set -e
ONLY=${ONLY:-"$*"}
-# bug number for skipped test: 3192 LU-1205 15528/3811 9977 15528/11549 18080
-ALWAYS_EXCEPT=" 14b 18c 19 28 29 35 $SANITYN_EXCEPT"
+# bug number for skipped test: LU-1205 9977/LU-7105 LU-9452
+ALWAYS_EXCEPT=" 18c 28 29 $SANITYN_EXCEPT"
# UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT!
SRCDIR=$(dirname $0)
build_test_filter
-mkdir -p $MOUNT2
-mount_client $MOUNT2
-
test_1a() {
touch $DIR1/f1
[ -f $DIR2/f1 ] || error
test_32a() { # bug 11270
local p="$TMP/$TESTSUITE-$TESTNAME.parameters"
save_lustre_params client "osc.*.lockless_truncate" > $p
- cancel_lru_locks osc
- enable_lockless_truncate 1
- rm -f $DIR1/$tfile
- lfs setstripe -c -1 $DIR1/$tfile
- dd if=/dev/zero of=$DIR1/$tfile count=$OSTCOUNT bs=$STRIPE_BYTES > \
- /dev/null 2>&1
- clear_osc_stats
-
- log "checking cached lockless truncate"
- $TRUNCATE $DIR1/$tfile 8000000
- $CHECKSTAT -s 8000000 $DIR2/$tfile || error "wrong file size"
- [ $(calc_osc_stats lockless_truncate) -ne 0 ] ||
+ cancel_lru_locks osc
+ enable_lockless_truncate 1
+ rm -f $DIR1/$tfile
+ lfs setstripe -c -1 $DIR1/$tfile
+ dd if=/dev/zero of=$DIR1/$tfile count=$OSTCOUNT bs=$STRIPE_BYTES > \
+ /dev/null 2>&1
+ clear_stats osc.*.osc_stats
+
+ log "checking cached lockless truncate"
+ $TRUNCATE $DIR1/$tfile 8000000
+ $CHECKSTAT -s 8000000 $DIR2/$tfile || error "wrong file size"
+ [ $(calc_stats osc.*.osc_stats lockless_truncate) -ne 0 ] ||
error "cached truncate isn't lockless"
- log "checking not cached lockless truncate"
- $TRUNCATE $DIR2/$tfile 5000000
- $CHECKSTAT -s 5000000 $DIR1/$tfile || error "wrong file size"
- [ $(calc_osc_stats lockless_truncate) -ne 0 ] ||
+ log "checking not cached lockless truncate"
+ $TRUNCATE $DIR2/$tfile 5000000
+ $CHECKSTAT -s 5000000 $DIR1/$tfile || error "wrong file size"
+ [ $(calc_stats osc.*.osc_stats lockless_truncate) -ne 0 ] ||
error "not cached truncate isn't lockless"
- log "disabled lockless truncate"
- enable_lockless_truncate 0
- clear_osc_stats
- $TRUNCATE $DIR2/$tfile 3000000
- $CHECKSTAT -s 3000000 $DIR1/$tfile || error "wrong file size"
- [ $(calc_osc_stats lockless_truncate) -eq 0 ] ||
- error "lockless truncate disabling failed"
- rm $DIR1/$tfile
- # restore lockless_truncate default values
- restore_lustre_params < $p
- rm -f $p
+ log "disabled lockless truncate"
+ enable_lockless_truncate 0
+ clear_stats osc.*.osc_stats
+ $TRUNCATE $DIR2/$tfile 3000000
+ $CHECKSTAT -s 3000000 $DIR1/$tfile || error "wrong file size"
+ [ $(calc_stats osc.*.osc_stats lockless_truncate) -eq 0 ] ||
+ error "lockless truncate disabling failed"
+ rm $DIR1/$tfile
+ # restore lockless_truncate default values
+ restore_lustre_params < $p
+ rm -f $p
}
run_test 32a "lockless truncate"
test_32b() { # bug 11270
- remote_ost_nodsh && skip "remote OST with nodsh" && return
+ remote_ost_nodsh && skip "remote OST with nodsh" && return
local node
local facets=$(get_facets OST)
"ldlm.namespaces.filter-*.contended_locks" >> $p
save_lustre_params $facets \
"ldlm.namespaces.filter-*.contention_seconds" >> $p
- clear_osc_stats
+ clear_stats osc.*.osc_stats
- # agressive lockless i/o settings
- for node in $(osts_nodes); do
- do_node $node 'lctl set_param -n ldlm.namespaces.filter-*.max_nolock_bytes 2000000; lctl set_param -n ldlm.namespaces.filter-*.contended_locks 0; lctl set_param -n ldlm.namespaces.filter-*.contention_seconds 60'
- done
- lctl set_param -n osc.*.contention_seconds 60
- for i in $(seq 5); do
- dd if=/dev/zero of=$DIR1/$tfile bs=4k count=1 conv=notrunc > /dev/null 2>&1
- dd if=/dev/zero of=$DIR2/$tfile bs=4k count=1 conv=notrunc > /dev/null 2>&1
- done
- [ $(calc_osc_stats lockless_write_bytes) -ne 0 ] || error "lockless i/o was not triggered"
- # disable lockless i/o (it is disabled by default)
- for node in $(osts_nodes); do
- do_node $node 'lctl set_param -n ldlm.namespaces.filter-*.max_nolock_bytes 0; lctl set_param -n ldlm.namespaces.filter-*.contended_locks 32; lctl set_param -n ldlm.namespaces.filter-*.contention_seconds 0'
- done
- # set contention_seconds to 0 at client too, otherwise Lustre still
- # remembers lock contention
- lctl set_param -n osc.*.contention_seconds 0
- clear_osc_stats
- for i in $(seq 1); do
- dd if=/dev/zero of=$DIR1/$tfile bs=4k count=1 conv=notrunc > /dev/null 2>&1
- dd if=/dev/zero of=$DIR2/$tfile bs=4k count=1 conv=notrunc > /dev/null 2>&1
- done
- [ $(calc_osc_stats lockless_write_bytes) -eq 0 ] ||
- error "lockless i/o works when disabled"
- rm -f $DIR1/$tfile
- restore_lustre_params <$p
- rm -f $p
+ # agressive lockless i/o settings
+ do_nodes $(comma_list $(osts_nodes)) \
+ "lctl set_param -n ldlm.namespaces.*.max_nolock_bytes=2000000 \
+ ldlm.namespaces.filter-*.contended_locks=0 \
+ ldlm.namespaces.filter-*.contention_seconds=60"
+ lctl set_param -n osc.*.contention_seconds=60
+ for i in {1..5}; do
+ dd if=/dev/zero of=$DIR1/$tfile bs=4k count=1 conv=notrunc > \
+ /dev/null 2>&1
+ dd if=/dev/zero of=$DIR2/$tfile bs=4k count=1 conv=notrunc > \
+ /dev/null 2>&1
+ done
+ [ $(calc_stats osc.*.osc_stats lockless_write_bytes) -ne 0 ] ||
+ error "lockless i/o was not triggered"
+ # disable lockless i/o (it is disabled by default)
+ do_nodes $(comma_list $(osts_nodes)) \
+ "lctl set_param -n ldlm.namespaces.filter-*.max_nolock_bytes=0 \
+ ldlm.namespaces.filter-*.contended_locks=32 \
+ ldlm.namespaces.filter-*.contention_seconds=0"
+ # set contention_seconds to 0 at client too, otherwise Lustre still
+ # remembers lock contention
+ lctl set_param -n osc.*.contention_seconds=0
+ clear_stats osc.*.osc_stats
+ for i in {1..1}; do
+ dd if=/dev/zero of=$DIR1/$tfile bs=4k count=1 conv=notrunc > \
+ /dev/null 2>&1
+ dd if=/dev/zero of=$DIR2/$tfile bs=4k count=1 conv=notrunc > \
+ /dev/null 2>&1
+ done
+ [ $(calc_stats osc.*.osc_stats lockless_write_bytes) -eq 0 ] ||
+ error "lockless i/o works when disabled"
+ rm -f $DIR1/$tfile
+ restore_lustre_params <$p
+ rm -f $p
}
run_test 32b "lockless i/o"
[ $fstype = ldiskfs ] && jbdold=$(print_jbd_stat)
echo "=== START createmany old: $jbdold transaction"
- local elapsed=$(do_and_time "do_nodes $CLIENT1,$CLIENT2 createmany -o $DIR1/$tdir-\\\$(hostname)-$i/f- -r $DIR2/$tdir-\\\$(hostname)-$i/f- $nfiles > /dev/null 2>&1")
+ local elapsed=$(do_and_time "do_nodes $CLIENT1,$CLIENT2 createmany -o $DIR1/$tdir-\\\$(hostname)-$i/f- -r$DIR2/$tdir-\\\$(hostname)-$i/f- $nfiles > /dev/null 2>&1")
[ $fstype = ldiskfs ] && jbdnew=$(print_jbd_stat)
[ $fstype = ldiskfs ] && jbd=$(( jbdnew - jbdold ))
echo "=== END createmany new: $jbdnew transaction : $jbd transactions nfiles $nfiles time $elapsed COS=$COS"
echo "=== START createmany old: $jbdold transaction"
local elapsed=$(do_and_time "do_nodes $CLIENT1,$CLIENT2\
createmany -o $DIR1/$tdir-\\\$(hostname)-$i/f- \
- -r $DIR2/$tdir-\\\$(hostname)-$i/f- $nfiles > \
+ -r$DIR2/$tdir-\\\$(hostname)-$i/f- $nfiles > \
/dev/null 2>&1")
jbdnew=$(print_jbd_stat)
jbd=$(( jbdnew - jbdold ))
}
run_test 39c "check truncate mtime update ======================"
+test_39d() { # LU-7310
+ touch $DIR1/$tfile
+ touch -m -d @$TEST_39_MTIME $DIR1/$tfile
+
+ local mtime1=$(stat -c %Y $DIR2/$tfile)
+ [ "$mtime1" = $TEST_39_MTIME ] ||
+ error "mtime: $mtime1, should be $TEST_39_MTIME"
+
+ # force sync write
+ # define OBD_FAIL_OSC_NO_GRANT 0x411
+ $LCTL set_param fail_loc=0x411
+
+ local d1=$(date +%s)
+ echo hello >> $DIR1/$tfile
+ local d2=$(date +%s)
+
+ $LCTL set_param fail_loc=0
+
+ cancel_lru_locks osc
+
+ local mtime2=$(stat -c %Y $DIR2/$tfile)
+ [ "$mtime2" -ge "$d1" ] && [ "$mtime2" -le "$d2" ] ||
+ error "mtime is not updated on write: $d1 <= $mtime2 <= $d2"
+}
+run_test 39d "sync write should update mtime"
+
# check that pid exists hence second operation wasn't blocked by first one
# if it is so then there is no conflict, return 0
# else second operation is conflicting with first one, return 1
local filesize
local origfile=/etc/hosts
- filesize=`stat -c %s $origfile`
+ filesize=$(stat -c %s $origfile)
# create an empty file
- $MCREATE $DIR1/$tfile
+ $MCREATE $DIR1/$tfile || error "can't create $DIR1/$tfile"
# cache layout lock on both mount point
- stat $DIR1/$tfile > /dev/null
- stat $DIR2/$tfile > /dev/null
+ stat $DIR1/$tfile > /dev/null || error "stat $DIR1/$tfile failed"
+ stat $DIR2/$tfile > /dev/null || error "stat $DIR2/$tfile failed"
# open and sleep 2 seconds then read
$MULTIOP $DIR2/$tfile o_2r${filesize}c &
sleep 1
# create the layout of testing file
- dd if=$origfile of=$DIR1/$tfile conv=notrunc > /dev/null
+ dd if=$origfile of=$DIR1/$tfile conv=notrunc > /dev/null ||
+ error "dd $DIR1/$tfile failed"
# MULTIOP proc should be able to read enough bytes and exit
sleep 2
- kill -0 $pid && error "multiop is still there"
- cmp $origfile $DIR2/$tfile || error "$MCREATE and $DIR2/$tfile differs"
+ kill -0 $pid 2> /dev/null && error "multiop is still there"
+ cmp $origfile $DIR2/$tfile || error "$origfile and $DIR2/$tfile differs"
rm -f $DIR1/$tfile
}
local tmpfile=`mktemp`
# create an empty file
- $MCREATE $DIR1/$tfile
+ $MCREATE $DIR1/$tfile || error "mcreate $DIR1/$tfile failed"
# delay glimpse so that layout has changed when glimpse finish
#define OBD_FAIL_GLIMPSE_DELAY 0x1404
sleep 1
# create layout of testing file
- dd if=/dev/zero of=$DIR1/$tfile bs=1k count=1 conv=notrunc > /dev/null
+ dd if=/dev/zero of=$DIR1/$tfile bs=1k count=1 conv=notrunc >/dev/null ||
+ error "dd $DIR1/$tfile failed"
wait $pid
- local fsize=`cat $tmpfile`
+ local fsize=$(cat $tmpfile)
[ x$fsize = x1024 ] || error "file size is $fsize, should be 1024"
[ $OSTCOUNT -ge 2 ] || { skip "need at least 2 osts"; return; }
# set default layout to have 1 stripe
- mkdir -p $DIR1/$tdir
+ mkdir $DIR1/$tdir
$LFS setstripe -c 1 $DIR1/$tdir
# create a file with empty layout
- $MCREATE $DIR1/$tdir/$tfile
+ $MCREATE $DIR1/$tdir/$tfile ||
+ error "$MCREATE $DIR1/$tdir/$tfile failed"
#define OBD_FAIL_MDS_LL_BLOCK 0x172
do_facet $SINGLEMDS $LCTL set_param fail_loc=0x172
# write something to the file, it should be blocked on fetching layout
dd if=/dev/zero of=$DIR2/$tdir/$tfile bs=1k count=1 conv=notrunc
- local cnt=$($LFS getstripe -c $DIR2/$tdir/$tfile)
- [ $cnt -eq $OSTCOUNT ] || error "have $cnt stripes, expected $OSTCOUNT"
+ local stripecnt=$($LFS getstripe -c $DIR2/$tdir/$tfile)
+ wait $pid
+
+ # lod_qos.c::min_stripe_count() allows setstripe with a default stripe
+ # count to succeed with only 3/4 of the number of stripes (rounded up),
+ # so creating striped files does not fail if an OST is offline or full
+ [ $stripecnt -ge $((OSTCOUNT - $OSTCOUNT / 4)) ] ||
+ error "layout wrong: getstripe -c $stripecnt < $OSTCOUNT * 3/4"
rm -fr $DIR1/$tdir
}
echo "==> rename vs getattr vs setxattr should not deadlock"
mkdir -p $DIR/d1/d2/d3 || error "(1) mkdir failed"
- do_facet mds $LCTL set_param fail_loc=$1
+ do_facet mds1 $LCTL set_param fail_loc=$1
mv -T $DIR/d1/d2/d3 $DIR/d1/d3 &
PID1=$!
echo "==> rename vs getattr vs open vs getattr should not deadlock"
mkdir -p $DIR/d1/d2/d3 || error "(1) mkdir failed"
- do_facet mds $LCTL set_param fail_loc=$1
+ do_facet mds1 $LCTL set_param fail_loc=$1
mv -T $DIR/d1/d2/d3 $DIR/d1/d3 &
PID1=$!
mkdir -p $DIR/d1/d2 $DIR/d3 || error "(1) mkdir failed"
#define OBD_FAIL_MDS_RENAME4 0x156
- do_facet mds $LCTL set_param fail_loc=0x80000156
+ do_facet mds1 $LCTL set_param fail_loc=0x80000156
mv -T $DIR/d1/d2 $DIR/d3/d2 &
PID1=$!
mkdir -p $DIR/d1/d2 $DIR/d3 || error "(1) mkdir failed"
#define OBD_FAIL_MDS_RENAME4 0x156
- do_facet mds $LCTL set_param fail_loc=0x80000156
+ do_facet mds1 $LCTL set_param fail_loc=0x80000156
mv -T $DIR/d1/d2 $DIR/d3/d2 &
PID1=$!
mkdir -p $DIR/d1/d2 $DIR/d3 || error "(1) mkdir failed"
#define OBD_FAIL_MDS_RENAME4 0x156
- do_facet mds $LCTL set_param fail_loc=0x156
+ do_facet mds1 $LCTL set_param fail_loc=0x156
mv -T $DIR/d1/d2 $DIR/d3/d2 &
PID1=$!
touch $DIR/f1
#define OBD_FAIL_MDS_RENAME3 0x155
- do_facet mds $LCTL set_param fail_loc=0x155
+ do_facet mds1 $LCTL set_param fail_loc=0x155
mv $DIR/f1 $DIR/$tdir &
PID1=$!
sleep 2
}
run_test 70b "remove files after calling rm_entry"
-test_71() {
+test_71a() {
local server_version=$(lustre_version_code $SINGLEMDS)
[[ $server_version -lt $(version_code 2.1.6) ]] &&
# Patch not applied to 2.2 and 2.3 branches
[[ $server_version -ge $(version_code 2.2.0) ]] &&
[[ $server_version -lt $(version_code 2.4.0) ]] &&
- skip "Need MDS version at least 2.4.0" && return
+ skip "Need MDS version earlier than 2.2.0 or at least 2.4.0" &&
+ return
checkfiemap --test ||
{ skip "checkfiemap not runnable: $?" && return; }
echo $can2
[ $can3 -eq $can4 ] || error $((can2-can1)) "cancel RPC occured."
}
-run_test 71 "correct file map just after write operation is finished"
+run_test 71a "correct file map just after write operation is finished"
+
+test_71b() {
+ local server_version=$(lustre_version_code $SINGLEMDS)
+
+ [[ $server_version -lt $(version_code 2.1.6) ]] &&
+ skip "Need MDS version at least 2.1.6" && return
+
+ # Patch not applied to 2.2 and 2.3 branches
+ [[ $server_version -ge $(version_code 2.2.0) ]] &&
+ [[ $server_version -lt $(version_code 2.4.0) ]] &&
+ skip "Need MDS version earlier than 2.2.0 or at least 2.4.0" &&
+ return
+ [[ $OSTCOUNT -ge 2 ]] || { skip "need at least 2 osts"; return; }
+
+ checkfiemap --test ||
+ { skip "error $?: checkfiemap failed" && return; }
+
+ mkdir -p $DIR1/$tdir
+
+ $LFS setstripe -c -1 $DIR1/$tdir || error "setstripe failed"
+ dd if=/dev/urandom of=$DIR1/$tdir/$tfile bs=40K count=1
+ [ "$(facet_fstype ost$(($($GETSTRIPE -i $DIR1/$tdir/$tfile) + 1)))" = \
+ "zfs" ] &&
+ skip "ORI-366/LU-1941: FIEMAP unimplemented on ZFS" && return 0
+ checkfiemap $DIR1/$tdir/$tfile 40960 || error "checkfiemap failed"
+}
+run_test 71b "check fiemap support for stripecount > 1"
test_72() {
local p="$TMP/sanityN-$TESTNAME.parameters"
error "setfattr1 failed"
getfattr -n user.attr1 $DIR2/$tfile || error "getfattr1 failed"
getfattr -n user.attr1 $DIR1/$tfile || error "getfattr2 failed"
- clear_llite_stats
+ clear_stats llite.*.stats
# PR lock should be cached by now on both clients
getfattr -n user.attr1 $DIR1/$tfile || error "getfattr3 failed"
# 2 hits for getfattr(0)+getfattr(size)
- [ $(calc_llite_stats getxattr_hits) -eq 2 ] || error "not cached in $DIR1"
+ [ $(calc_stats llite.*.stats getxattr_hits) -eq 2 ] ||
+ error "not cached in $DIR1"
getfattr -n user.attr1 $DIR2/$tfile || error "getfattr4 failed"
# 4 hits for more getfattr(0)+getfattr(size)
- [ $(calc_llite_stats getxattr_hits) -eq 4 ] || error "not cached in $DIR2"
+ [ $(calc_stats llite.*.stats getxattr_hits) -eq 4 ] ||
+ error "not cached in $DIR2"
rm -f $DIR2/$tfile
restore_lustre_params < $p
skip "Need MDS version at least 2.5.53" && return
remote_mds_nodsh && skip "remote MDS with nodsh" && return
- local fcount=2048
+ local fcount=$((MDSCOUNT * 256))
declare -a fd_list
declare -a fid_list
# drop all open locks and close any cached "open" files on the client
cancel_lru_locks mdc
- echo -n "open files "
- ulimit -n 8096
+ local open_fids_cmd="$LCTL get_param -n mdt.*.exports.'$nid'.open_files"
+ local fid_list=($(do_nodes $(comma_list $(mdts_nodes)) $open_fids_cmd))
+ local already=${#fid_list[@]}
+ for (( i = 0; i < $already; i++ )) ; do
+ log "already open[$i]: $($LFS fid2path $DIR2 ${fid_list[i]})"
+ done
+
+ echo -n "opening files: "
+ ulimit -n $((fcount + 50))
for ((i = 0; i < $fcount; i++)); do
touch $DIR/$tdir/f_$i
- local fd=$(free_fd)
- local cmd="exec $fd<$DIR/$tdir/f_$i"
- eval $cmd
+ local fd=$(free_fd ${fd_list[i]})
+ local open_cmd="exec $fd<$DIR/$tdir/f_$i"
+ eval $open_cmd
+
fd_list[i]=$fd
- echo -n "."
+
+ (( $i % 32 == 0 )) && echo -n "."
done
echo
- local get_open_fids="$LCTL get_param -n mdt.*.exports.'$nid'.open_files"
- local fid_list=($(do_nodes $(comma_list $(mdts_nodes)) $get_open_fids))
+ fid_list=($(do_nodes $(comma_list $(mdts_nodes)) $open_fids_cmd))
# Possible errors in openfiles FID list.
# 1. Missing FIDs. Check 1
# 5. Valid FID, points to some other file. Check 3
# Check 1
- [ ${#fid_list[@]} -ne $fcount ] &&
- error "${#fid_list[@]} != $fcount open files"
-
- for (( i = 0; i < $fcount; i++ )) ; do
- cmd="exec ${fd_list[i]}</dev/null"
- eval $cmd
- filename=$($LFS fid2path $DIR2 ${fid_list[i]})
+ [ ${#fid_list[@]} -ne $((fcount + already)) ] &&
+ error "${#fid_list[@]} != $fcount (+$already old) open files"
+
+ echo -n "closing files: "
+ for (( fd = 0, fid = 0; fd < $fcount; fd++, fid++ )) ; do
+ local close_cmd="exec ${fd_list[fd]}<&-"
+ eval $close_cmd
+ filename=$($LFS fid2path $DIR2 ${fid_list[fid]})
+
+ while [[ ! "$filename" =~ "$DIR2/$tdir/f_" ]]; do
+ echo "skip old open file $filename"
+ ((fid++))
+ filename=$($LFS fid2path $DIR2 ${fid_list[fid]})
+ done
# Check 2
rm --interactive=no $filename
[ $? -ne 0 ] &&
- error "Nonexisting fid ${fid_list[i]} listed."
+ error "Nonexisting fid ${fid_list[fid]} listed."
+ (( $fd % 32 == 0 )) && echo -n "."
done
+ echo
# Check 3
ls_op=$(ls $DIR2/$tdir | wc -l)
rm -rf $DIR/$tdir
}
-run_test 76 "Verify open file for 2048 files"
+run_test 76 "Verify MDT open_files listing"
nrs_write_read() {
local n=16
$LFS setstripe -c $OSTCOUNT $dir || error "setstripe to $dir failed"
chmod 777 $dir
- do_nodes $CLIENTS $myRUNAS dd if=/dev/zero of="$dir/nrs_r_$HOSTNAME"\
- bs=1M count=$n > /dev/null 2>&1
+ do_nodes $CLIENTS $myRUNAS \
+ dd if=/dev/zero of="$dir/nrs_r_$HOSTNAME" bs=1M count=$n ||
+ error "dd at 0 on client failed (1)"
for ((i = 0; i < $n; i++)); do
- do_nodes $CLIENTS $myRUNAS dd if=/dev/zero\
- of="$dir/nrs_w_$HOSTNAME" bs=1M seek=$i count=1\
- > /dev/null 2>&1 &
+ do_nodes $CLIENTS $myRUNAS dd if=/dev/zero \
+ of="$dir/nrs_w_$HOSTNAME" bs=1M seek=$i count=1 ||
+ error "dd at ${i}MB on client failed (2)" &
local pids_w[$i]=$!
done
do_nodes $CLIENTS sync;
cancel_lru_locks osc
for ((i = 0; i < $n; i++)); do
- do_nodes $CLIENTS $myRUNAS dd if="$dir/nrs_w_$HOSTNAME"\
- of=/dev/zero bs=1M seek=$i count=1 > /dev/null 2>&1 &
+ do_nodes $CLIENTS $myRUNAS dd if="$dir/nrs_w_$HOSTNAME" \
+ of=/dev/zero bs=1M seek=$i count=1 > /dev/null ||
+ error "dd at ${i}MB on client failed (3)" &
local pids_r[$i]=$!
done
cancel_lru_locks osc
}
test_77a() { #LU-3266
- do_facet $SINGLEMDS lctl set_param ost.OSS.*.nrs_policies="fifo"
+ oss=$(comma_list $(osts_nodes))
+ do_nodes $oss lctl set_param ost.OSS.*.nrs_policies="fifo"
nrs_write_read
return 0
}
run_test 77a "check FIFO NRS policy"
-
test_77b() { #LU-3266
- do_facet $SINGLEMDS lctl set_param ost.OSS.*.nrs_policies="crrn"
- do_facet $SINGLEMDS lctl set_param ost.OSS.*.nrs_crrn_quantum=1
+ oss=$(comma_list $(osts_nodes))
+
+ do_nodes $oss lctl set_param ost.OSS.*.nrs_policies="crrn" \
+ ost.OSS.*.nrs_crrn_quantum=1
echo "policy: crr-n, crrn_quantum 1"
nrs_write_read
- do_facet $SINGLEMDS lctl set_param ost.OSS.*.nrs_crrn_quantum=64
+ do_nodes $oss lctl set_param ost.OSS.*.nrs_crrn_quantum=64
echo "policy: crr-n, crrn_quantum 64"
nrs_write_read
+ # cleanup
+ do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="fifo"
return 0
}
run_test 77b "check CRR-N NRS policy"
orr_trr() {
local policy=$1
- for i in $(seq 1 $OSTCOUNT)
- do
- do_facet ost"$i" lctl set_param \
- ost.OSS.ost_io.nrs_policies=$policy
- do_facet ost"$i" lctl set_param \
- ost.OSS.*.nrs_"$policy"_quantum=1
- do_facet ost"$i" lctl set_param \
- ost.OSS.*.nrs_"$policy"_offset_type="physical"
- do_facet ost"$i" lctl set_param \
- ost.OSS.*.nrs_"$policy"_supported="reads"
- done
+ oss=$(comma_list $(osts_nodes))
+
+ do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies=$policy \
+ ost.OSS.*.nrs_"$policy"_quantum=1 \
+ ost.OSS.*.nrs_"$policy"_offset_type="physical" \
+ ost.OSS.*.nrs_"$policy"_supported="reads"
- echo "policy: $policy, ${policy}_quantum 1, ${policy}_offset_type \
- physical, ${policy}_supported reads"
+ echo "policy: $policy, ${policy}_quantum 1, ${policy}_offset_type physical, ${policy}_supported reads"
nrs_write_read
- for i in $(seq 1 $OSTCOUNT)
- do
- do_facet ost"$i" lctl set_param \
- ost.OSS.*.nrs_${policy}_supported="writes"
- do_facet ost"$i" lctl set_param \
- ost.OSS.*.nrs_${policy}_quantum=64
- done
- echo "policy: $policy, ${policy}_quantum 64, \
- ${policy}_offset_type physical, ${policy}_supported writes"
+ do_nodes $oss lctl set_param ost.OSS.*.nrs_${policy}_supported="writes" \
+ ost.OSS.*.nrs_${policy}_quantum=64
+
+ echo "policy: $policy, ${policy}_quantum 64, ${policy}_offset_type physical, ${policy}_supported writes"
nrs_write_read
- for i in $(seq 1 $OSTCOUNT)
- do
- do_facet ost"$i" lctl set_param \
- ost.OSS.*.nrs_${policy}_supported="reads_and_writes"
- do_facet ost"$i" lctl set_param \
- ost.OSS.*.nrs_${policy}_offset_type="logical"
- done
- echo "policy: $policy, ${policy}_quantum 64, \
- ${policy}_offset_type logical, ${policy}_supported reads_and_writes"
+ do_nodes $oss lctl set_param ost.OSS.*.nrs_${policy}_supported="reads_and_writes" \
+ ost.OSS.*.nrs_${policy}_offset_type="logical"
+ echo "policy: $policy, ${policy}_quantum 64, ${policy}_offset_type logical, ${policy}_supported reads_and_writes"
nrs_write_read
+ # cleanup
+ do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="fifo"
return 0
}
do_facet $facet lctl set_param \
ost.OSS.ost_io.nrs_tbf_rule="$*"
[ $? -ne 0 ] &&
- error "failed to operate on TBF rules"
+ error "failed to run operate '$*' on TBF rules"
+}
+
+tbf_verify() {
+ local dir=$DIR/$tdir
+ local client1=${CLIENT1:-$(hostname)}
+ local myRUNAS="$3"
+
+ mkdir $dir || error "mkdir $dir failed"
+ $LFS setstripe -c 1 $dir || error "setstripe to $dir failed"
+ chmod 777 $dir
+
+ echo "Limited write rate: $1, read rate: $2"
+ echo "Verify the write rate is under TBF control"
+ local rate=$(do_node $client1 $myRUNAS dd if=/dev/zero of=$dir/tbf \
+ bs=1M count=100 oflag=direct 2>&1 | awk '/bytes/ {print $8}')
+ echo "Write speed is $rate"
+
+ # verify the write rate does not exceed 110% of TBF limited rate
+ [ $(bc <<< "$rate < 1.1 * $1") -eq 1 ] ||
+ error_ignore LU-9140 "The write rate ($rate) exceeds 110% of preset rate ($1)"
+
+ cancel_lru_locks osc
+
+ echo "Verify the read rate is under TBF control"
+ rate=$(do_node $client1 $myRUNAS dd if=$dir/tbf of=/dev/null \
+ bs=1M count=100 iflag=direct 2>&1 | awk '/bytes/ {print $8}')
+ echo "Read speed is $rate"
+
+ # verify the read rate does not exceed 110% of TBF limited rate
+ [ $(bc <<< "$rate < 1.1 * $2") -eq 1 ] ||
+ error_ignore LU-9140 "The read rate ($rate) exceeds 110% of preset rate ($2)"
+
+ cancel_lru_locks osc
+ rm -rf $dir || error "rm -rf $dir failed"
}
test_77e() {
- for i in $(seq 1 $OSTCOUNT)
- do
- do_facet ost"$i" lctl set_param \
- ost.OSS.ost_io.nrs_policies="tbf\ nid"
- [ $? -ne 0 ] &&
- error "failed to set TBF policy"
- done
+ local server_version=$(lustre_version_code ost1)
+ [[ $server_version -ge $(version_code 2.7.58) ]] ||
+ { skip "Need server version newer than 2.7.57"; return 0; }
+
+ oss=$(comma_list $(osts_nodes))
+
+ do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="tbf\ nid"
+ [ $? -ne 0 ] && error "failed to set TBF policy"
local idis
local rateis
nrs_write_read
# Cleanup the TBF 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
+ do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="fifo"
+ [ $? -ne 0 ] && error "failed to set policy back to fifo"
nrs_write_read
return 0
}
run_test 77e "check TBF NID nrs policy"
test_77f() {
+ local server_version=$(lustre_version_code ost1)
+ [[ $server_version -ge $(version_code 2.7.58) ]] ||
+ { skip "Need server version newer than 2.7.57"; return 0; }
+
+ oss=$(comma_list $(osts_nodes))
+
# Configure jobid_var
local saved_jobid_var=$($LCTL get_param -n jobid_var)
if [ $saved_jobid_var != procname_uid ]; then
"$FSNAME.sys.jobid_var" procname_uid
fi
- for i in $(seq 1 $OSTCOUNT)
- do
- do_facet ost"$i" lctl set_param \
- ost.OSS.ost_io.nrs_policies="tbf\ jobid"
- [ $? -ne 0 ] &&
- error "failed to set TBF policy"
- done
+ do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="tbf\ jobid"
+ [ $? -ne 0 ] && error "failed to set TBF policy"
local idis
local rateis
nrs_write_read "$RUNAS"
# Cleanup the TBF 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
+ do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="fifo"
+ [ $? -ne 0 ] && error "failed to set policy back to fifo"
nrs_write_read "$RUNAS"
local current_jobid_var=$($LCTL get_param -n jobid_var)
run_test 77f "check TBF JobID nrs policy"
test_77g() {
- for i in $(seq 1 $OSTCOUNT)
- do
- do_facet ost"$i" lctl set_param \
- ost.OSS.ost_io.nrs_policies="tbf\ nid"
- [ $? -ne 0 ] &&
- error "failed to set TBF policy"
- done
+ local server_version=$(lustre_version_code ost1)
+ [[ $server_version -ge $(version_code 2.7.58) ]] ||
+ { skip "Need server version newer than 2.7.57"; return 0; }
- for i in $(seq 1 $OSTCOUNT)
- do
- do_facet ost"$i" lctl set_param \
- ost.OSS.ost_io.nrs_policies="tbf\ jobid"
- [ $? -ne 0 ] &&
- error "failed to set TBF policy"
- done
+ oss=$(comma_list $(osts_nodes))
+
+ do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="tbf\ nid"
+ [ $? -ne 0 ] && error "failed to set TBF policy"
+
+ do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="tbf\ jobid"
+ [ $? -ne 0 ] && error "failed to set TBF policy"
local idis
local rateis
tbf_rule_operate ost1 "start\ dd_runas\ ${idis}{dd.$RUNAS_ID}\ ${rateis}50"
# Cleanup the TBF 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
+ do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="fifo"
+ [ $? -ne 0 ] && error "failed to set policy back to fifo"
return 0
}
run_test 77g "Change TBF type directly"
[ $? -eq 0 ] && error "should return error"
do_facet ost1 lctl set_param \
- ost.OSS.ost_io.nrs_policies="tbf\ reg"
- [ $? -eq 0 ] && error "should return error"
-
- do_facet ost1 lctl set_param \
ost.OSS.ost_io.nrs_policies="tbf\ reg\ abc"
[ $? -eq 0 ] && error "should return error"
}
run_test 77h "Wrong policy name should report error, not LBUG"
-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"
+tbf_rule_check()
+{
+ local facet=$1
+ local expected=$2
+ local error_message=$3
+ local rule_number=0
+ for rule in $expected; do
+ rule_number=$((rule_number + 1))
done
+ local stop_line=$(($rule_number + 3))
+ local awk_command="awk 'NR >= 4 && NR <= $stop_line {print \$1}'"
+
+ local output=$(do_facet $facet lctl get_param \
+ ost.OSS.ost_io.nrs_tbf_rule |
+ eval $awk_command |
+ tr "\n" " " |
+ sed 's/[ ]*$//')
+ if [ "$output" != "$expected" ]; then
+ error "$error_message, expected '$expected', got '$output'"
+ fi
+}
+
+test_77i() {
+ [ $(lustre_version_code ost1) -ge $(version_code 2.8.55) ] ||
+ { skip "Need OST version at least 2.8.55"; return 0; }
- # 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"
+ ost.OSS.ost_io.nrs_policies="tbf\ jobid"
[ $? -ne 0 ] &&
- error "failed to set policy back to fifo"
+ error "failed to set TBF policy"
done
+
+ tbf_rule_check ost1 "default" "error before inserting any rule"
+
+ tbf_rule_operate ost1 "start\ before\ jobid={jobid}\ rate=1000"
+ tbf_rule_check ost1 "before default" \
+ "error when inserting rule 'before'"
+
+ tbf_rule_operate ost1 "start\ after\ jobid={jobid}\ rate=1000\ rank=default"
+ tbf_rule_check ost1 "before after default" \
+ "error when inserting rule 'after'"
+
+ tbf_rule_operate ost1 "start\ target\ jobid={jobid}\ rate=1000\ rank=after"
+ tbf_rule_check ost1 "before target after default" \
+ "error when inserting rule 'target'"
+
+ echo "Move before itself"
+ tbf_rule_operate ost1 "change\ target\ rank=target"
+ tbf_rule_check ost1 "before target after default" \
+ "error when moving before itself"
+
+ echo "Move to higher rank"
+ tbf_rule_operate ost1 "change\ target\ rank=before"
+ tbf_rule_check ost1 "target before after default" \
+ "error when moving to higher rank"
+
+ echo "Move to lower rank"
+ tbf_rule_operate ost1 "change\ target\ rank=after"
+ tbf_rule_check ost1 "before target after default" \
+ "error when moving to lower rank"
+
+ echo "Move before default"
+ tbf_rule_operate ost1 "change\ target\ rank=default"
+ tbf_rule_check ost1 "before after target default" \
+ error "error when moving before default"
+
+ # Cleanup the TBF policy
+ do_nodes $(comma_list $(osts_nodes)) \
+ $LCTL set_param ost.OSS.ost_io.nrs_policies=fifo
+ return 0
+}
+run_test 77i "Change rank of TBF rule"
+
+test_77j() {
+ local idis
+ local rateis
+ if [ $(lustre_version_code ost1) -ge $(version_code 2.8.60) ]; then
+ idis="opcode="
+ rateis="rate="
+ fi
+
+ do_nodes $(comma_list $(osts_nodes)) \
+ lctl set_param jobid_var=procname_uid \
+ ost.OSS.ost_io.nrs_policies="tbf\ opcode" \
+ ost.OSS.ost_io.nrs_tbf_rule="start\ ost_r\ ${idis}{ost_read}\ ${rateis}5" \
+ ost.OSS.ost_io.nrs_tbf_rule="start\ ost_w\ ${idis}{ost_write}\ ${rateis}20"
+ [ $? -ne 0 ] && error "failed to set TBF OPCode policy"
+
+ nrs_write_read
+ tbf_verify 20 5
+
+ do_nodes $(comma_list $(osts_nodes)) \
+ lctl set_param ost.OSS.ost_io.nrs_tbf_rule="stop\ ost_r" \
+ ost.OSS.ost_io.nrs_tbf_rule="stop\ ost_w" \
+ ost.OSS.ost_io.nrs_policies="fifo"
+
+ # sleep 3 seconds to wait the tbf policy stop completely,
+ # or the next test case is possible get -EAGAIN when
+ # setting the tbf policy
+ sleep 3
+}
+run_test 77j "check TBF-OPCode NRS policy"
+
+test_77k() {
+ [[ $(lustre_version_code ost1) -ge $(version_code 2.9.53) ]] ||
+ { skip "Need OST version at least 2.9.53"; return 0; }
+
+ do_nodes $(comma_list $(osts_nodes)) \
+ lctl set_param ost.OSS.ost_io.nrs_policies="tbf" \
+ ost.OSS.ost_io.nrs_tbf_rule="start\ ext_w\ jobid={dd.$RUNAS_ID}\&opcode={ost_write}\ rate=20" \
+ ost.OSS.ost_io.nrs_tbf_rule="start\ ext_r\ jobid={dd.$RUNAS_ID}\&opcode={ost_read}\ rate=10"
+
+ nrs_write_read "$RUNAS"
+ tbf_verify 20 10 "$RUNAS"
+
+ local address=$(comma_list "$(host_nids_address $CLIENTS $NETTYPE)")
+ local client_nids=$(nids_list $address "\\")
+ do_nodes $(comma_list $(osts_nodes)) \
+ lctl set_param ost.OSS.ost_io.nrs_tbf_rule="stop\ ext_w" \
+ ost.OSS.ost_io.nrs_tbf_rule="stop\ ext_r" \
+ ost.OSS.ost_io.nrs_tbf_rule="start\ ext_w\ nid={0@lo\ $client_nids}\&opcode={ost_write}\ rate=20" \
+ ost.OSS.ost_io.nrs_tbf_rule="start\ ext_r\ nid={0@lo\ $client_nids}\&opcode={ost_read}\ rate=10"
+
+ nrs_write_read
+ tbf_verify 20 10
+
+ do_nodes $(comma_list $(osts_nodes)) \
+ lctl set_param ost.OSS.ost_io.nrs_tbf_rule="stop\ ext_w" \
+ ost.OSS.ost_io.nrs_tbf_rule="stop\ ext_r" \
+ ost.OSS.ost_io.nrs_tbf_rule="start\ ext\ nid={0@lo\ $client_nids}\&jobid={dd.$RUNAS_ID}\ rate=20"
+
+ nrs_write_read "$RUNAS"
+ tbf_verify 20 20 "$RUNAS"
+
+ do_nodes $(comma_list $(osts_nodes)) \
+ lctl set_param ost.OSS.ost_io.nrs_tbf_rule="stop\ ext" \
+ ost.OSS.ost_io.nrs_tbf_rule="start\ ext_a\ jobid={dd.$RUNAS_ID},opcode={ost_write}\ rate=20" \
+ ost.OSS.ost_io.nrs_tbf_rule="start\ ext_b\ jobid={dd.$RUNAS_ID},opcode={ost_read}\ rate=10"
+
+ nrs_write_read "$RUNAS"
+ # with parameter "RUNAS", it will match the latest rule
+ # "ext_b" first, so the limited write rate is 10.
+ tbf_verify 10 10 "$RUNAS"
+ tbf_verify 20 10
+
+ do_nodes $(comma_list $(osts_nodes)) \
+ lctl set_param ost.OSS.ost_io.nrs_tbf_rule="stop\ ext_a" \
+ ost.OSS.ost_io.nrs_tbf_rule="stop\ ext_b" \
+ ost.OSS.ost_io.nrs_policies="fifo"
+
+ sleep 3
+}
+run_test 77k "check the extended TBF policy with NID/JobID/OPCode expression"
+
+test_77l() {
+ if [ $(lustre_version_code ost1) -lt $(version_code 2.9.54) ]; then
+ skip "Need OST version at least 2.9.54"
+ return 0
+ fi
+
+ local dir=$DIR/$tdir
+
+ mkdir $dir || error "mkdir $dir failed"
+ $LFS setstripe -c $OSTCOUNT $dir || error "setstripe to $dir failed"
+ chmod 777 $dir
+
+ local nodes=$(comma_list $(osts_nodes))
+ do_nodes $nodes lctl set_param ost.OSS.ost_io.nrs_policies=delay \
+ ost.OSS.ost_io.nrs_delay_min=4 \
+ ost.OSS.ost_io.nrs_delay_max=4 \
+ ost.OSS.ost_io.nrs_delay_pct=100
+ [ $? -ne 0 ] && error "Failed to set delay policy"
+
+ local start=$SECONDS
+ do_nodes "${SINGLECLIENT:-$HOSTNAME}" "$RUNAS" \
+ dd if=/dev/zero of="$dir/nrs_delay_$HOSTNAME" bs=1M count=1 \
+ oflag=direct conv=fdatasync ||
+ { do_nodes $nodes lctl set_param ost.OSS.ost_io.nrs_policies="fifo";
+ error "dd on client failed (1)"; }
+ local elapsed=$((SECONDS - start))
+
+ # NRS delay doesn't do sub-second timing, so a request enqueued at
+ # 0.9 seconds can be dequeued at 4.0
+ [ $elapsed -lt 3 ] &&
+ { do_nodes $nodes lctl set_param ost.OSS.ost_io.nrs_policies="fifo";
+ error "Single 1M write should take at least 3 seconds"; }
+
+ start=$SECONDS
+ do_nodes "${SINGLECLIENT:-$HOSTNAME}" "$RUNAS" \
+ dd if=/dev/zero of="$dir/nrs_delay_$HOSTNAME" bs=1M count=10 \
+ oflag=direct conv=fdatasync ||
+ { do_nodes $nodes lctl set_param ost.OSS.ost_io.nrs_policies="fifo";
+ error "dd on client failed (2)"; }
+ elapsed=$((SECONDS - start))
+
+ [ $elapsed -lt 30 ] &&
+ { do_nodes $nodes lctl set_param ost.OSS.ost_io.nrs_policies="fifo";
+ error "Ten 1M writes should take at least 30 seconds"; }
+
+ do_nodes $nodes lctl set_param ost.OSS.ost_io.nrs_policies="fifo"
+ [ $? -ne 0 ] && error "failed to set policy back to fifo"
+
+ return 0
+}
+run_test 77l "check NRS Delay slows write RPC processing"
+
+test_78() { #LU-6673
+ local server_version=$(lustre_version_code ost1)
+ [[ $server_version -ge $(version_code 2.7.58) ]] ||
+ { skip "Need server version newer than 2.7.57"; return 0; }
+
+ local rc
+
+ oss=$(comma_list $(osts_nodes))
+ do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="orr" &
+ do_nodes $oss lctl set_param ost.OSS.*.nrs_orr_quantum=1
+ rc=$?
+ # 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"
+
+ # Cleanup the ORR policy
+ do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="fifo"
+ [ $? -ne 0 ] && error "failed to set policy back to fifo"
return 0
}
run_test 78 "Enable policy and specify tunings right away"
+test_79() {
+ remote_mds_nodsh && skip "remote MDS with nodsh" && return
+ test_mkdir -p $DIR/$tdir
+
+ # Prevent interference from layout intent RPCs due to
+ # asynchronous writeback. These will be tested in 130c below.
+ do_nodes ${CLIENTS:-$HOSTNAME} sync
+
+ setfattr -n trusted.name1 -v value1 $DIR/$tdir ||
+ error "setfattr -n trusted.name1=value1 $DIR/$tdir failed"
+
+#define OBD_FAIL_MDS_INTENT_DELAY 0x160
+ local mdtidx=$($LFS getstripe -M $DIR/$tdir)
+ local facet=mds$((mdtidx + 1))
+ stat $DIR/$tdir
+ set_nodes_failloc $(facet_active_host $facet) 0x80000160
+ getfattr -n trusted.name1 $DIR/$tdir 2> /dev/null &
+ local pid=$!
+ sleep 2
+
+#define OBD_FAIL_MDS_GETXATTR_PACK 0x131
+ set_nodes_failloc $(facet_active_host $facet) 0x80000131
+
+ wait $pid
+ return 0
+}
+run_test 79 "xattr: intent error"
+
test_80a() {
[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
local MDTIDX=1
}
run_test 92 "create remote directory under orphan directory"
+test_93() {
+ local rc1=0
+ local rc2=0
+ local old_rr
+
+ mkdir -p $DIR1/$tfile-1/
+ mkdir -p $DIR2/$tfile-2/
+ local old_rr=$(do_facet $SINGLEMDS lctl get_param -n \
+ 'lod.lustre-MDT*/qos_threshold_rr' | sed -e 's/%//')
+ do_facet $SINGLEMDS lctl set_param -n \
+ 'lod.lustre-MDT*/qos_threshold_rr' 100
+ #define OBD_FAIL_MDS_LOV_CREATE_RACE 0x163
+ do_facet $SINGLEMDS "lctl set_param fail_loc=0x00000163"
+
+ $SETSTRIPE -c -1 $DIR1/$tfile-1/file1 &
+ local PID1=$!
+ sleep 1
+ $SETSTRIPE -c -1 $DIR2/$tfile-2/file2 &
+ local PID2=$!
+ wait $PID2
+ wait $PID1
+ do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
+ do_facet $SINGLEMDS "lctl set_param -n \
+ 'lod.lustre-MDT*/qos_threshold_rr' $old_rr"
+
+ $GETSTRIPE $DIR1/$tfile-1/file1
+ rc1=$($GETSTRIPE -q $DIR1/$tfile-1/file1 |
+ awk '{if (/[0-9]/) print $1 }' | sort | uniq -d | wc -l)
+ $GETSTRIPE $DIR2/$tfile-2/file2
+ rc2=$($GETSTRIPE -q $DIR2/$tfile-2/file2 |
+ awk '{if (/[0-9]/) print $1 }' | sort | uniq -d | wc -l)
+ echo "rc1=$rc1 and rc2=$rc2 "
+ [ $rc1 -eq 0 ] && [ $rc2 -eq 0 ] ||
+ error "object allocate on same ost detected"
+}
+run_test 93 "alloc_rr should not allocate on same ost"
+
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 $MOUNT2
+[ "$(mount | grep $MOUNT2)" ] && wait_update $HOSTNAME "fuser -m $MOUNT2" "" ||
+ true
complete $SECONDS
rm -f $SAMPLE_FILE