-#!/bin/sh
+#!/bin/bash
set -e
+#set -v
#
# This test needs to be run on the client
LUSTRE=${LUSTRE:-`dirname $0`/..}
. $LUSTRE/tests/test-framework.sh
-
init_test_env $@
+. ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
-. ${CONFIG:=$LUSTRE/tests/cfg/lmv.sh}
-
-build_test_filter
-
-assert_env MDSCOUNT
# Skip these tests
-ALWAYS_EXCEPT=""
+# bug number: 4176
+ALWAYS_EXCEPT="39 $REPLAY_SINGLE_EXCEPT"
gen_config() {
rm -f $XMLCONFIG
-
- if [ "$MDSCOUNT" -gt 1 ]; then
- add_lmv lmv1_svc
- for mds in `mds_list`; do
- MDSDEV=$TMP/${mds}-`hostname`
- add_mds $mds --dev $MDSDEV --size $MDSSIZE --lmv lmv1_svc
- done
- add_lov_to_lmv lov1 lmv1_svc --stripe_sz $STRIPE_BYTES \
- --stripe_cnt $STRIPES_PER_OBJ --stripe_pattern 0
- MDS=lmv1
- else
- add_mds mds1 --dev $MDSDEV --size $MDSSIZE
- if [ ! -z "$mds1failover_HOST" ]; then
- add_mdsfailover mds1 --dev $MDSDEV --size $MDSSIZE
- fi
- add_lov lov1 mds1 --stripe_sz $STRIPE_BYTES \
- --stripe_cnt $STRIPES_PER_OBJ --stripe_pattern 0
- MDS=mds1_svc
+ add_mds mds --dev $MDSDEV --size $MDSSIZE
+ if [ ! -z "$mdsfailover_HOST" ]; then
+ add_mdsfailover mds --dev $MDSDEV --size $MDSSIZE
fi
+ add_lov lov1 mds --stripe_sz $STRIPE_BYTES \
+ --stripe_cnt $STRIPES_PER_OBJ --stripe_pattern 0
add_ost ost --lov lov1 --dev $OSTDEV --size $OSTSIZE
add_ost ost2 --lov lov1 --dev ${OSTDEV}-2 --size $OSTSIZE
- add_client client $MDS --lov lov1 --path $MOUNT
+ add_client client mds --lov lov1 --path $MOUNT
}
build_test_filter
-cleanup() {
- # make sure we are using the primary MDS, so the config log will
- # be able to clean up properly.
- activemds=`facet_active mds1`
- if [ $activemds != "mds1" ]; then
- fail mds1
- fi
- zconf_umount `hostname` $MOUNT
- for mds in `mds_list`; do
- stop $mds ${FORCE} $MDSLCONFARGS
- done
- stop ost2 ${FORCE} --dump cleanup.log
- stop ost ${FORCE} --dump cleanup.log
-}
+SETUP=${SETUP:-"setup"}
+CLEANUP=${CLEANUP:-"cleanupall"}
if [ "$ONLY" == "cleanup" ]; then
- sysctl -w portals.debug=0 || true
- cleanup
- exit
+ sysctl -w lnet.debug=0 || true
+ $CLEANUP
+ exit 0
fi
-SETUP=${SETUP:-"setup"}
-CLEANUP=${CLEANUP:-"cleanup"}
-
setup() {
- gen_config
-
- start ost --reformat $OSTLCONFARGS
- start ost2 --reformat $OSTLCONFARGS
- [ "$DAEMONFILE" ] && $LCTL debug_daemon start $DAEMONFILE $DAEMONSIZE
- for mds in `mds_list`; do
- start $mds --reformat $MDSLCONFARGS
- done
- grep " $MOUNT " /proc/mounts || zconf_mount `hostname` $MOUNT
+ formatall
+ setupall
}
$SETUP
mkdir -p $DIR
test_0() {
- replay_barrier mds1
- fail mds1
+ replay_barrier mds
+ fail mds
}
run_test 0 "empty replay"
test_0b() {
# this test attempts to trigger a race in the precreation code,
# and must run before any other objects are created on the filesystem
- fail ost
+ fail ost1
createmany -o $DIR/$tfile 20 || return 1
unlinkmany $DIR/$tfile 20 || return 2
}
run_test 0b "ensure object created after recover exists. (3284)"
test_1() {
- replay_barrier mds1
+ replay_barrier mds
mcreate $DIR/$tfile
- fail mds1
+ fail mds
$CHECKSTAT -t file $DIR/$tfile || return 1
rm $DIR/$tfile
}
run_test 1 "simple create"
+test_1a() {
+ do_facet ost1 "sysctl -w lustre.fail_loc=0"
+
+ rm -fr $DIR/$tfile
+ local old_last_id=`cat $LPROC/obdfilter/*/last_id`
+ touch -o $DIR/$tfile 1
+ sync
+ local new_last_id=`cat $LPROC/obdfilter/*/last_id`
+
+ test "$old_last_id" = "$new_last_id" || {
+ echo "OST object create is caused by MDS"
+ return 1
+ }
+
+ old_last_id=`cat $LPROC/obdfilter/*/last_id`
+ echo "data" > $DIR/$tfile
+ sync
+ new_last_id=`cat $LPROC/obdfilter/*/last_id`
+ test "$old_last_id" = "$new_last_id "&& {
+ echo "CROW does not work on write"
+ return 1
+ }
+
+ rm -fr $DIR/$tfile
+
+#define OBD_FAIL_OST_CROW_EIO | OBD_FAIL_ONCE
+ do_facet ost1 "sysctl -w lustre.fail_loc=0x80000801"
+
+ rm -fr $DIR/1a1
+ old_last_id=`cat $LPROC/obdfilter/*/last_id`
+ echo "data" > $DIR/1a1
+ sync
+ new_last_id=`cat $LPROC/obdfilter/*/last_id`
+ test "$old_last_id" = "$new_last_id" || {
+ echo "CROW does work with fail_loc=0x80000801"
+ return 1
+ }
+
+ rm -fr $DIR/1a1
+
+ do_facet ost1 "sysctl -w lustre.fail_loc=0"
+}
+#CROW run_test 1a "CROW object create (check OST last_id)"
+
test_2a() {
- replay_barrier mds1
+ replay_barrier mds
touch $DIR/$tfile
- fail mds1
+ fail mds
$CHECKSTAT -t file $DIR/$tfile || return 1
rm $DIR/$tfile
}
test_2b() {
./mcreate $DIR/$tfile
- replay_barrier mds1
+ replay_barrier mds
touch $DIR/$tfile
- fail mds1
+ fail mds
$CHECKSTAT -t file $DIR/$tfile || return 1
rm $DIR/$tfile
}
run_test 2b "touch"
test_3a() {
- replay_barrier mds1
+ replay_barrier mds
mcreate $DIR/$tfile
o_directory $DIR/$tfile
- fail mds1
+ fail mds
$CHECKSTAT -t file $DIR/$tfile || return 2
rm $DIR/$tfile
}
run_test 3a "replay failed open(O_DIRECTORY)"
test_3b() {
- replay_barrier mds1
+ replay_barrier mds
#define OBD_FAIL_MDS_OPEN_PACK | OBD_FAIL_ONCE
do_facet mds "sysctl -w lustre.fail_loc=0x80000114"
touch $DIR/$tfile
do_facet mds "sysctl -w lustre.fail_loc=0"
- fail mds1
+ fail mds
$CHECKSTAT -t file $DIR/$tfile && return 2
return 0
}
run_test 3b "replay failed open -ENOMEM"
test_3c() {
- replay_barrier mds1
+ replay_barrier mds
#define OBD_FAIL_MDS_ALLOC_OBDO | OBD_FAIL_ONCE
do_facet mds "sysctl -w lustre.fail_loc=0x80000128"
touch $DIR/$tfile
do_facet mds "sysctl -w lustre.fail_loc=0"
- fail mds1
+ fail mds
$CHECKSTAT -t file $DIR/$tfile && return 2
return 0
run_test 3c "replay failed open -ENOMEM"
test_4() {
- replay_barrier mds1
+ replay_barrier mds
for i in `seq 10`; do
echo "tag-$i" > $DIR/$tfile-$i
done
- fail mds1
+ fail mds
for i in `seq 10`; do
grep -q "tag-$i" $DIR/$tfile-$i || error "$tfile-$i"
done
run_test 4 "|x| 10 open(O_CREAT)s"
test_4b() {
- replay_barrier mds1
+ replay_barrier mds
rm -rf $DIR/$tfile-*
- fail mds1
+ fail mds
$CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
}
run_test 4b "|x| rm 10 files"
# The idea is to get past the first block of precreated files on both
# osts, and then replay.
test_5() {
- replay_barrier mds1
+ replay_barrier mds
for i in `seq 220`; do
echo "tag-$i" > $DIR/$tfile-$i
done
- fail mds1
+ fail mds
for i in `seq 220`; do
grep -q "tag-$i" $DIR/$tfile-$i || error "f1c-$i"
done
test_6() {
- replay_barrier mds1
+ replay_barrier mds
mkdir $DIR/$tdir
mcreate $DIR/$tdir/$tfile
- fail mds1
+ fail mds
$CHECKSTAT -t dir $DIR/$tdir || return 1
$CHECKSTAT -t file $DIR/$tdir/$tfile || return 2
sleep 2
run_test 6 "mkdir + contained create"
test_6b() {
- replay_barrier mds1
+ replay_barrier mds
rm -rf $DIR/$tdir
- fail mds1
+ fail mds
$CHECKSTAT -t dir $DIR/$tdir && return 1 || true
}
run_test 6b "|X| rmdir"
test_7() {
mkdir $DIR/$tdir
- replay_barrier mds1
+ replay_barrier mds
mcreate $DIR/$tdir/$tfile
- fail mds1
+ fail mds
$CHECKSTAT -t dir $DIR/$tdir || return 1
$CHECKSTAT -t file $DIR/$tdir/$tfile || return 2
rm -fr $DIR/$tdir
run_test 7 "mkdir |X| contained create"
test_8() {
- replay_barrier mds1
+ replay_barrier mds
multiop $DIR/$tfile mo_c &
MULTIPID=$!
sleep 1
- fail mds1
+ fail mds
ls $DIR/$tfile
$CHECKSTAT -t file $DIR/$tfile || return 1
kill -USR1 $MULTIPID || return 2
run_test 8 "creat open |X| close"
test_9() {
- replay_barrier mds1
+ replay_barrier mds
mcreate $DIR/$tfile
local old_inum=`ls -i $DIR/$tfile | awk '{print $1}'`
- fail mds1
+ fail mds
local new_inum=`ls -i $DIR/$tfile | awk '{print $1}'`
echo " old_inum == $old_inum, new_inum == $new_inum"
test_10() {
mcreate $DIR/$tfile
- replay_barrier mds1
+ replay_barrier mds
mv $DIR/$tfile $DIR/$tfile-2
rm -f $DIR/$tfile
- fail mds1
-
+ fail mds
$CHECKSTAT $DIR/$tfile && return 1
- $CHECKSTAT $DIR/$tfile-2 || return 2
+ $CHECKSTAT $DIR/$tfile-2 ||return 2
rm $DIR/$tfile-2
return 0
}
mcreate $DIR/$tfile
echo "old" > $DIR/$tfile
mv $DIR/$tfile $DIR/$tfile-2
- replay_barrier mds1
+ replay_barrier mds
echo "new" > $DIR/$tfile
grep new $DIR/$tfile
grep old $DIR/$tfile-2
- fail mds1
+ fail mds
grep new $DIR/$tfile || return 1
grep old $DIR/$tfile-2 || return 2
}
# give multiop a chance to open
sleep 1
rm -f $DIR/$tfile
- replay_barrier mds1
+ replay_barrier mds
kill -USR1 $pid
wait $pid || return 1
- fail mds1
+ fail mds
[ -e $DIR/$tfile ] && return 2
return 0
}
sleep 1
chmod 0 $DIR/$tfile
$CHECKSTAT -p 0 $DIR/$tfile
- replay_barrier mds1
- fail mds1
+ replay_barrier mds
+ fail mds
kill -USR1 $pid
wait $pid || return 1
# give multiop a chance to open
sleep 1
rm -f $DIR/$tfile
- replay_barrier mds1
+ replay_barrier mds
kill -USR1 $pid || return 1
wait $pid || return 2
- fail mds1
+ fail mds
[ -e $DIR/$tfile ] && return 3
return 0
}
# give multiop a chance to open
sleep 1
rm -f $DIR/$tfile
- replay_barrier mds1
+ replay_barrier mds
touch $DIR/g11 || return 1
kill -USR1 $pid
wait $pid || return 2
- fail mds1
+ fail mds
[ -e $DIR/$tfile ] && return 3
touch $DIR/h11 || return 4
return 0
test_16() {
- replay_barrier mds1
+ replay_barrier mds
mcreate $DIR/$tfile
munlink $DIR/$tfile
mcreate $DIR/$tfile-2
- fail mds1
+ fail mds
[ -e $DIR/$tfile ] && return 1
[ -e $DIR/$tfile-2 ] || return 2
munlink $DIR/$tfile-2 || return 3
run_test 16 "|X| open(O_CREAT), unlink, touch new, unlink new"
test_17() {
- replay_barrier mds1
+ replay_barrier mds
multiop $DIR/$tfile O_c &
pid=$!
# give multiop a chance to open
sleep 1
- fail mds1
+ fail mds
kill -USR1 $pid || return 1
wait $pid || return 2
$CHECKSTAT -t file $DIR/$tfile || return 3
run_test 17 "|X| open(O_CREAT), |replay| close"
test_18() {
- replay_barrier mds1
+ replay_barrier mds
multiop $DIR/$tfile O_tSc &
pid=$!
# give multiop a chance to open
kill -USR1 $pid
wait $pid || return 2
- fail mds1
+ fail mds
[ -e $DIR/$tfile ] && return 3
[ -e $DIR/$tfile-2 ] || return 4
# this touch frequently fails
# bug 1855 (a simpler form of test_11 above)
test_19() {
- replay_barrier mds1
+ replay_barrier mds
mcreate $DIR/$tfile
echo "old" > $DIR/$tfile
mv $DIR/$tfile $DIR/$tfile-2
grep old $DIR/$tfile-2
- fail mds1
+ fail mds
grep old $DIR/$tfile-2 || return 2
}
run_test 19 "|X| mcreate, open, write, rename "
test_20() {
- replay_barrier mds1
+ replay_barrier mds
multiop $DIR/$tfile O_tSc &
pid=$!
# give multiop a chance to open
sleep 1
rm -f $DIR/$tfile
- fail mds1
+ fail mds
kill -USR1 $pid
wait $pid || return 1
[ -e $DIR/$tfile ] && return 2
}
run_test 20 "|X| open(O_CREAT), unlink, replay, close (test mds_cleanup_orphans)"
+test_20b() { # bug 10480
+ BEFOREUSED=`df -P $DIR | tail -1 | awk '{ print $3 }'`
+
+ dd if=/dev/zero of=$DIR/$tfile bs=4k count=10000 &
+ pid=$!
+ while [ ! -e $DIR/$tfile ] ; do
+ usleep 60 # give dd a chance to start
+ done
+
+ lfs getstripe $DIR/$tfile || return 1
+ rm -f $DIR/$tfile || return 2 # make it an orphan
+ mds_evict_client
+ df -P $DIR || df -P $DIR || true # reconnect
+
+ fail mds # start orphan recovery
+ df -P $DIR || df -P $DIR || true # reconnect
+ sleep 2
+
+ AFTERUSED=`df -P $DIR | tail -1 | awk '{ print $3 }'`
+ log "before $BEFOREUSED, after $AFTERUSED"
+ [ $AFTERUSED -gt $((BEFOREUSED + 20)) ] && \
+ error "after $AFTERUSED > before $BEFOREUSED" && return 5
+ return 0
+}
+run_test 20b "write, unlink, eviction, replay, (test mds_cleanup_orphans)"
+
+test_20c() { # bug 10480
+ dd if=/dev/zero of=$DIR/$tfile bs=4k count=10000
+
+ exec 100< $DIR/$tfile
+
+ ls -la $DIR/$tfile
+
+ mds_evict_client
+
+ df -P $DIR || df -P $DIR || true # reconnect
+
+ exec 100<&-
+
+ test -s $DIR/$tfile || error "File was truncated"
+
+ return 0
+}
+run_test 20c "check that client eviction does not affect file content"
+
test_21() {
- replay_barrier mds1
+ replay_barrier mds
multiop $DIR/$tfile O_tSc &
pid=$!
# give multiop a chance to open
rm -f $DIR/$tfile
touch $DIR/g11 || return 1
- fail mds1
+ fail mds
kill -USR1 $pid
wait $pid || return 2
[ -e $DIR/$tfile ] && return 3
# give multiop a chance to open
sleep 1
- replay_barrier mds1
+ replay_barrier mds
rm -f $DIR/$tfile
- fail mds1
+ fail mds
kill -USR1 $pid
wait $pid || return 1
[ -e $DIR/$tfile ] && return 2
# give multiop a chance to open
sleep 1
- replay_barrier mds1
+ replay_barrier mds
rm -f $DIR/$tfile
touch $DIR/g11 || return 1
- fail mds1
+ fail mds
kill -USR1 $pid
wait $pid || return 2
[ -e $DIR/$tfile ] && return 3
# give multiop a chance to open
sleep 1
- replay_barrier mds1
- fail mds1
+ replay_barrier mds
+ fail mds
rm -f $DIR/$tfile
kill -USR1 $pid
wait $pid || return 1
sleep 1
rm -f $DIR/$tfile
- replay_barrier mds1
- fail mds1
+ replay_barrier mds
+ fail mds
kill -USR1 $pid
wait $pid || return 1
[ -e $DIR/$tfile ] && return 2
run_test 25 "open(O_CREAT), unlink, replay, close (test mds_cleanup_orphans)"
test_26() {
- replay_barrier mds1
+ replay_barrier mds
multiop $DIR/$tfile-1 O_tSc &
pid1=$!
multiop $DIR/$tfile-2 O_tSc &
kill -USR1 $pid2
wait $pid2 || return 1
- fail mds1
+ fail mds
kill -USR1 $pid1
wait $pid1 || return 2
[ -e $DIR/$tfile-1 ] && return 3
run_test 26 "|X| open(O_CREAT), unlink two, close one, replay, close one (test mds_cleanup_orphans)"
test_27() {
- replay_barrier mds1
+ replay_barrier mds
multiop $DIR/$tfile-1 O_tSc &
pid1=$!
multiop $DIR/$tfile-2 O_tSc &
rm -f $DIR/$tfile-1
rm -f $DIR/$tfile-2
- fail mds1
+ fail mds
kill -USR1 $pid1
wait $pid1 || return 1
kill -USR1 $pid2
pid2=$!
# give multiop a chance to open
sleep 1
- replay_barrier mds1
+ replay_barrier mds
rm -f $DIR/$tfile-1
rm -f $DIR/$tfile-2
kill -USR1 $pid2
wait $pid2 || return 1
- fail mds1
+ fail mds
kill -USR1 $pid1
wait $pid1 || return 2
[ -e $DIR/$tfile-1 ] && return 3
pid2=$!
# give multiop a chance to open
sleep 1
- replay_barrier mds1
+ replay_barrier mds
rm -f $DIR/$tfile-1
rm -f $DIR/$tfile-2
- fail mds1
+ fail mds
kill -USR1 $pid1
wait $pid1 || return 1
kill -USR1 $pid2
rm -f $DIR/$tfile-1
rm -f $DIR/$tfile-2
- replay_barrier mds1
- fail mds1
+ replay_barrier mds
+ fail mds
kill -USR1 $pid1
wait $pid1 || return 1
kill -USR1 $pid2
sleep 1
rm -f $DIR/$tfile-1
- replay_barrier mds1
+ replay_barrier mds
rm -f $DIR/$tfile-2
- fail mds1
+ fail mds
kill -USR1 $pid1
wait $pid1 || return 1
kill -USR1 $pid2
# Abort recovery before client complete
test_33() {
- replay_barrier mds1
+ replay_barrier mds
touch $DIR/$tfile
- fail_abort mds1
+ fail_abort mds
# this file should be gone, because the replay was aborted
- $CHECKSTAT -t file $DIR/$tfile && return 1
+ $CHECKSTAT -t file $DIR/$tfile && return 3
return 0
}
run_test 33 "abort recovery before client does replay"
sleep 1
rm -f $DIR/$tfile
- replay_barrier mds1
- fail_abort mds1
+ replay_barrier mds
+ fail_abort mds
kill -USR1 $pid
[ -e $DIR/$tfile ] && return 1
sync
sync
sleep 1
# give a chance to remove from MDS
- fail_abort mds1
+ fail_abort mds
$CHECKSTAT -t file $DIR/$tfile && return 1 || true
}
run_test 35 "test recovery from llog for unlink op"
# b=2432 resent cancel after replay uses wrong cookie,
# so don't resend cancels
test_36() {
- replay_barrier mds1
+ replay_barrier mds
touch $DIR/$tfile
checkstat $DIR/$tfile
- facet_failover mds1
- cancel_lru_locks MDC
+ facet_failover mds
+ cancel_lru_locks mdc
if dmesg | grep "unknown lock cookie"; then
echo "cancel after replay failed"
return 1
sleep 1
rmdir $DIR/$tfile
- replay_barrier mds1
+ replay_barrier mds
# clear the dmesg buffer so we only see errors from this recovery
dmesg -c >/dev/null
- fail_abort mds1
+ fail_abort mds
kill -USR1 $pid
dmesg | grep "mds_unlink_orphan.*error .* unlinking orphan" && return 1
sync
test_38() {
createmany -o $DIR/$tfile-%d 800
unlinkmany $DIR/$tfile-%d 0 400
- replay_barrier mds1
- fail mds1
+ replay_barrier mds
+ fail mds
unlinkmany $DIR/$tfile-%d 400 400
sleep 2
$CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
}
run_test 38 "test recovery from unlink llog (test llog_gen_rec) "
-test_39() {
+test_39() { # bug 4176
createmany -o $DIR/$tfile-%d 800
- replay_barrier mds1
+ replay_barrier mds
unlinkmany $DIR/$tfile-%d 0 400
- fail mds1
+ fail mds
unlinkmany $DIR/$tfile-%d 400 400
sleep 2
+ ls -1f $DIR/$tfile-*
$CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
}
run_test 39 "test recovery from unlink llog (test llog_gen_rec) "
count_ost_writes() {
- cat /proc/fs/lustre/osc/*/stats |
- awk -vwrites=0 '/ost_write/ { writes += $2 } END { print writes; }'
+ awk -vwrites=0 '/ost_write/ { writes += $2 } END { print writes; }' $LPROC/osc/*/stats
}
#b=2477,2532
writeme -s $MOUNT/${tfile}-2 &
WRITE_PID=$!
sleep 1
- facet_failover mds1
+ facet_failover mds
#define OBD_FAIL_MDS_CONNECT_NET 0x117
do_facet mds "sysctl -w lustre.fail_loc=0x80000117"
kill -USR1 $PID
# make sure the start of the file is ost1
lfs setstripe $f $((128 * 1024)) 0 0
do_facet client dd if=/dev/zero of=$f bs=4k count=1 || return 3
- cancel_lru_locks OSC
+ cancel_lru_locks osc
# fail ost2 and read from ost1
- local osc2_dev=`$LCTL device_list | \
- awk '(/ost2.*client_facet/){print $4}' `
- $LCTL --device %$osc2_dev deactivate
+ local osc2dev=`grep ${ost2_svc}-osc- $LPROC/devices | awk '{print $1}'`
+ [ "$osc2dev" ] || return 4
+ $LCTL --device $osc2dev deactivate || return 1
do_facet client dd if=$f of=/dev/null bs=4k count=1 || return 3
- $LCTL --device %$osc2_dev activate
+ $LCTL --device $osc2dev activate || return 2
return 0
}
run_test 41 "read from a valid osc while other oscs are invalid"
# test MDS recovery after ost failure
test_42() {
- blocks=`df $MOUNT | tail -n 1 | awk '{ print $1 }'`
+ blocks=`df -P $MOUNT | tail -n 1 | awk '{ print $2 }'`
createmany -o $DIR/$tfile-%d 800
- replay_barrier ost
+ replay_barrier ost1
unlinkmany $DIR/$tfile-%d 0 400
- facet_failover ost
+ DEBUG42="`sysctl -n lnet.debug`"
+ sysctl -w lnet.debug=-1
+ facet_failover ost1
- # osc is evicted, fs is smaller
- blocks_after=`df $MOUNT | tail -n 1 | awk '{ print $1 }'`
- [ $blocks_after -lt $blocks ] || return 1
+ # osc is evicted, fs is smaller (but only with failout OSTs (bug 7287)
+ #blocks_after=`df -P $MOUNT | tail -n 1 | awk '{ print $2 }'`
+ #[ $blocks_after -lt $blocks ] || return 1
echo wait for MDS to timeout and recover
sleep $((TIMEOUT * 2))
+ sysctl -w lnet.debug=$DEBUG42
unlinkmany $DIR/$tfile-%d 400 400
$CHECKSTAT -t file $DIR/$tfile-* && return 2 || true
}
run_test 42 "recovery after ost failure"
-# b=2530
# timeout in MDS/OST recovery RPC will LBUG MDS
-test_43() {
- replay_barrier mds1
+test_43() { # bug 2530
+ replay_barrier mds
# OBD_FAIL_OST_CREATE_NET 0x204
- do_facet ost "sysctl -w lustre.fail_loc=0x80000204"
- facet_failover mds1
- df $MOUNT || return 1
+ do_facet ost1 "sysctl -w lustre.fail_loc=0x80000204"
+ fail mds
sleep 10
- do_facet ost "sysctl -w lustre.fail_loc=0"
+ do_facet ost1 "sysctl -w lustre.fail_loc=0"
return 0
}
run_test 43 "mds osc import failure during recovery; don't LBUG"
test_44() {
- mdcdev=`awk '/mds_svc_MNT/ {print $1}' < /proc/fs/lustre/devices`
- do_facet mds "sysctl -w lustre.fail_loc=0x80000701"
- $LCTL --device $mdcdev recover
- df $MOUNT
+ mdcdev=`awk '/-mdc-/ {print $1}' $LPROC/devices`
+ [ "$mdcdev" ] || exit 2
+ for i in `seq 1 10`; do
+ #define OBD_FAIL_TGT_CONN_RACE 0x701
+ do_facet mds "sysctl -w lustre.fail_loc=0x80000701"
+ $LCTL --device $mdcdev recover
+ df $MOUNT
+ done
do_facet mds "sysctl -w lustre.fail_loc=0"
return 0
}
run_test 44 "race in target handle connect"
+test_44b() {
+ mdcdev=`awk '/-mdc-/ {print $1}' $LPROC/devices`
+ [ "$mdcdev" ] || exit 2
+ for i in `seq 1 10`; do
+ #define OBD_FAIL_TGT_DELAY_RECONNECT 0x704
+ do_facet mds "sysctl -w lustre.fail_loc=0x80000704"
+ $LCTL --device $mdcdev recover
+ df $MOUNT
+ done
+ do_facet mds "sysctl -w lustre.fail_loc=0"
+ return 0
+}
+run_test 44b "race in target handle connect"
+
# Handle failed close
test_45() {
- mdcdev=`awk '/mds_svc_MNT/ {print $1}' < /proc/fs/lustre/devices`
+ mdcdev=`awk '/-mdc-/ {print $1}' $LPROC/devices`
+ [ "$mdcdev" ] || exit 2
$LCTL --device $mdcdev recover
multiop $DIR/$tfile O_c &
# This will cause the CLOSE to fail before even
# allocating a reply buffer
- $LCTL --device $mdcdev deactivate
+ $LCTL --device $mdcdev deactivate || return 4
# try the close
kill -USR1 $pid
wait $pid || return 1
- $LCTL --device $mdcdev activate
+ $LCTL --device $mdcdev activate || return 5
+ sleep 1
$CHECKSTAT -t file $DIR/$tfile || return 2
return 0
test_46() {
dmesg -c >/dev/null
drop_reply "touch $DIR/$tfile"
- fail mds1
+ fail mds
# ironically, the previous test, 45, will cause a real forced close,
# so just look for one for this test
dmesg | grep -i "force closing client file handle for $tfile" && return 1
}
run_test 46 "Don't leak file handle after open resend (3325)"
-# b=2824
-test_47() {
-
+test_47() { # bug 2824
# create some files to make sure precreate has been done on all
# OSTs. (just in case this test is run independently)
createmany -o $DIR/$tfile 20 || return 1
# OBD_FAIL_OST_CREATE_NET 0x204
- fail ost
- do_facet ost "sysctl -w lustre.fail_loc=0x80000204"
+ fail ost1
+ do_facet ost1 "sysctl -w lustre.fail_loc=0x80000204"
df $MOUNT || return 2
# let the MDS discover the OST failure, attempt to recover, fail
createmany -o $DIR/$tfile 20 || return 3
unlinkmany $DIR/$tfile 20 || return 4
- do_facet ost "sysctl -w lustre.fail_loc=0"
+ do_facet ost1 "sysctl -w lustre.fail_loc=0"
return 0
}
run_test 47 "MDS->OSC failure during precreate cleanup (2824)"
-
test_48() {
- createmany -o $DIR/${tfile}- 100
- $CHECKSTAT $DIR/${tfile}-99 || return 1
- mds_evict_client
- df $MOUNT || df $MOUNT || return 2
- sleep 1
- $CHECKSTAT $DIR/${tfile}-99 || return 3
+ replay_barrier mds
+ createmany -o $DIR/$tfile 20 || return 1
+ # OBD_FAIL_OST_EROFS 0x216
+ fail mds
+ do_facet ost1 "sysctl -w lustre.fail_loc=0x80000216"
+ df $MOUNT || return 2
- dmesg -c >/dev/null
- replay_barrier mds1
- fail mds1
- unlinkmany $DIR/${tfile}- 100 || return 4
- if dmesg | grep "back in time"; then
- echo "server went back in time!"
- return 5
- fi
- return 0
-}
-run_test 48 "Don't lose transno when client is evicted (2525)"
+ createmany -o $DIR/$tfile 20 20 || return 2
+ unlinkmany $DIR/$tfile 40 || return 3
-# b=3550 - replay of unlink
-test_49() {
- replay_barrier mds1
- createmany -o $DIR/$tfile-%d 400 || return 1
- unlinkmany $DIR/$tfile-%d 0 400 || return 2
- fail mds1
- $CHECKSTAT -t file $DIR/$tfile-* && return 3 || true
+ do_facet ost1 "sysctl -w lustre.fail_loc=0"
+ return 0
}
-run_test 49 "re-write records to llog as written during fail"
+run_test 48 "MDS->OSC failure during precreate cleanup (2824)"
test_50() {
- local osc_dev=`$LCTL device_list | \
- awk '(/ost_svc_mds1_svc/){print $4}' `
- $LCTL --device %$osc_dev recover && $LCTL --device %$osc_dev recover
+ local oscdev=`grep ${ost1_svc}-osc- $LPROC/devices | awk '{print $1}'`
+ [ "$oscdev" ] || return 1
+ $LCTL --device $oscdev recover && $LCTL --device $oscdev recover
# give the mds_lov_sync threads a chance to run
sleep 5
}
# b3764 timed out lock replay
test_52() {
touch $DIR/$tfile
- cancel_lru_locks MDC
+ cancel_lru_locks mdc
- multiop $DIR/$tfile s
- replay_barrier mds1
- do_facet mds1 "sysctl -w lustre.fail_loc=0x8000030c"
- fail mds1
- do_facet mds1 "sysctl -w lustre.fail_loc=0x0"
+ multiop $DIR/$tfile s || return 1
+ replay_barrier mds
+#define OBD_FAIL_LDLM_REPLY 0x30c
+ do_facet mds "sysctl -w lustre.fail_loc=0x8000030c"
+ fail mds || return 2
+ do_facet mds "sysctl -w lustre.fail_loc=0x0"
$CHECKSTAT -t file $DIR/$tfile-* && return 3 || true
}
run_test 52 "time out lock replay (3764)"
-equals_msg test complete, cleaning up
-$CLEANUP
+#b_cray 53 "|X| open request and close reply while two MDC requests in flight"
+#b_cray 54 "|X| open request and close reply while two MDC requests in flight"
+
+#b3761 ASSERTION(hash != 0) failed
+test_55() {
+# OBD_FAIL_MDS_OPEN_CREATE | OBD_FAIL_ONCE
+ do_facet mds "sysctl -w lustre.fail_loc=0x8000012b"
+ touch $DIR/$tfile &
+ # give touch a chance to run
+ sleep 5
+ do_facet mds "sysctl -w lustre.fail_loc=0x0"
+ rm $DIR/$tfile
+ return 0
+}
+run_test 55 "let MDS_CHECK_RESENT return the original return code instead of 0"
+
+#b3440 ASSERTION(rec->ur_fid2->id) failed
+test_56() {
+ ln -s foo $DIR/$tfile
+ replay_barrier mds
+ #drop_reply "cat $DIR/$tfile"
+ fail mds
+ sleep 10
+}
+run_test 56 "don't replay a symlink open request (3440)"
+#recovery one mds-ost setattr from llog
+test_57() {
+#define OBD_FAIL_MDS_OST_SETATTR 0x12c
+ do_facet mds "sysctl -w lustre.fail_loc=0x8000012c"
+ touch $DIR/$tfile
+ replay_barrier mds
+ fail mds
+ sleep 1
+ $CHECKSTAT -t file $DIR/$tfile || return 1
+ do_facet mds "sysctl -w lustre.fail_loc=0x0"
+ rm $DIR/$tfile
+}
+run_test 57 "test recovery from llog for setattr op"
+
+#recovery many mds-ost setattr from llog
+test_58() {
+#define OBD_FAIL_MDS_OST_SETATTR 0x12c
+ do_facet mds "sysctl -w lustre.fail_loc=0x8000012c"
+ mkdir $DIR/$tdir
+ createmany -o $DIR/$tdir/$tfile-%d 2500
+ replay_barrier mds
+ fail mds
+ sleep 2
+ $CHECKSTAT -t file $DIR/$tdir/$tfile-* || return 1
+ do_facet mds "sysctl -w lustre.fail_loc=0x0"
+ unlinkmany $DIR/$tdir/$tfile-%d 2500
+ rmdir $DIR/$tdir
+}
+run_test 58 "test recovery from llog for setattr op (test llog_gen_rec)"
+
+equals_msg `basename $0`: test complete, cleaning up
+$CLEANUP