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
echo "copytools still running on $hosts"
done
+ # 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 ${wait_timeout}s"
return 1
local lustre_mntpnt=${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 $HSMTOOL_UPDATE_INTERVAL"
local agt_facet=$SINGLEAGT
local agt_hosts=${1:-$(facet_active_host $agt_facet)}
local hsm_root=$(copytool_device $agt_facet)
+
+ [ -z "${hsm_root// /}" ] && error "copytool_cleanup: hsm_root empty!"
+
local i
local facet
local param
done
if do_facet $agt_facet "df $hsm_root" >/dev/null 2>&1 ; then
- do_facet $agt_facet "rm -rf $hsm_root/*"
+ do_facet $agt_facet "rm -rf $hsm_root/$HSMTMP/*"
fi
}
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
kill_copytools
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
$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"
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
}
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
}
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) ] &&