# LU-2829 / LU-2887 - make allowances for ZFS slowness
TEST33_NFILES=${TEST33_NFILES:-1000}
fi
-
+# 23 (min)"
[ "$SLOW" = "no" ] && EXCEPT_SLOW="33a"
FAIL_ON_ERROR=false
check_and_setup_lustre
-LOVNAME=$($LCTL get_param -n llite.*.lov.common_name | tail -n 1)
-OSTCOUNT=$($LCTL get_param -n lov.$LOVNAME.numobd)
-
assert_DIR
rm -rf $DIR1/[df][0-9]* $DIR1/lnk $DIR/[df].${TESTSUITE}*
}
run_test 13 "test directory page revocation ===================="
-test_14() {
+test_14aa() {
test_mkdir -p $DIR1/$tdir
cp -p /bin/ls $DIR1/$tdir/$tfile
multiop_bg_pause $DIR1/$tdir/$tfile Ow_c || return 1
kill -USR1 $MULTIPID
wait $MULTIPID || return 2
}
-run_test 14 "execution of file open for write returns -ETXTBSY ="
+run_test 14aa "execution of file open for write returns -ETXTBSY"
-test_14a() {
+test_14ab() {
test_mkdir -p $DIR1/d14
cp -p `which multiop` $DIR1/d14/multiop || error "cp failed"
MULTIOP_PROG=$DIR1/d14/multiop multiop_bg_pause $TMP/test14.junk O_c || return 1
wait $MULTIOP_PID || return 3
rm $TMP/test14.junk $DIR1/d14/multiop || error "removing multiop"
}
-run_test 14a "open(RDWR) of executing file returns -ETXTBSY ===="
+run_test 14ab "open(RDWR) of executing file returns -ETXTBSY"
test_14b() { # bug 3192, 7040
test_mkdir -p $DIR1/d14
avgjbd=0
avgtime=0
for i in 1 2 3; do
- do_node $CLIENT1 "$LFS mkdir -i $MDTIDX -p \
+ do_node $CLIENT1 "$LFS mkdir -i $MDTIDX \
$DIR1/$tdir-\\\$(hostname)-$i"
jbdold=$(print_jbd_stat)
}
run_test 33b "COS: cross create/delete, 2 clients, benchmark under remote dir"
+test_33c() {
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
+ [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.7.63) ] &&
+ skip "DNE CoS not supported" && return
+
+ sync
+
+ mkdir $DIR/$tdir
+ # remote mkdir is done on MDT2, which enqueued lock of $tdir on MDT1
+ $LFS mkdir -i 1 $DIR/$tdir/d1
+ do_facet mds1 "lctl set_param -n mdt.*.sync_count=0"
+ mkdir $DIR/$tdir/d2
+ local sync_count=$(do_facet mds1 \
+ "lctl get_param -n mdt.*MDT0000.sync_count")
+ [ $sync_count -eq 1 ] || error "Sync-Lock-Cancel not triggered"
+
+ $LFS mkdir -i 1 $DIR/$tdir/d3
+ do_facet mds1 "lctl set_param -n mdt.*.sync_count=0"
+ # during sleep remote mkdir should have been committed and canceled
+ # remote lock spontaneously, which shouldn't trigger sync
+ sleep 6
+ mkdir $DIR/$tdir/d4
+ local sync_count=$(do_facet mds1 \
+ "lctl get_param -n mdt.*MDT0000.sync_count")
+ [ $sync_count -eq 0 ] || error "Sync-Lock-Cancel triggered"
+}
+run_test 33c "Cancel cross-MDT lock should trigger Sync-Lock-Cancel"
+
+ops_do_cos() {
+ local nodes=$(comma_list $(mdts_nodes))
+ do_nodes $nodes "lctl set_param -n mdt.*.async_commit_count=0"
+ sh -c "$@"
+ local async_commit_count=$(do_nodes $nodes \
+ "lctl get_param -n mdt.*.async_commit_count" | calc_sum)
+ [ $async_commit_count -gt 0 ] || error "CoS not triggerred"
+
+ rm -rf $DIR/$tdir
+ sync
+}
+
+test_33d() {
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
+ [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.7.63) ] &&
+ skip "DNE CoS not supported" && return
+
+ sync
+ # remote directory create
+ mkdir $DIR/$tdir
+ ops_do_cos "$LFS mkdir -i 1 $DIR/$tdir/subdir"
+ # remote directory unlink
+ $LFS mkdir -i 1 $DIR/$tdir
+ ops_do_cos "rmdir $DIR/$tdir"
+ # striped directory create
+ mkdir $DIR/$tdir
+ ops_do_cos "$LFS mkdir -c 2 $DIR/$tdir/subdir"
+ # striped directory setattr
+ $LFS mkdir -c 2 $DIR/$tdir
+ touch $DIR/$tdir
+ ops_do_cos "chmod 713 $DIR/$tdir"
+ # striped directory unlink
+ $LFS mkdir -c 2 $DIR/$tdir
+ touch $DIR/$tdir
+ ops_do_cos "rmdir $DIR/$tdir"
+ # cross-MDT link
+ $LFS mkdir -c 2 $DIR/$tdir
+ $LFS mkdir -i 0 $DIR/$tdir/d1
+ $LFS mkdir -i 1 $DIR/$tdir/d2
+ touch $DIR/$tdir/d1/tgt
+ ops_do_cos "ln $DIR/$tdir/d1/tgt $DIR/$tdir/d2/src"
+ # cross-MDT rename
+ $LFS mkdir -c 2 $DIR/$tdir
+ $LFS mkdir -i 0 $DIR/$tdir/d1
+ $LFS mkdir -i 1 $DIR/$tdir/d2
+ touch $DIR/$tdir/d1/src
+ ops_do_cos "mv $DIR/$tdir/d1/src $DIR/$tdir/d2/tgt"
+ # migrate
+ $LFS mkdir -i 0 $DIR/$tdir
+ ops_do_cos "$LFS migrate -m 1 $DIR/$tdir"
+ return 0
+}
+run_test 33d "DNE distributed operation should trigger COS"
+
+test_33e() {
+ [ -n "$CLIENTS" ] || { skip "Need two or more clients" && return 0; }
+ [ $CLIENTCOUNT -ge 2 ] ||
+ { skip "Need two or more clients, have $CLIENTCOUNT" &&
+ return 0; }
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
+ [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.7.63) ] &&
+ skip "DNE CoS not supported" && return
+
+ local client2=${CLIENT2:-$(hostname)}
+
+ sync
+
+ local nodes=$(comma_list $(mdts_nodes))
+ do_nodes $nodes "lctl set_param -n mdt.*.async_commit_count=0"
+
+ $LFS mkdir -c 2 $DIR/$tdir
+ mkdir $DIR/$tdir/subdir
+ echo abc > $DIR/$tdir/$tfile
+ do_node $client2 echo dfg >> $DIR/$tdir/$tfile
+ do_node $client2 touch $DIR/$tdir/subdir
+
+ local async_commit_count=$(do_nodes $nodes \
+ "lctl get_param -n mdt.*.async_commit_count" | calc_sum)
+ [ $async_commit_count -gt 0 ] && error "CoS triggerred"
+
+ return 0
+}
+run_test 33e "DNE local operation shouldn't trigger COS"
+
# End commit on sharing tests
get_ost_lock_timeouts() {
}
test_34() { #16129
+ remote_ost_nodsh && skip "remote OST with nodsh" && return
local OPER
local lock_in
local lock_out
sync; sleep 5; sync # wait for delete thread
while [ $i -le 10 ]; do
- lctl mark "start test"
+ lctl mark "$HOSTNAME start test loop $i"
local before=$(lfs_df $MOUNT1 | awk '/^filesystem/{ print $4; exit }')
dd if=/dev/zero of=$DIR1/$tdir/$tfile bs=1M count=$SIZE ||
error "dd $DIR1/$tdir/$tfile ${SIZE}MB failed"
# pdirop tests
# test 40: check non-blocking operations
test_40a() {
+ remote_mds_nodsh && skip "remote MDS with nodsh" && return
#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
touch $DIR2
run_test 40a "pdirops: create vs others =============="
test_40b() {
+ remote_mds_nodsh && skip "remote MDS with nodsh" && return
#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
touch $DIR1/$tfile &
run_test 40b "pdirops: open|create and others =============="
test_40c() {
+ remote_mds_nodsh && skip "remote MDS with nodsh" && return
touch $DIR1/$tfile
#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
run_test 40c "pdirops: link and others =============="
test_40d() {
+ remote_mds_nodsh && skip "remote MDS with nodsh" && return
touch $DIR1/$tfile
#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
run_test 40d "pdirops: unlink and others =============="
test_40e() {
+ remote_mds_nodsh && skip "remote MDS with nodsh" && return
touch $DIR1/$tfile
#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
error "failed to set TBF policy"
done
- # Only operate rules on ost0 since OSTs might run on the same OSS
+ # Only operate rules on ost1 since OSTs might run on the same OSS
# Add some rules
- tbf_rule_operate ost0 "start\ localhost\ {0@lo}\ 1000"
+ tbf_rule_operate ost1 "start\ localhost\ {0@lo}\ 1000"
local address=$(comma_list "$(host_nids_address $CLIENTS $NETTYPE)")
local client_nids=$(nids_list $address "\\")
- tbf_rule_operate ost0 "start\ clients\ {$client_nids}\ 100"
- tbf_rule_operate ost0 "start\ others\ {*.*.*.*@$NETTYPE}\ 50"
+ tbf_rule_operate ost1 "start\ clients\ {$client_nids}\ 100"
+ tbf_rule_operate ost1 "start\ others\ {*.*.*.*@$NETTYPE}\ 50"
nrs_write_read
# Change the rules
- tbf_rule_operate ost0 "change\ localhost\ 1001"
- tbf_rule_operate ost0 "change\ clients\ 101"
- tbf_rule_operate ost0 "change\ others\ 51"
+ tbf_rule_operate ost1 "change\ localhost\ 1001"
+ tbf_rule_operate ost1 "change\ clients\ 101"
+ tbf_rule_operate ost1 "change\ others\ 51"
nrs_write_read
# Stop the rules
- tbf_rule_operate ost0 "stop\ localhost"
- tbf_rule_operate ost0 "stop\ clients"
- tbf_rule_operate ost0 "stop\ others"
+ tbf_rule_operate ost1 "stop\ localhost"
+ tbf_rule_operate ost1 "stop\ clients"
+ tbf_rule_operate ost1 "stop\ others"
nrs_write_read
# Cleanup the TBF policy
error "failed to set TBF policy"
done
- # Only operate rules on ost0 since OSTs might run on the same OSS
+ # Only operate rules on ost1 since OSTs might run on the same OSS
# Add some rules
- tbf_rule_operate ost0 "start\ runas\ {iozone.$RUNAS_ID\ dd.$RUNAS_ID\ tiotest.$RUNAS_ID}\ 1000"
- tbf_rule_operate ost0 "start\ iozone_runas\ {iozone.$RUNAS_ID}\ 100"
- tbf_rule_operate ost0 "start\ dd_runas\ {dd.$RUNAS_ID}\ 50"
+ tbf_rule_operate ost1 "start\ runas\ {iozone.$RUNAS_ID\ dd.$RUNAS_ID\ tiotest.$RUNAS_ID}\ 1000"
+ tbf_rule_operate ost1 "start\ iozone_runas\ {iozone.$RUNAS_ID}\ 100"
+ tbf_rule_operate ost1 "start\ dd_runas\ {dd.$RUNAS_ID}\ 50"
nrs_write_read "$RUNAS"
# Change the rules
- tbf_rule_operate ost0 "change\ runas\ 1001"
- tbf_rule_operate ost0 "change\ iozone_runas\ 101"
- tbf_rule_operate ost0 "change\ dd_runas\ 51"
+ tbf_rule_operate ost1 "change\ runas\ 1001"
+ tbf_rule_operate ost1 "change\ iozone_runas\ 101"
+ tbf_rule_operate ost1 "change\ dd_runas\ 51"
nrs_write_read "$RUNAS"
# Stop the rules
- tbf_rule_operate ost0 "stop\ runas"
- tbf_rule_operate ost0 "stop\ iozone_runas"
- tbf_rule_operate ost0 "stop\ dd_runas"
+ tbf_rule_operate ost1 "stop\ runas"
+ tbf_rule_operate ost1 "stop\ iozone_runas"
+ tbf_rule_operate ost1 "stop\ dd_runas"
nrs_write_read "$RUNAS"
# Cleanup the TBF policy
# Add a rule that only valid for Jobid TBF. If direct change between
# TBF types is not supported, this operation will fail.
- tbf_rule_operate ost0 "start\ dd_runas\ {dd.$RUNAS_ID}\ 50"
+ tbf_rule_operate ost1 "start\ dd_runas\ {dd.$RUNAS_ID}\ 50"
# Cleanup the TBF policy
for i in $(seq 1 $OSTCOUNT)
}
run_test 77g "Change TBF type directly"
-test_80() {
+test_78() { #LU-6673
+ local rc
+
+ for i in $(seq 1 $OSTCOUNT)
+ do
+ do_facet ost"$i" lctl set_param \
+ ost.OSS.ost_io.nrs_policies="orr" &
+ do_facet ost"$i" lctl set_param \
+ ost.OSS.*.nrs_orr_quantum=1
+ rc=$?
+ # Valid return codes are:
+ # 0: Tuning succeeded
+ # ENODEV: Policy is still stopped
+ # EAGAIN: Policy is being initialized
+ [ $rc -eq 0 -o $rc -eq 19 -o $rc -eq 11 ] ||
+ error "Expected set_param to return 0|ENODEV|EAGAIN"
+ done
+
+ # Cleanup the ORR policy
+ for i in $(seq 1 $OSTCOUNT)
+ do
+ do_facet ost"$i" lctl set_param \
+ ost.OSS.ost_io.nrs_policies="fifo"
+ [ $? -ne 0 ] &&
+ error "failed to set policy back to fifo"
+ done
+ return 0
+}
+run_test 78 "Enable policy and specify tunings right away"
+
+test_80a() {
[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
local MDTIDX=1
local mdt_index
# give multiop a chance to open
sleep 1
- $LFS mv -M $MDTIDX $DIR1/$tdir &&
+ $LFS migrate -m $MDTIDX $DIR1/$tdir &&
error "migrate open files should failed with open files"
kill -USR1 $pid
- $LFS mv -M $MDTIDX $DIR1/$tdir ||
+ $LFS migrate -m $MDTIDX $DIR1/$tdir ||
error "migrate remote dir error"
echo "Finish migration, then checking.."
rm -rf $DIR1/$tdir || error "rm dir failed after migration"
}
-run_test 80 "migrate directory when some children is being opened"
+run_test 80a "migrate directory when some children is being opened"
+
+cleanup_80b() {
+ trap 0
+ kill -9 $migrate_pid
+}
+
+test_80b() {
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
+ local migrate_dir1=$DIR1/$tdir/migrate_dir
+ local migrate_dir2=$DIR2/$tdir/migrate_dir
+ local migrate_run=$LUSTRE/tests/migrate.sh
+ local start_time
+ local end_time
+ local show_time=1
+ local mdt_idx
+ local rc=0
+ local rc1=0
+
+ trap cleanup_80b EXIT
+ #prepare migrate directory
+ mkdir -p $migrate_dir1
+ for F in {1,2,3,4,5}; do
+ echo "$F$F$F$F$F" > $migrate_dir1/file$F
+ echo "$F$F$F$F$F" > $DIR/$tdir/file$F
+ done
+
+ #migrate the directories among MDTs
+ (
+ while true; do
+ mdt_idx=$((RANDOM % MDSCOUNT))
+ $LFS migrate -m $mdt_idx $migrate_dir1 2&>/dev/null ||
+ rc=$?
+ [ $rc -ne 0 -o $rc -ne 16 ] || break
+ done
+ ) &
+ migrate_pid=$!
+
+ echo "start migration thread $migrate_pid"
+ #Access the files at the same time
+ start_time=$(date +%s)
+ echo "accessing the migrating directory for 5 minutes..."
+ while true; do
+ ls $migrate_dir2 > /dev/null || {
+ echo "read dir fails"
+ break
+ }
+ diff -u $DIR2/$tdir/file1 $migrate_dir2/file1 || {
+ echo "access file1 fails"
+ break
+ }
+
+ cat $migrate_dir2/file2 > $migrate_dir2/file3 || {
+ echo "access file2/3 fails"
+ break
+ }
+
+ echo "aaaaa" > $migrate_dir2/file4 > /dev/null || {
+ echo "access file4 fails"
+ break
+ }
+
+ stat $migrate_dir2/file5 > /dev/null || {
+ echo "stat file5 fails"
+ break
+ }
+
+ touch $migrate_dir2/source_file > /dev/null || rc1=$?
+ [ $rc1 -ne 0 -o $rc1 -ne 1 ] || {
+ echo "touch file failed with $rc1"
+ break;
+ }
+
+ if [ -e $migrate_dir2/source_file ]; then
+ ln $migrate_dir2/source_file $migrate_dir2/link_file \
+ 2&>/dev/null || rc1=$?
+ if [ -e $migrate_dir2/link_file ]; then
+ rm -rf $migrate_dir2/link_file
+ fi
+
+ mrename $migrate_dir2/source_file \
+ $migrate_dir2/target_file 2&>/dev/null || rc1=$?
+ [ $rc1 -ne 0 -o $rc1 -ne 1 ] || {
+ echo "rename failed with $rc1"
+ break
+ }
+
+ if [ -e $migrate_dir2/target_file ]; then
+ rm -rf $migrate_dir2/target_file 2&>/dev/null ||
+ rc1=$?
+ else
+ rm -rf $migrate_dir2/source_file 2&>/dev/null ||
+ rc1=$?
+ fi
+ [ $rc1 -ne 0 -o $rc1 -ne 1 ] || {
+ echo "unlink failed with $rc1"
+ break
+ }
+ fi
+
+ end_time=$(date +%s)
+ duration=$((end_time - start_time))
+ if [ $((duration % 10)) -eq 0 ]; then
+ if [ $show_time -eq 1 ]; then
+ echo "...$duration seconds"
+ show_time=0
+ fi
+ else
+ show_time=1
+ fi
+
+ kill -0 $migrate_pid || {
+ echo "migration stopped 1"
+ break
+ }
+
+ [ $duration -ge 300 ] && break
+ done
+
+ #check migration are still there
+ kill -0 $migrate_pid || error "migration stopped 2"
+ cleanup_80b
+}
+run_test 80b "Accessing directory during migration"
test_81() {
[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
}
run_test 82 "fsetxattr and fgetxattr on orphan files"
+test_83() {
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
+ local pid1
+ local pid2
+
+ (
+ cd $DIR1
+ while true; do
+ $LFS mkdir -i1 -c2 $tdir
+ rmdir $tdir
+ done
+ ) &
+ pid1=$!
+ echo "start pid $pid1 to create/unlink striped directory"
+
+ # Access the directory at the same time
+ (
+ cd $DIR2
+ while true; do
+ stat $tdir > /dev/null 2>&1
+ done
+ ) &
+ pid2=$!
+ echo "start pid $pid2 to stat striped directory"
+
+ sleep 120
+ kill $pid1 $pid2
+ wait $pid1 $pid2
+
+ return 0
+}
+run_test 83 "access striped directory while it is being created/unlinked"
+
+test_90() {
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
+ local pid1
+ local pid2
+ local duration=180
+
+ [ "$SLOW" = "yes" ] && duration=600
+ # Open/Create under striped directory
+ (
+ cd $DIR1
+ while true; do
+ $LFS mkdir -c$MDSCOUNT $tdir > /dev/null 2>&1
+ touch $tdir/f{0..3} > /dev/null 2>&1
+ done
+ ) &
+ pid1=$!
+ echo "start pid $pid1 to open/create under striped directory"
+
+ # unlink the striped directory at the same time
+ (
+ cd $DIR2
+ while true; do
+ rm -rf $tdir > /dev/null 2>&1
+ done
+ ) &
+ pid2=$!
+ echo "start pid $pid2 to unlink striped directory"
+
+ sleep $duration
+
+ kill $pid1 $pid2
+ wait $pid1 $pid2
+
+ return 0
+}
+run_test 90 "open/create and unlink striped directory"
+
+test_91() {
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
+ local pid1
+ local pid2
+ local duration=180
+
+ [ "$SLOW" = "yes" ] && duration=600
+ # chmod striped directory
+ (
+ cd $DIR1
+ while true; do
+ $LFS mkdir -c$MDSCOUNT $tdir > /dev/null 2>&1
+ chmod go+w $tdir > /dev/null 2>&1
+ done
+ ) &
+ pid1=$!
+ echo "start pid $pid1 to chmod striped directory"
+
+ # unlink the striped directory at the same time
+ (
+ cd $DIR2
+ while true; do
+ rm -rf $tdir > /dev/null 2>&1
+ done
+ ) &
+ pid2=$!
+ echo "start pid $pid2 to unlink striped directory"
+
+ sleep $duration
+
+ kill $pid1 $pid2
+ wait $pid1 $pid2
+
+ return 0
+}
+run_test 91 "chmod and unlink striped directory"
+
+test_92() {
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
+
+ local fd=$(free_fd)
+ local cmd="exec $fd<$DIR1/$tdir"
+ $LFS setdirstripe -c$MDSCOUNT $DIR1/$tdir || error "mkdir $tdir fails"
+ eval $cmd
+ cmd="exec $fd<&-"
+ trap "eval $cmd" EXIT
+ cd $DIR1/$tdir || error "cd $DIR1/$tdir fails"
+ rmdir ../$tdir || error "rmdir ../$tdir fails"
+
+ #define OBD_FAIL_LLITE_NO_CHECK_DEAD 0x1408
+ $LCTL set_param fail_loc=0x1408
+ mkdir $DIR2/$tdir/dir && error "create dir succeeds"
+ $LFS setdirstripe -i1 $DIR2/$tdir/remote_dir &&
+ error "create remote dir succeeds"
+ $LCTL set_param fail_loc=0
+ eval $cmd
+ return 0
+}
+run_test 92 "create remote directory under orphan directory"
+
log "cleanup: ======================================================"
+# kill and wait in each test only guarentee script finish, but command in script
+# like 'rm' 'chmod' may still be running, wait for all commands to finish
+# otherwise umount below will fail
+wait_update $HOSTNAME "fuser -m $MOUNT2" "" || true
+
[ "$(mount | grep $MOUNT2)" ] && umount $MOUNT2
complete $SECONDS
git://git.whamcloud.com / fs / lustre-release.git / blobdiff