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 $PWD)}
MULTIOP=${MULTIOP:-multiop}
OPENFILE=${OPENFILE:-openfile}
-MCREATE=${MCREATE:-mcreate}
MOUNT_2=${MOUNT_2:-"yes"}
FAIL_ON_ERROR=false
# archive is purged at copytool setup
HSM_ARCHIVE_PURGE=true
+
+ # Don't allow copytool error upon start/setup
+ HSMTOOL_NOERROR=false
}
# Get the backend root path for the given agent facet.
if [[ -z "$arc_id" ]] &&
do_facet $facet "pkill -CONT -x $HSMTOOL_BASE"; then
- echo "Wakeup copytool $facet on $agent"
+ echo "Only wakeup running copytool $facet on $agent"
return 0
fi
[[ -z "$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 $agent failed"
+ fi
+
trap cleanup EXIT
}
local oldstate
local mdt_hsmctrl
local hsm_root=$(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.
"$oldstate" 20 ||
error "mds${mdtno} cdt state is not $oldstate"
done
- do_facet $facet "rm -rf $hsm_root"
+ if do_facet $facet "df $hsm_root" >/dev/null 2>&1 ; then
+ do_facet $facet "rm -rf $hsm_root/*"
+ fi
}
copytool_suspend() {
local file=$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() {
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)"
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" " |
[[ $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=
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"
}
[ $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"
}
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"
}
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"
}
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 $1"
}
local cmd="$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 $mds"
}
error "user could not change hsm flags"
dd if=/etc/passwd of=$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 $f.append "0x00000003"
# Modify a file sets it dirty
error "user could not change hsm flags"
dd if=/dev/zero of=$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
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
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
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
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
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
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
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
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 $f
local f=$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
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
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
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 $f ||
error "could not archive file"
test_60() {
# This test validates the fix for LU-4512. Ensure that the -u
- # option changes the progress reporting interval from the default
- # (30 seconds) to the user-specified interval.
+ # option changes the progress reporting interval from the
+ # default (30 seconds) to the user-specified interval.
local interval=5
- local progress_timeout=$((interval * 3))
+ local progress_timeout=$((interval * 4))
# test needs a new running copytool
copytool_cleanup
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="
# 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
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 $f ||
error "could not archive file"
fid=$(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
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
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
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
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
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
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
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
# deactivate all mdc on agent1
mdc_change_state $SINGLEAGT "$FSNAME-MDT000." "deactivate"
- copytool_setup $SINGLEAGT
+ HSMTOOL_NOERROR=true copytool_setup $SINGLEAGT
check_agent_unregistered "uuid" # match any agent
}
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