytool @@ -1071,7 +1155,10 @@ test_12c() { mkdir -p $DIR/$tdir local f=$DIR/$tdir/$tfile $LFS setstripe -c 2 $f - local fid=$(make_large_for_striping $f) + local fid + fid=$(make_large_for_striping $f) + [ $? != 0 ] && skip "not enough free space" && return + local FILE_CRC=$(md5sum $f) $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f @@ -1298,6 +1385,26 @@ test_12o() { } run_test 12o "Layout-swap failure during Restore leaves file released" +test_12p() { + # test needs a running copytool + copytool_setup + + mkdir $DIR/$tdir + local f=$DIR/$tdir/$tfile + local fid=$(copy_file /etc/hosts $f) + + $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f + wait_request_state $fid ARCHIVE SUCCEED + do_facet $SINGLEAGT cat $f > /dev/null || error "cannot cat $f" + $LFS hsm_release $f || error "cannot release $f" + do_facet $SINGLEAGT cat $f > /dev/null || error "cannot cat $f" + $LFS hsm_release $f || error "cannot release $f" + do_facet $SINGLEAGT cat $f > /dev/null || error "cannot cat $f" + + copytool_cleanup +} +run_test 12p "implicit restore of a file on copytool mount point" + test_13() { # test needs a running copytool copytool_setup @@ -1727,7 +1834,7 @@ test_24b() { copytool_setup mkdir -p $DIR/$tdir - # Check that root can do HSM actions on a ordinary user's file. + # Check that root can do HSM actions on a regular user's file. rm -f $file fid=$(make_small $file) sum0=$(md5sum $file) @@ -1946,7 +2053,10 @@ test_26() { mkdir -p $DIR/$tdir local f=$DIR/$tdir/$tfile - local fid=$(make_large_for_progress $f) + local fid + fid=$(make_large_for_progress $f) + [ $? != 0 ] && skip "not enough free space" && return + $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f wait_request_state $fid ARCHIVE SUCCEED @@ -1983,7 +2093,10 @@ test_27b() { mkdir -p $DIR/$tdir local f=$DIR/$tdir/$tfile - local fid=$(make_large_for_progress $f) + local fid + fid=$(make_large_for_progress $f) + [ $? != 0 ] && skip "not enough free space" && return + $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f wait_request_state $fid ARCHIVE SUCCEED $LFS hsm_release $f @@ -2002,7 +2115,10 @@ test_28() { mkdir -p $DIR/$tdir local f=$DIR/$tdir/$tfile - local fid=$(make_large_for_progress $f) + local fid + fid=$(make_large_for_progress $f) + [ $? != 0 ] && skip "not enough free space" && return + $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f wait_request_state $fid ARCHIVE SUCCEED @@ -2091,6 +2207,7 @@ test_30c() { mkdir -p $DIR/$tdir local f=$DIR/$tdir/SLEEP + local slp_sum1=$(md5sum /bin/sleep) local fid=$(copy_file /bin/sleep $f) chmod 755 $f $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f @@ -2107,7 +2224,12 @@ test_30c() { wait $pid [[ $? == 0 ]] || error "Execution failed during run" cmp /bin/sleep $f - [[ $? == 0 ]] || error "Binary overwritten during exec" + if [[ $? != 0 ]]; then + local slp_sum2=$(md5sum /bin/sleep) + # in case sleep file is modified during the test + [[ $slp_sum1 == $slp_sum2 ]] && + error "Binary overwritten during exec" + fi # cleanup # remove no try action mode @@ -2180,7 +2302,10 @@ test_31b() { mkdir -p $DIR/$tdir local f=$DIR/$tdir/$tfile - local fid=$(make_large_for_progress $f) + local fid + fid=$(make_large_for_progress $f) + [ $? != 0 ] && skip "not enough free space" && return + $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f wait_request_state $fid ARCHIVE SUCCEED $LFS hsm_release $f @@ -2201,7 +2326,10 @@ test_31c() { mkdir -p $DIR/$tdir local f=$DIR/$tdir/$tfile - local fid=$(make_large_for_progress_aligned $f) + local fid + fid=$(make_large_for_progress_aligned $f) + [ $? != 0 ] && skip "not enough free space" && return + $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f wait_request_state $fid ARCHIVE SUCCEED $LFS hsm_release $f @@ -2222,7 +2350,10 @@ test_33() { mkdir -p $DIR/$tdir local f=$DIR/$tdir/$tfile - local fid=$(make_large_for_progress $f) + local fid + fid=$(make_large_for_progress $f) + [ $? != 0 ] && skip "not enough free space" && return + $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f wait_request_state $fid ARCHIVE SUCCEED $LFS hsm_release $f @@ -2286,7 +2417,10 @@ test_34() { mkdir -p $DIR/$tdir local f=$DIR/$tdir/$tfile - local fid=$(make_large_for_progress $f) + local fid + fid=$(make_large_for_progress $f) + [ $? != 0 ] && skip "not enough free space" && return + $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f wait_request_state $fid ARCHIVE SUCCEED $LFS hsm_release $f @@ -2319,7 +2453,10 @@ test_35() { local f=$DIR/$tdir/$tfile local f1=$DIR/$tdir/$tfile-1 - local fid=$(make_large_for_progress $f) + local fid + fid=$(make_large_for_progress $f) + [ $? != 0 ] && skip "not enough free space" && return + local fid1=$(copy_file /etc/passwd $f1) $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f wait_request_state $fid ARCHIVE SUCCEED @@ -2355,7 +2492,10 @@ test_36() { mkdir -p $DIR/$tdir local f=$DIR/$tdir/$tfile - local fid=$(make_large_for_progress $f) + local fid + fid=$(make_large_for_progress $f) + [ $? != 0 ] && skip "not enough free space" && return + $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f wait_request_state $fid ARCHIVE SUCCEED $LFS hsm_release $f @@ -2409,7 +2549,7 @@ test_40() { done # force copytool to use a local/temp archive dir to ensure best # performance vs remote/NFS mounts used in auto-tests - if df --local $HSM_ARCHIVE >/dev/null 2>&1 ; then + if do_facet $SINGLEAGT "df --local $HSM_ARCHIVE" >/dev/null 2>&1 ; then copytool_setup else copytool_setup $SINGLEAGT $MOUNT $HSM_ARCHIVE_NUMBER $TMP/$tdir @@ -2548,7 +2688,9 @@ test_56() { mkdir -p $DIR/$tdir local f=$DIR/$tdir/$tfile - local fid=$(make_large_for_progress $f) + local fid + fid=$(make_large_for_progress $f) + [ $? != 0 ] && skip "not enough free space" && return $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f || error "could not archive file" @@ -2659,10 +2801,10 @@ run_test 58 "Truncate a released file will trigger restore" test_60() { # This test validates the fix for LU-4512. Ensure that the -u - # option changes the progress reporting interval from the default - # (30 seconds) to the user-specified interval. + # option changes the progress reporting interval from the + # default (30 seconds) to the user-specified interval. local interval=5 - local progress_timeout=$((interval * 3)) + local progress_timeout=$((interval * 4)) # test needs a new running copytool copytool_cleanup @@ -2670,15 +2812,32 @@ test_60() { mkdir -p $DIR/$tdir local f=$DIR/$tdir/$tfile - local fid=$(make_large_for_progress $f) + local fid + fid=$(make_large_for_progress $f) + [ $? != 0 ] && skip "not enough free space" && return + + local mdtidx=0 + local mdt=${MDT_PREFIX}${mdtidx} + local mds=mds$((mdtidx + 1)) + + # Wait for copytool to register + wait_update_facet $mds \ + "$LCTL get_param -n ${mdt}.hsm.agents | grep -o ^uuid" \ + uuid 100 || error "coyptool failed to register with $mdt" local start_at=$(date +%s) $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f || error "could not archive file" - local mdtidx=0 - local mdt=${MDT_PREFIX}${mdtidx} - local mds=mds$((mdtidx + 1)) + local agent=$(facet_active_host $SINGLEAGT) + local prefix=$TESTLOG_PREFIX + [[ -z "$TESTNAME" ]] || prefix=$prefix.$TESTNAME + local copytool_log=$prefix.copytool_log.$agent.log + + + wait_update $agent \ + "grep -o start.copy $copytool_log" "start copy" 100 || + error "copytool failed to start" local cmd="$LCTL get_param -n ${mdt}.hsm.active_requests" cmd+=" | awk '/'$fid'.*action=ARCHIVE/ {print \\\$12}' | cut -f2 -d=" @@ -2722,6 +2881,13 @@ test_70() { # Just start and stop the copytool to generate events. cdt_clear_no_retry + + # Wait for the copytool to register. + wait_update --verbose $(facet_active_host mds1) \ + "$LCTL get_param -n ${MDT_PREFIX}0.hsm.agents | grep -o ^uuid" \ + uuid 100 || + error "copytool failed to register with MDT0000" + copytool_cleanup local REGISTER_EVENT @@ -2765,7 +2931,9 @@ test_71() { mkdir -p $DIR/$tdir local f=$DIR/$tdir/$tfile - local fid=$(make_large_for_progress $f) + local fid + fid=$(make_large_for_progress $f) + [ $? != 0 ] && skip "not enough free space" && return $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f || error "could not archive file" @@ -2947,7 +3115,14 @@ test_90() { fid=$(copy_file /etc/hosts $f.$i) echo $f.$i >> $FILELIST done - copytool_setup + # force copytool to use a local/temp archive dir to ensure best + # performance vs remote/NFS mounts used in auto-tests + if do_facet $SINGLEAGT "df --local $HSM_ARCHIVE" >/dev/null 2>&1 ; then + copytool_setup + else + local dai=$(get_hsm_param default_archive_id) + copytool_setup $SINGLEAGT $MOUNT $dai $TMP/$tdir + fi # to be sure wait_all_done will not be mislead by previous tests cdt_purge wait_for_grace_delay @@ -3034,7 +3209,10 @@ test_104() { mkdir -p $DIR/$tdir local f=$DIR/$tdir/$tfile - local fid=$(make_large_for_progress $f) + local fid + fid=$(make_large_for_progress $f) + [ $? != 0 ] && skip "not enough free space" && return + # if cdt is on, it can serve too quickly the request cdt_disable $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER --data $DATA $f @@ -3345,7 +3523,10 @@ test_200() { mkdir -p $DIR/$tdir local f=$DIR/$tdir/$tfile - local fid=$(make_large_for_cancel $f) + local fid + fid=$(make_large_for_cancel $f) + [ $? != 0 ] && skip "not enough free space" && return + # test with cdt on is made in test_221 cdt_disable $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f @@ -3386,7 +3567,10 @@ test_202() { mkdir -p $DIR/$tdir local f=$DIR/$tdir/$tfile - local fid=$(make_large_for_progress $f) + local fid + fid=$(make_large_for_progress $f) + [ $? != 0 ] && skip "not enough free space" && return + $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f wait_request_state $fid ARCHIVE SUCCEED @@ -3431,7 +3615,9 @@ test_221() { mkdir -p $DIR/$tdir local f=$DIR/$tdir/$tfile - local fid=$(make_large_for_cancel $f) + local fid + fid=$(make_large_for_cancel $f) + [ $? != 0 ] && skip "not enough free space" && return changelog_setup @@ -3538,7 +3724,9 @@ test_223b() { mkdir -p $DIR/$tdir local f=$DIR/$tdir/$tfile - local fid=$(make_large_for_progress $f) + local fid + fid=$(make_large_for_progress $f) + [ $? != 0 ] && skip "not enough free space" && return changelog_setup $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f @@ -3598,7 +3786,9 @@ test_225() { mkdir -p $DIR/$tdir local f=$DIR/$tdir/$tfile - local fid=$(make_large_for_progress $f) + local fid + fid=$(make_large_for_progress $f) + [ $? != 0 ] && skip "not enough free space" && return changelog_setup $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f @@ -3801,7 +3991,9 @@ test_251() { mkdir -p $DIR/$tdir local f=$DIR/$tdir/$tfile - local fid=$(make_large_for_cancel $f) + local fid + fid=$(make_large_for_cancel $f) + [ $? != 0 ] && skip "not enough free space" && return cdt_disable # to have a short test @@ -3813,6 +4005,10 @@ test_251() { set_hsm_param loop_period 2 cdt_enable + # clear locks to avoid extra delay caused by flush/cancel + # and thus prevent early copytool death to timeout. + cancel_lru_locks osc + $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f wait_request_state $fid ARCHIVE STARTED sleep 5 @@ -3978,9 +4174,9 @@ test_402() { copytool_cleanup # deactivate all mdc on agent1 - mdc_change_state $SINGLEAGT "MDT000." "deactivate" + mdc_change_state $SINGLEAGT "$FSNAME-MDT000." "deactivate" - copytool_setup $SINGLEAGT + HSMTOOL_NOERROR=true copytool_setup $SINGLEAGT check_agent_unregistered "uuid" # match any agent @@ -3988,7 +4184,7 @@ test_402() { search_copytools $agent && error "Copytool start should have failed" # reactivate MDCs - mdc_change_state $SINGLEAGT "MDT000." "activate" + mdc_change_state $SINGLEAGT "$FSNAME-MDT000." "activate" } run_test 402 "Copytool start fails if all MDTs are inactive" @@ -4002,7 +4198,7 @@ test_403() { local uuid=$(do_rpc_nodes $agent get_client_uuid | cut -d' ' -f2) # deactivate all mdc for MDT0001 - mdc_change_state $SINGLEAGT "MDT0001" "deactivate" + mdc_change_state $SINGLEAGT "$FSNAME-MDT0001" "deactivate" copytool_setup # check the agent is registered on MDT0000, and not on MDT0001 @@ -4013,7 +4209,7 @@ test_403() { search_copytools $agent || error "No running copytools on $agent" # reactivate all mdc for MDT0001 - mdc_change_state $SINGLEAGT "MDT0001" "activate" + mdc_change_state $SINGLEAGT "$FSNAME-MDT0001" "activate" # make sure the copytool is now registered to all MDTs check_agent_registered $uuid @@ -4037,7 +4233,7 @@ test_404() { local fid1=$(make_small $dir_mdt0/$tfile) # deactivate all mdc for MDT0001 - mdc_change_state $SINGLEAGT "MDT0001" "deactivate" + mdc_change_state $SINGLEAGT "$FSNAME-MDT0001" "deactivate" # send an HSM request for files in MDT0000 $LFS hsm_archive $dir_mdt0/$tfile || error "lfs hsm_archive" @@ -4047,7 +4243,7 @@ test_404() { echo "archive successful on mdt0" # reactivate all mdc for MDT0001 - mdc_change_state $SINGLEAGT "MDT0001" "activate" + mdc_change_state $SINGLEAGT "$FSNAME-MDT0001" "activate" copytool_cleanup # clean test files and directories @@ -4055,6 +4251,69 @@ test_404() { } run_test 404 "Inactive MDT does not block requests for active MDTs" +test_405() { + [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return + + copytool_setup + + mkdir -p $DIR/$tdir + + local striped_dir=$DIR/$tdir/striped_dir + + # create striped dir on all of MDTs + $LFS mkdir -i 0 -c $MDSCOUNT $striped_dir || error "lfs mkdir" + + local fid1=$(make_small_sync $striped_dir/${tfile}_0) + local fid2=$(make_small_sync $striped_dir/${tfile}_1) + local fid3=$(make_small_sync $striped_dir/${tfile}_2) + local fid4=$(make_small_sync $striped_dir/${tfile}_3) + + local idx1=$($LFS getstripe -M $striped_dir/${tfile}_0) + local idx2=$($LFS getstripe -M $striped_dir/${tfile}_1) + local idx3=$($LFS getstripe -M $striped_dir/${tfile}_2) + local idx4=$($LFS getstripe -M $striped_dir/${tfile}_3) + + # check that compound requests are shunt to the rights MDTs + $LFS hsm_archive $striped_dir/${tfile}_0 $striped_dir/${tfile}_1 \ + $striped_dir/${tfile}_2 $striped_dir/${tfile}_3 || + error "lfs hsm_archive" + + wait_request_state $fid1 ARCHIVE SUCCEED $idx1 && + echo "archive successful on $fid1" + wait_request_state $fid2 ARCHIVE SUCCEED $idx2 && + echo "archive successful on $fid2" + wait_request_state $fid3 ARCHIVE SUCCEED $idx3 && + echo "archive successful on $fid3" + wait_request_state $fid4 ARCHIVE SUCCEED $idx4 && + echo "archive successful on $fid4" + + $LFS hsm_release $striped_dir/${tfile}_0 || error "lfs hsm_release 1" + $LFS hsm_release $striped_dir/${tfile}_1 || error "lfs hsm_release 2" + $LFS hsm_release $striped_dir/${tfile}_2 || error "lfs hsm_release 3" + $LFS hsm_release $striped_dir/${tfile}_3 || error "lfs hsm_release 4" + + cat $striped_dir/${tfile}_0 > /dev/null || error "cat ${tfile}_0 failed" + cat $striped_dir/${tfile}_1 > /dev/null || error "cat ${tfile}_1 failed" + cat $striped_dir/${tfile}_2 > /dev/null || error "cat ${tfile}_2 failed" + cat $striped_dir/${tfile}_3 > /dev/null || error "cat ${tfile}_3 failed" + + copytool_cleanup +} +run_test 405 "archive and release under striped directory" + +test_500() +{ + [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.6.92) ] && + skip "HSM migrate is not supported" && return + + # Stop the existing copytool + copytool_cleanup + + test_mkdir -p $DIR/$tdir + llapi_hsm_test -d $DIR/$tdir || error "One llapi HSM test failed" +} +run_test 500 "various LLAPI HSM tests" + copytool_cleanup complete $SECONDS