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
}
echo $raw_event | python -c "$json_parser"
}
-# populate MDT device array
-get_mdt_devices
-
# 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
}
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 $f > /dev/null &
}
run_test 58 "Truncate a released file will trigger restore"
+test_59() {
+ local fid
+ local server_version=$(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
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
# test with cdt on is made in test_221
cdt_disable
$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
+ # wait archive to register at CDT
+ wait_request_state $fid ARCHIVE WAITING
$LFS hsm_cancel $f
cdt_enable
wait_request_state $fid ARCHIVE CANCELED
# test with cdt on is made in test_222
cdt_disable
$LFS hsm_restore $f
+ # wait restore to register at CDT
+ wait_request_state $fid RESTORE WAITING
$LFS hsm_cancel $f
cdt_enable
wait_request_state $fid RESTORE CANCELED
cdt_disable
$LFS hsm_remove $f
+ # wait remove to register at CDT
+ wait_request_state $fid REMOVE WAITING
$LFS hsm_cancel $f
cdt_enable
wait_request_state $fid REMOVE CANCELED
}
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) ] &&