export PATH=$PWD/$SRCDIR:$SRCDIR:$PWD/$SRCDIR/utils:$PATH:/sbin:/usr/sbin
ONLY=${ONLY:-"$*"}
-# bug number for skipped test: LU-3815
-ALWAYS_EXCEPT="$SANITY_HSM_EXCEPT 34 35 36"
+# bug number for skipped test:
+ALWAYS_EXCEPT="$SANITY_HSM_EXCEPT"
# UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT!
LUSTRE=${LUSTRE:-$(cd $(dirname $0)/..; echo $PWD)}
MULTIOP=${MULTIOP:-multiop}
OPENFILE=${OPENFILE:-openfile}
+MMAP_CAT=${MMAP_CAT:-mmap_cat}
MOUNT_2=${MOUNT_2:-"yes"}
FAIL_ON_ERROR=false
export HSMTOOL_EVENT_FIFO=${HSMTOOL_EVENT_FIFO:=""}
export HSMTOOL_TESTDIR
export HSMTOOL_BASE=$(basename "$HSMTOOL" | cut -f1 -d" ")
+ # $hsm_root/$HSMTMP Makes $hsm_root dir path less generic to ensure
+ # rm -rf $hsm_root/* is safe even if $hsm_root becomes unset to avoid
+ # deleting everything in filesystem, independent of any copytool.
+ export HSMTMP=${HSMTMP:-"shsm"}
+
HSM_ARCHIVE=$(copytool_device $SINGLEAGT)
+
+ [ -z "${HSM_ARCHIVE// /}" ] && error "HSM_ARCHIVE is empty!"
+ HSM_ARCHIVE=$HSM_ARCHIVE/$HSMTMP
+
HSM_ARCHIVE_NUMBER=2
# The test only support up to 10 MDTs
}
search_copytools() {
- local agents=${1:-$(facet_active_host $SINGLEAGT)}
- do_nodesv $agents "pgrep -x $HSMTOOL_BASE"
+ local hosts=${1:-$(facet_active_host $SINGLEAGT)}
+ do_nodesv $hosts "pgrep -x $HSMTOOL_BASE"
}
-search_and_kill_copytool() {
- local agents=${1:-$(facet_active_host $SINGLEAGT)}
+kill_copytools() {
+ local hosts=${1:-$(facet_active_host $SINGLEAGT)}
+
+ echo "Killing existing copytools on $hosts"
+ do_nodesv $hosts "killall -q $HSMTOOL_BASE" || true
+}
+
+wait_copytools() {
+ local hosts=${1:-$(facet_active_host $SINGLEAGT)}
+ local wait_timeout=200
+ local wait_start=$SECONDS
+ local wait_end=$((wait_start + wait_timeout))
+
+ while ((SECONDS < wait_end)); do
+ sleep 2
+ if ! search_copytools $hosts; then
+ echo "copytools stopped in $((SECONDS - wait_start))s"
+ return 0
+ fi
+
+ echo "copytools still running on $hosts"
+ done
- echo "Killing existing copytools on $agents"
- do_nodesv $agents "killall -q $HSMTOOL_BASE" || true
+ # try to dump Copytool's stack
+ do_nodesv $hosts "echo 1 >/proc/sys/kernel/sysrq ; " \
+ "echo t >/proc/sysrq-trigger"
+
+ echo "copytools failed to stop in ${wait_timeout}s"
+
+ return 1
}
copytool_monitor_setup() {
local lustre_mntpnt=${2:-${MOUNT2:-$MOUNT}}
local arc_id=$3
local hsm_root=${4:-$(copytool_device $facet)}
+
+ [ -z "${hsm_root// /}" ] && error "copytool_setup: hsm_root empty!"
+
local agent=$(facet_active_host $facet)
if [[ -z "$arc_id" ]] &&
if $HSM_ARCHIVE_PURGE; then
echo "Purging archive on $agent"
- do_facet $facet "rm -rf $hsm_root/*"
+ do_facet $facet "rm -rf $hsm_root/$HSMTMP/*"
fi
echo "Starting copytool $facet on $agent"
- do_facet $facet "mkdir -p $hsm_root" || error "mkdir '$hsm_root' failed"
+ do_facet $facet "mkdir -p $hsm_root/$HSMTMP/" ||
+ error "mkdir '$hsm_root/$HSMTMP' failed"
# bandwidth is limited to 1MB/s so the copy time is known and
# independent of hardware
- local cmd="$HSMTOOL $HSMTOOL_VERBOSE --daemon --hsm-root $hsm_root"
+ local cmd="$HSMTOOL $HSMTOOL_VERBOSE --daemon"
+ cmd+=" --hsm-root $hsm_root/$HSMTMP"
[[ -z "$arc_id" ]] || cmd+=" --archive $arc_id"
[[ -z "$HSMTOOL_UPDATE_INTERVAL" ]] ||
cmd+=" --update-interval $HSMTOOL_UPDATE_INTERVAL"
copytool_cleanup() {
trap - EXIT
- local facet=$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 ))
+ local agt_facet=$SINGLEAGT
+ local agt_hosts=${1:-$(facet_active_host $agt_facet)}
+ local hsm_root=$(copytool_device $agt_facet)
- do_nodesv $agents "pkill -INT -x $HSMTOOL_BASE" || return 0
+ [ -z "${hsm_root// /}" ] && error "copytool_cleanup: hsm_root empty!"
- 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"
+ local i
+ local facet
+ local param
+ local -a state
+
+ kill_copytools $agt_hosts
+ wait_copytools $agt_hosts || error "copytools failed to stop"
+
+ # Clean all CDTs orphans requests from previous tests that
+ # would otherwise need to timeout to clear.
+ for ((i = 0; i < MDSCOUNT; i++)); do
+ facet=mds$((i + 1))
+ param=$(printf 'mdt.%s-MDT%04x.hsm_control' $FSNAME $i)
+ state[$i]=$(do_facet $facet "$LCTL get_param -n $param")
+
+ # Skip already stopping or stopped CDTs.
+ [[ "${state[$i]}" =~ ^stop ]] && continue
+
+ do_facet $facet "$LCTL set_param $param=shutdown"
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.
- for mdtno in $(seq 1 $MDSCOUNT); do
- idx=$(($mdtno - 1))
- mdt_hsmctrl="mdt.$FSNAME-MDT000${idx}.hsm_control"
- oldstate=$(do_facet mds${mdtno} "$LCTL get_param -n " \
- "$mdt_hsmctrl")
- # skip already stop[ed,ing] CDTs
- echo $oldstate | grep stop && continue
-
- do_facet mds${mdtno} "$LCTL set_param $mdt_hsmctrl=shutdown"
- wait_result mds${mdtno} "$LCTL get_param -n $mdt_hsmctrl" \
- "stopped" 20 ||
- error "mds${mdtno} cdt state is not stopped"
- do_facet mds${mdtno} "$LCTL set_param $mdt_hsmctrl=$oldstate"
- wait_result mds${mdtno} "$LCTL get_param -n $mdt_hsmctrl" \
- "$oldstate" 20 ||
- error "mds${mdtno} cdt state is not $oldstate"
+ for ((i = 0; i < MDSCOUNT; i++)); do
+ # Only check and restore CDTs that we stopped in the first loop.
+ [[ "${state[$i]}" =~ ^stop ]] && continue
+
+ facet=mds$((i + 1))
+ param=$(printf 'mdt.%s-MDT%04x.hsm_control' $FSNAME $i)
+
+ wait_result $facet "$LCTL get_param -n $param" stopped 20 ||
+ error "$facet CDT state is not stopped"
+
+ # Restore old CDT state.
+ do_facet $facet "$LCTL set_param $param=${state[$i]}"
+ done
+
+ for ((i = 0; i < MDSCOUNT; i++)); do
+ # Only check CDTs that we stopped in the first loop.
+ [[ "${state[$i]}" =~ ^stop ]] && continue
+
+ facet=mds$((i + 1))
+ param=$(printf 'mdt.%s-MDT%04x.hsm_control' $FSNAME $i)
+
+ # Check that the old CDT state was restored.
+ wait_result $facet "$LCTL get_param -n $param" "${state[$i]}" \
+ 20 || error "$facet CDT state is not '${state[$i]}'"
done
- if do_facet $facet "df $hsm_root" >/dev/null 2>&1 ; then
- do_facet $facet "rm -rf $hsm_root/*"
+
+ if do_facet $agt_facet "df $hsm_root" >/dev/null 2>&1 ; then
+ do_facet $agt_facet "rm -rf $hsm_root/$HSMTMP/*"
fi
}
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 ||
- file_creation_failure dd $file2 $?
-
- path2fid $1 || error "cannot get fid on $1"
-}
-
-make_large_for_progress() {
- local file2=${1/$DIR/$DIR2}
-
- 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 ||
- file_creation_failure dd $file2 $?
-
- path2fid $1 || error "cannot get fid on $1"
-}
-
-make_large_for_progress_aligned() {
+make_custom_file_for_progress() {
local file2=${1/$DIR/$DIR2}
+ local fsize=${2:-"39"}
+ local blksz=$($LCTL get_param -n lov.*-clilov-*.stripesize | head -n1)
+ blksz=${3:-$blksz}
- 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 ||
- file_creation_failure dd $file2 $?
- path2fid $1 || error "cannot get fid on $1"
-}
-
-make_large_for_cancel() {
- local file2=${1/$DIR/$DIR2}
+ [[ $fsize -gt 0 ]] || error "Invalid file size"
+ [[ $blksz -gt 0 ]] || error "Invalid stripe size"
cleanup_large_files
-
- check_enough_free_space 103 1048576
+ check_enough_free_space $fsize $blksz
[ $? != 0 ] && return $?
-
- # Copy timeout is 100s. 105MB => 105s
- dd if=/dev/urandom of=$file2 count=103 bs=1M conv=fsync ||
+ dd if=/dev/zero of=$file2 count=$fsize bs=$blksz conv=fsync ||
file_creation_failure dd $file2 $?
path2fid $1 || error "cannot get fid on $1"
}
sleep $val
}
+wait_for_loop_period() {
+ local val=$(get_hsm_param loop_period)
+ sleep $val
+}
+
parse_json_event() {
local raw_event=$1
echo $raw_event | python -c "$json_parser"
}
-# populate MDT device array
-get_mdt_devices
+get_agent_by_uuid_mdt() {
+ local uuid=$1
+ local mdtidx=$2
+ local mds=mds$(($mdtidx + 1))
+ do_facet $mds "$LCTL get_param -n ${MDT_PREFIX}${mdtidx}.hsm.agents |\
+ grep $uuid"
+}
+
+check_agent_registered_by_mdt() {
+ local uuid=$1
+ local mdtidx=$2
+ local mds=mds$(($mdtidx + 1))
+ local agent=$(get_agent_by_uuid_mdt $uuid $mdtidx)
+ if [[ ! -z "$agent" ]]; then
+ echo "found agent $agent on $mds"
+ else
+ error "uuid $uuid not found in agent list on $mds"
+ fi
+}
+
+check_agent_unregistered_by_mdt() {
+ local uuid=$1
+ local mdtidx=$2
+ local mds=mds$(($mdtidx + 1))
+ local agent=$(get_agent_by_uuid_mdt $uuid $mdtidx)
+ if [[ -z "$agent" ]]; then
+ echo "uuid not found in agent list on $mds"
+ else
+ error "uuid found in agent list on $mds: $agent"
+ fi
+}
+
+check_agent_registered() {
+ local uuid=$1
+ local mdsno
+ for mdsno in $(seq 1 $MDSCOUNT); do
+ check_agent_registered_by_mdt $uuid $((mdsno - 1))
+ done
+}
+
+check_agent_unregistered() {
+ local uuid=$1
+ local mdsno
+ for mdsno in $(seq 1 $MDSCOUNT); do
+ check_agent_unregistered_by_mdt $uuid $((mdsno - 1))
+ done
+}
+
+get_agent_uuid() {
+ local agent=${1:-$(facet_active_host $SINGLEAGT)}
+
+ # Lustre mount-point is mandatory and last parameter on
+ # copytool cmd-line.
+ local mntpnt=$(do_rpc_nodes $agent ps -C $HSMTOOL_BASE -o args= |
+ awk '{print $NF}')
+ [ -n "$mntpnt" ] || error "Found no Agent or with no mount-point "\
+ "parameter"
+ do_rpc_nodes $agent get_client_uuid $mntpnt | cut -d' ' -f2
+}
# initiate variables
init_agt_vars
+# populate MDT device array
+get_mdt_devices
+
# cleanup from previous bad setup
-search_and_kill_copytool
+kill_copytools
# for recovery tests, coordinator needs to be started at mount
# so force it
}
run_test 1 "lfs hsm flags root/non-root access"
+test_1a() {
+ mkdir -p $DIR/$tdir
+ local f=$DIR/$tdir/$tfile
+ local fid=$(make_small $f)
+
+ $LFS hsm_archive $f || error "could not archive file"
+ wait_request_state $fid ARCHIVE SUCCEED
+
+ # Release and check states
+ $LFS hsm_release $f || error "could not release file"
+ echo -n "Verifying released state: "
+ check_hsm_flags $f "0x0000000d"
+
+ $MMAP_CAT $f > /dev/null || error "failed mmap & cat release file"
+}
+run_test 1a "mmap & cat a HSM released file"
+
test_2() {
mkdir -p $DIR/$tdir
local f=$DIR/$tdir/$tfile
run_test 8 "Test default archive number"
test_9() {
- mkdir -p $DIR/$tdir
- local f=$DIR/$tdir/$tfile
- local fid=$(copy_file /etc/passwd $f)
# we do not use the default one to be sure
local new_an=$((HSM_ARCHIVE_NUMBER + 1))
copytool_cleanup
copytool_setup $SINGLEAGT $MOUNT $new_an
+
+ # give time for CT to register with MDTs
+ sleep $(($MDSCOUNT*2))
+ local uuid=$(get_agent_uuid $(facet_active_host $SINGLEAGT))
+ check_agent_registered $uuid
+
+ mkdir -p $DIR/$tdir
+ local f=$DIR/$tdir/$tfile
+ local fid=$(copy_file /etc/passwd $f)
$LFS hsm_archive --archive $new_an $f
wait_request_state $fid ARCHIVE SUCCEED
local f=$DIR/$tdir/$tfile
$LFS setstripe -c 2 $f
local fid
- fid=$(make_large_for_striping $f)
+ fid=$(make_custom_file_for_progress $f 5)
[ $? != 0 ] && skip "not enough free space" && return
local FILE_CRC=$(md5sum $f)
}
test_12q() {
+ [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.7.58) ] &&
+ skip "need MDS version at least 2.7.58" && return 0
zconf_mount $(facet_host $SINGLEAGT) $MOUNT3 ||
error "cannot mount $MOUNT3 on $SINGLEAGT"
$LFS hsm_release $f || error "could not release file"
check_hsm_flags $f "0x0000000d"
- search_and_kill_copytool
- sleep 5
- search_copytools && error "Copytool should have stopped"
+ kill_copytools
+ wait_copytools || error "copytool failed to stop"
cat $f > /dev/null &
$LFS hsm_archive $f
wait_request_state $fid ARCHIVE SUCCEED
local end=$(date +%s)
- local duration=$((end - start))
+ # Add 1 to account for rounding errors between start and end (LU-8155)
+ local duration=$((end - start + 1))
[[ $duration -ge $goal ]] ||
error "Transfer is too fast $duration < $goal"
[ $ctime0 -eq $ctime1 ] ||
error "release changed ctime from $ctime0 to $ctime1"
- # Restore should not change atime or mtime and should not
- # decrease ctime.
+ # Restore should not change any timestamps.
$LFS hsm_restore $file
wait_request_state $fid RESTORE SUCCEED
chown $RUNAS_ID:nobody $file ||
error "cannot chown '$file' to '$RUNAS_ID:nobody'"
- set_hsm_param user_request_mask ""
$RUNAS $LFS hsm_$action $file &&
error "$action by user should fail"
chown nobody:$RUNAS_GID $file ||
error "cannot chown '$file' to 'nobody:$RUNAS_GID'"
- set_hsm_param group_request_mask ""
$RUNAS $LFS hsm_$action $file &&
error "$action by group should fail"
chown nobody:nobody $file ||
error "cannot chown '$file' to 'nobody:nobody'"
- set_hsm_param other_request_mask ""
$RUNAS $LFS hsm_$action $file &&
error "$action by other should fail"
wait_request_state $fid1 RESTORE SUCCEED
$LFS hsm_release $file1 || error "cannot release '$file1'"
- dd if=$file2 of=/dev/null bs=1M || "cannot read '$file2'"
+ dd if=$file2 of=/dev/null bs=1M || error "cannot read '$file2'"
$LFS hsm_release $file2 &&
error "release should fail on read-only mount"
mkdir -p $DIR/$tdir
local f=$DIR/$tdir/$tfile
local fid
- fid=$(make_large_for_progress $f)
+ fid=$(make_custom_file_for_progress $f 39 1000000)
[ $? != 0 ] && skip "not enough free space" && return
$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
mkdir -p $DIR/$tdir
local f=$DIR/$tdir/$tfile
local fid
- fid=$(make_large_for_progress $f)
+ fid=$(make_custom_file_for_progress $f 39 1000000)
[ $? != 0 ] && skip "not enough free space" && return
$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
mkdir -p $DIR/$tdir
local f=$DIR/$tdir/$tfile
local fid
- fid=$(make_large_for_progress $f)
+ fid=$(make_custom_file_for_progress $f 39 1000000)
[ $? != 0 ] && skip "not enough free space" && return
$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
}
run_test 29c "Archive/delete/remove by FID, using a file list."
+test_29d() {
+ # test needs more than one CT
+ needclients 3 || return 0
+
+ local n
+ local file
+ local fid
+
+ copytool_cleanup $(comma_list $(agts_nodes))
+
+ # start all of the copytools
+ for n in $(seq $AGTCOUNT); do
+ copytool_setup agt$n $MOUNT2 $n
+ done
+
+ trap "copytool_cleanup $(comma_list $(agts_nodes))" EXIT
+ # archive files
+ mkdir -p $DIR/$tdir
+ file=$DIR/$tdir/$tfile
+ fid=$(make_small $file)
+
+ $LFS hsm_archive $file
+ wait_request_state $fid ARCHIVE SUCCEED
+ check_hsm_flags $file "0x00000009"
+
+ rm -f $file
+
+ $LFS hsm_remove -a 0 $fid
+
+ # give time for CDT to handle remove request and create broadcasted
+ sleep 2
+
+ # remove request has been broadcasted ?
+ local cnt=$(get_request_count $fid REMOVE)
+ # broadcasted requests + original
+ [[ $cnt -eq $((AGTCOUNT + 1)) ]] ||
+ error "remove not broadcasted to all CTs"
+
+ # give time for CDT and CTs to handle broadcasted
+ wait_for_loop_period
+
+ # each agent serves one different archive_id, so broadcasted
+ # hsm_remove request should only succeed once and fail at all others
+ local res
+ local scnt=0
+ local fcnt=0
+ for n in $(seq $AGTCOUNT); do
+ res=$(do_facet $SINGLEMDS "$LCTL get_param -n \
+ $HSM_PARAM.actions | awk \
+ '/'$fid'.*action=REMOVE archive#='$n'/ \
+ {print \\\$13}' | cut -f2 -d=")
+ if [[ "$res" == "SUCCEED" ]]; then
+ scnt=$((scnt + 1))
+ elif [[ "$res" == "FAILED" ]]; then
+ fcnt=$((fcnt + 1))
+ fi
+ done
+
+ [[ $scnt -ne 1 ]] &&
+ error "one and only CT should have removed successfully"
+
+ [[ $AGTCOUNT -ne $((scnt + fcnt)) ]] &&
+ error "all but one CT should have failed to remove"
+
+ trap - EXIT
+ copytool_cleanup $(comma_list $(agts_nodes))
+
+}
+run_test 29d "hsm_remove by FID with archive_id 0 for unlinked file cause "\
+ "request to be sent once for each registered archive_id"
+
test_30a() {
# restore at exec cannot work on agent node (because of Linux kernel
# protection of executables)
while [[ "$st" != "0x00000009" && $cpt -le 10 ]]
do
n=$(stat -c "%s" $f)
- # we echo in both cases to show stat is not
- # hang
+ # we echo in both cases to show stat is not hang
if [[ $n != $s ]]; then
echo "size seen is $n != $s"
err=1
else
echo "size seen is right: $n == $s"
fi
- st=$(get_hsm_flags $f)
sleep 10
cpt=$((cpt + 1))
+ st=$(get_hsm_flags $f)
done
- if [[ $cpt -lt 10 ]]; then
+ if [[ "$st" = "0x00000009" ]]; then
echo " "done
else
echo " restore is too long"
local f=$DIR/$tdir/$tfile
local fid
- fid=$(make_large_for_progress $f)
+ fid=$(make_custom_file_for_progress $f 39 1000000)
[ $? != 0 ] && skip "not enough free space" && return
$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
local f=$DIR/$tdir/$tfile
local fid
- fid=$(make_large_for_progress_aligned $f)
+ fid=$(make_custom_file_for_progress $f 33 1048576)
[ $? != 0 ] && skip "not enough free space" && return
$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
local f=$DIR/$tdir/$tfile
local fid
- fid=$(make_large_for_progress $f)
+ fid=$(make_custom_file_for_progress $f 39 1000000)
[ $? != 0 ] && skip "not enough free space" && return
$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
local f=$DIR/$tdir/$tfile
local fid
- fid=$(make_large_for_progress $f)
+ fid=$(make_custom_file_for_progress $f 39 1000000)
[ $? != 0 ] && skip "not enough free space" && return
$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
local f=$DIR/$tdir/$tfile
local f1=$DIR/$tdir/$tfile-1
local fid
- fid=$(make_large_for_progress $f)
+ fid=$(make_custom_file_for_progress $f 39 1000000)
[ $? != 0 ] && skip "not enough free space" && return
local fid1=$(copy_file /etc/passwd $f1)
local f=$DIR/$tdir/$tfile
local fid
- fid=$(make_large_for_progress $f)
+ fid=$(make_custom_file_for_progress $f 39 1000000)
[ $? != 0 ] && skip "not enough free space" && return
$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
mkdir -p $DIR/$tdir
local f=$DIR/$tdir/$tfile
- local fid=$(make_large_for_progress $f)
+ local fid=$(make_custom_file_for_progress $f 39 1000000)
$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
error "could not archive file"
mkdir -p $DIR/$tdir
local f=$DIR/$tdir/$tfile
- local fid=$(make_large_for_progress $f)
+ local fid=$(make_custom_file_for_progress $f 39 1000000)
$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
error "could not archive file"
mkdir -p $DIR/$tdir
local f=$DIR/$tdir/$tfile
local fid
- fid=$(make_large_for_progress $f)
+ fid=$(make_custom_file_for_progress $f 39 1000000)
[ $? != 0 ] && skip "not enough free space" && return
$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
}
run_test 58 "Truncate a released file will trigger restore"
+test_59() {
+ local fid
+ local server_version=$(lustre_version_code $SINGLEMDS)
+ [[ $server_version -lt $(version_code 2.7.63) ]] &&
+ skip "Need MDS version at least 2.7.63" && return
+
+ copytool_setup
+ $MCREATE $DIR/$tfile || error "mcreate failed"
+ $TRUNCATE $DIR/$tfile 42 || error "truncate failed"
+ $LFS hsm_archive $DIR/$tfile || error "archive request failed"
+ fid=$(path2fid $DIR/$tfile)
+ wait_request_state $fid ARCHIVE SUCCEED
+ $LFS hsm_release $DIR/$tfile || error "release failed"
+ copytool_cleanup
+}
+run_test 59 "Release stripeless file with non-zero size"
+
test_60() {
# This test validates the fix for LU-4512. Ensure that the -u
# option changes the progress reporting interval from the
mkdir -p $DIR/$tdir
local f=$DIR/$tdir/$tfile
local fid
- fid=$(make_large_for_progress $f)
+ fid=$(make_custom_file_for_progress $f 10)
[ $? != 0 ] && skip "not enough free space" && return
local mdtidx=0
local elapsed=$((finish_at - start_at))
# Ensure that the progress update occurred within the expected window.
- if [ $elapsed -lt $interval ]; then
+ if [ $elapsed -lt $((interval - 1)) ]; then
error "Expected progress update after at least $interval seconds"
fi
+ echo "Wait for on going archive hsm action to complete"
+ wait_update $agent "grep -o copied $copytool_log" "copied" 10 ||
+ echo "File archiving not completed even after 10 secs"
+
cdt_clear_no_retry
copytool_cleanup
}
mkdir -p $DIR/$tdir
local f=$DIR/$tdir/$tfile
local fid
- fid=$(make_large_for_progress $f)
+ fid=$(make_custom_file_for_progress $f 39 1000000)
[ $? != 0 ] && skip "not enough free space" && return
$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
mkdir -p $DIR/$tdir
local f=$DIR/$tdir/$tfile
local fid
- fid=$(make_large_for_progress $f)
+ fid=$(make_custom_file_for_progress $f 39 1000000)
[ $? != 0 ] && skip "not enough free space" && return
# if cdt is on, it can serve too quickly the request
}
run_test 105 "Restart of coordinator"
-get_agent_by_uuid_mdt() {
- local uuid=$1
- local mdtidx=$2
- local mds=mds$(($mdtidx + 1))
- do_facet $mds "$LCTL get_param -n ${MDT_PREFIX}${mdtidx}.hsm.agents |\
- grep $uuid"
-}
-
-check_agent_registered_by_mdt() {
- local uuid=$1
- local mdtidx=$2
- local mds=mds$(($mdtidx + 1))
- local agent=$(get_agent_by_uuid_mdt $uuid $mdtidx)
- if [[ ! -z "$agent" ]]; then
- echo "found agent $agent on $mds"
- else
- error "uuid $uuid not found in agent list on $mds"
- fi
-}
-
-check_agent_unregistered_by_mdt() {
- local uuid=$1
- local mdtidx=$2
- local mds=mds$(($mdtidx + 1))
- local agent=$(get_agent_by_uuid_mdt $uuid $mdtidx)
- if [[ -z "$agent" ]]; then
- echo "uuid not found in agent list on $mds"
- else
- error "uuid found in agent list on $mds: $agent"
- fi
-}
-
-check_agent_registered() {
- local uuid=$1
- local mdsno
- for mdsno in $(seq 1 $MDSCOUNT); do
- check_agent_registered_by_mdt $uuid $((mdsno - 1))
- done
-}
-
-check_agent_unregistered() {
- local uuid=$1
- local mdsno
- for mdsno in $(seq 1 $MDSCOUNT); do
- check_agent_unregistered_by_mdt $uuid $((mdsno - 1))
- done
-}
-
-get_agent_uuid() {
- local agent=${1:-$(facet_active_host $SINGLEAGT)}
-
- # Lustre mount-point is mandatory and last parameter on
- # copytool cmd-line.
- local mntpnt=$(do_rpc_nodes $agent pgrep -fl $HSMTOOL_BASE |
- grep -v pgrep | awk '{print $NF}')
- [ -n "$mntpnt" ] || error "Found no Agent or with no mount-point "\
- "parameter"
- do_rpc_nodes $agent get_client_uuid $mntpnt | cut -d' ' -f2
-}
-
test_106() {
# test needs a running copytool
copytool_setup
mkdir -p $DIR/$tdir
local f=$DIR/$tdir/$tfile
local fid
- fid=$(make_large_for_cancel $f)
+ fid=$(make_custom_file_for_progress $f 103 1048576)
[ $? != 0 ] && skip "not enough free space" && return
# test with cdt on is made in test_221
mkdir -p $DIR/$tdir
local f=$DIR/$tdir/$tfile
local fid
- fid=$(make_large_for_progress $f)
+ fid=$(make_custom_file_for_progress $f 39 1000000)
[ $? != 0 ] && skip "not enough free space" && return
$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
local f=$DIR/$tdir/$tfile
local fid
- fid=$(make_large_for_cancel $f)
+ fid=$(make_custom_file_for_progress $f 103 1048576)
[ $? != 0 ] && skip "not enough free space" && return
changelog_setup
local f=$DIR/$tdir/$tfile
local fid
- fid=$(make_large_for_progress $f)
+ fid=$(make_custom_file_for_progress $f 39 1000000)
[ $? != 0 ] && skip "not enough free space" && return
changelog_setup
mkdir -p $DIR/$tdir
local f=$DIR/$tdir/$tfile
local fid
- fid=$(make_large_for_progress $f)
+ fid=$(make_custom_file_for_progress $f 39 1000000)
[ $? != 0 ] && skip "not enough free space" && return
changelog_setup
mkdir -p $DIR/$tdir
local f=$DIR/$tdir/$tfile
local fid
- fid=$(make_large_for_cancel $f)
+ fid=$(make_custom_file_for_progress $f 103 1048576)
[ $? != 0 ] && skip "not enough free space" && return
cdt_disable
}
run_test 405 "archive and release under striped directory"
+test_406() {
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+
+ [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.7.64) ] &&
+ skip "need MDS version at least 2.7.64" && return 0
+
+ local fid
+ local mdt_index
+
+ copytool_setup
+ mkdir -p $DIR/$tdir
+ fid=$(make_small $DIR/$tdir/$tfile)
+ echo "old fid $fid"
+
+ $LFS hsm_archive $DIR/$tdir/$tfile
+ wait_request_state "$fid" ARCHIVE SUCCEED
+ $LFS hsm_release $DIR/$tdir/$tfile
+
+ # Should migrate $tdir but not $tfile.
+ $LFS mv -M1 $DIR/$tdir &&
+ error "migrating HSM an archived file should fail"
+
+ $LFS hsm_restore $DIR/$tdir/$tfile
+ wait_request_state "$fid" RESTORE SUCCEED
+
+ $LFS hsm_remove $DIR/$tdir/$tfile
+ wait_request_state "$fid" REMOVE SUCCEED
+
+ cat $DIR/$tdir/$tfile > /dev/null ||
+ error "cannot read $DIR/$tdir/$tfile"
+
+ $LFS mv -M1 $DIR/$tdir ||
+ error "cannot complete migration after HSM remove"
+
+ mdt_index=$($LFS getstripe -M $DIR/$tdir)
+ if ((mdt_index != 1)); then
+ error "expected MDT index 1, got $mdt_index"
+ fi
+
+ # Refresh fid after migration.
+ fid=$(path2fid $DIR/$tdir/$tfile)
+ echo "new fid $fid"
+
+ $LFS hsm_archive $DIR/$tdir/$tfile
+ wait_request_state "$fid" ARCHIVE SUCCEED 1
+
+ lctl set_param debug=+trace
+ $LFS hsm_release $DIR/$tdir/$tfile ||
+ error "cannot release $DIR/$tdir/$tfile"
+
+ $LFS hsm_restore $DIR/$tdir/$tfile
+ wait_request_state "$fid" RESTORE SUCCEED 1
+
+ cat $DIR/$tdir/$tfile > /dev/null ||
+ error "cannot read $DIR/$tdir/$tfile"
+
+ copytool_cleanup
+}
+run_test 406 "attempting to migrate HSM archived files is safe"
+
test_500()
{
[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.6.92) ] &&