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 $p
}
run_test 32b "lockless i/o"
}
run_test 70b "remove files after calling rm_entry"
-test_71() {
+test_71a() {
local server_version=$(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 MDS open_files tracking for 2048 files"
+run_test 76 "Verify MDT open_files listing"
nrs_write_read() {
local n=16
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 ||
- error "dd at ${i}MB on client failed (3)"
+ error "dd at ${i}MB on client failed (3)" &
local pids_r[$i]=$!
done
cancel_lru_locks osc
tbf_verify() {
local dir=$DIR/$tdir
- local client1=${CLIENT1:-`hostname`}
+ local client1=${CLIENT1:-$(hostname)}
local myRUNAS="$3"
mkdir $dir || error "mkdir $dir failed"
# verify the write rate does not exceed 110% of TBF limited rate
[ $(bc <<< "$rate < 1.1 * $1") -eq 1 ] ||
- error "The write rate ($rate) exceeds 110% of preset rate ($1)"
+ error_ignore LU-9140 "The write rate ($rate) exceeds 110% of preset rate ($1)"
cancel_lru_locks osc
# verify the read rate does not exceed 110% of TBF limited rate
[ $(bc <<< "$rate < 1.1 * $2") -eq 1 ] ||
- error "The read rate ($rate) exceeds 110% of preset rate ($2)"
+ error_ignore LU-9140 "The read rate ($rate) exceeds 110% of preset rate ($2)"
+ cancel_lru_locks osc
rm -rf $dir || error "rm -rf $dir failed"
}
[ $? -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"
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
}
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) ]] ||
# 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