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 $PWD)}
return 0
}
-make_large_for_striping() {
+make_custom_file_for_progress() {
local file2=${1/$DIR/$DIR2}
- local sz=$($LCTL get_param -n lov.*-clilov-*.stripesize | head -n1)
+ 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 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 $1"
-}
-
-make_large_for_progress_aligned() {
- local file2=${1/$DIR/$DIR2}
-
- 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"
}
+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
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)
$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"
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 ||
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
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 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
-}
-
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