LUSTRE=${LUSTRE:-`dirname $0`/..}
-UPCALL=${UPCALL:-$PWD/recovery-small-upcall.sh}
+
. $LUSTRE/tests/test-framework.sh
init_test_env $@
-. ${CONFIG:=$LUSTRE/tests/cfg/lmv.sh}
+. ${CONFIG:=$LUSTRE/tests/cfg/local.sh}
build_test_filter
# setting SETUP=" " and CLEANUP=" "
SETUP=${SETUP:-"setup"}
CLEANUP=${CLEANUP:-"cleanup"}
+FORCE=${FORCE:-"--force"}
make_config() {
rm -f $XMLCONFIG
-
- if [ "$MDSCOUNT" -gt 1 ]; then
- add_lmv lmv1
- for num in `seq $MDSCOUNT`; do
- MDSDEV=$TMP/mds${num}-`hostname`
- add_mds mds$num --dev $MDSDEV --size $MDSSIZE --master --lmv lmv1
- done
- add_lov_to_lmv lov1 lmv1 --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 --lmv lmv1 --lov lov1 --path $MOUNT
- else
- add_mds mds1 --dev $MDSDEV --size $MDSSIZE
- add_lov lov1 mds1 --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 mds1_svc --lov lov1 --path $MOUNT
- fi
+ add_mds mds --dev $MDSDEV --size $MDSSIZE
+ 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
}
setup() {
cleanup() {
zconf_umount `hostname` $MOUNT
- if [ "$MDSCOUNT" -gt 1 ]; then
- for num in `seq $MDSCOUNT`; do
- stop mds$num ${FORCE} $MDSLCONFARGS
- done
- else
- stop mds ${FORCE} $MDSLCONFARGS
- fi
- stop ost2 ${FORCE} --dump cleanup.log
- stop ost ${FORCE} --dump cleanup.log
+ stop mds ${FORCE} $MDSLCONFARGS
+ stop ost2 ${FORCE}
+ stop ost ${FORCE} --dump $TMP/recovery-small-`hostname`.log
+}
+
+replay() {
+ do_mds "sync"
+ do_mds 'echo -e "device \$mds1\\nprobe\\nnotransno\\nreadonly" | lctl'
+ do_client "$1" &
+ shutdown_mds -f
+ start_mds
+ wait
+ do_client "df -h $MOUNT" # trigger failover, if we haven't already
}
if [ ! -z "$EVAL" ]; then
exit $?
fi
-if [ "$ONLY" == "setup" ]; then
- setup
- exit
-fi
-
if [ "$ONLY" == "cleanup" ]; then
sysctl -w portals.debug=0 || true
- cleanup
+ FORCE=--force cleanup
exit
fi
REFORMAT=--reformat $SETUP
unset REFORMAT
+[ "$ONLY" == "setup" ] && exit
+
test_1() {
drop_request "mcreate $MOUNT/1" || return 1
drop_reint_reply "mcreate $MOUNT/2" || return 2
run_test 10 "finish request on server after client eviction (bug 1521)"
#bug 2460
-# wake up a thead waiting for completion after eviction
+# wake up a thread waiting for completion after eviction
test_11(){
do_facet client multiop $MOUNT/$tfile Ow || return 1
do_facet client multiop $MOUNT/$tfile or || return 2
do_facet client munlink $MOUNT/$tfile || return 4
}
-run_test 11 "wake up a thead waiting for completion after eviction (b=2460)"
+run_test 11 "wake up a thread waiting for completion after eviction (b=2460)"
#b=2494
test_12(){
# Bug 113, check that readdir lost recv timeout works.
test_13() {
- mkdir /mnt/lustre/readdir
- touch /mnt/lustre/readdir/newentry
+ mkdir /mnt/lustre/readdir || return 1
+ touch /mnt/lustre/readdir/newentry || return
# OBD_FAIL_MDS_READPAGE_NET|OBD_FAIL_ONCE
do_facet mds "sysctl -w lustre.fail_loc=0x80000104"
- ls /mnt/lustre/readdir || return 1
+ ls /mnt/lustre/readdir || return 3
do_facet mds "sysctl -w lustre.fail_loc=0"
- rm -rf /mnt/lustre/readdir
+ rm -rf /mnt/lustre/readdir || return 4
}
run_test 13 "mdc_readpage restart test (bug 1138)"
}
run_test 15 "failed open (-ENOMEM)"
+READ_AHEAD=`cat /proc/fs/lustre/llite/*/max_read_ahead_mb | head -n 1`
+stop_read_ahead() {
+ for f in /proc/fs/lustre/llite/*/max_read_ahead_mb; do
+ echo 0 > $f
+ done
+}
+
+start_read_ahead() {
+ for f in /proc/fs/lustre/llite/*/max_read_ahead_mb; do
+ echo $READ_AHEAD > $f
+ done
+}
+
test_16() {
do_facet client cp /etc/termcap $MOUNT
sync
+ stop_read_ahead
#define OBD_FAIL_PTLRPC_BULK_PUT_NET 0x504 | OBD_FAIL_ONCE
sysctl -w lustre.fail_loc=0x80000504
# give recovery a chance to finish (shouldn't take long)
sleep $TIMEOUT
do_facet client "cmp /etc/termcap $MOUNT/termcap" || return 2
+ start_read_ahead
}
run_test 16 "timeout bulk put, evict client (2732)"
# OBD_FAIL_PTLRPC_BULK_GET_NET 0x0503 | OBD_FAIL_ONCE
# client will get evicted here
sysctl -w lustre.fail_loc=0x80000503
- do_facet client cp /etc/termcap $DIR/$tfile
- sysctl -w lustre.fail_loc=0
-
+ # need to write enough to ensure we send an RPC
+ do_facet client dd if=/dev/zero of=$DIR/$tfile bs=1024k count=2
sleep $TIMEOUT
+ sysctl -w lustre.fail_loc=0
+ do_facet client "df $DIR"
# expect cmp to fail
do_facet client "cmp /etc/termcap $DIR/$tfile" && return 1
do_facet client "rm $DIR/$tfile" || return 2
drop_ldlm_cancel "chmod 0777 $f" || echo evicted
do_facet client checkstat -v -p 0777 $f || echo evicted
+ # let the client reconnect
+ sleep 5
do_facet client "munlink $f"
}
run_test 19a "test expired_lock_main on mds (2867)"
mkdir -p $DIR/$tdir
multiop $DIR/$tdir/${tfile} O_wc &
MULTI_PID=$!
- usleep 500
+ sleep 1
cancel_lru_locks OSC
#define OBD_FAIL_LDLM_ENQUEUE_EXTENT_ERR 0x308
do_facet ost sysctl -w lustre.fail_loc=0x80000308
}
run_test 20b "ldlm_handle_enqueue error (should return error)"
-$CLEANUP
+#b_cray run_test 21a "drop close request while close and open are both in flight"
+#b_cray run_test 21b "drop open request while close and open are both in flight"
+#b_cray run_test 21c "drop both request while close and open are both in flight"
+#b_cray run_test 21d "drop close reply while close and open are both in flight"
+#b_cray run_test 21e "drop open reply while close and open are both in flight"
+#b_cray run_test 21f "drop both reply while close and open are both in flight"
+#b_cray run_test 21g "drop open reply and close request while close and open are both in flight"
+#b_cray run_test 21h "drop open request and close reply while close and open are both in flight"
+#b_cray run_test 22 "drop close request and do mknod"
+#b_cray run_test 23 "client hang when close a file after mds crash"
+
+test_24() { # bug 2248 - eviction fails writeback but app doesn't see it
+ mkdir -p $DIR/$tdir
+ cancel_lru_locks OSC
+ multiop $DIR/$tdir/$tfile Owy_wyc &
+ MULTI_PID=$!
+ usleep 500
+# OBD_FAIL_PTLRPC_BULK_PUT_NET|OBD_FAIL_ONCE
+ sysctl -w lustre.fail_loc=0x80000503
+ usleep 500
+ kill -USR1 $MULTI_PID
+ wait $MULTI_PID
+ rc=$?
+ sysctl -w lustre.fail_loc=0x0
+ client_reconnect
+ [ $rc -eq 0 ] && error "multiop didn't fail fsync: rc $rc" || true
+}
+run_test 24 "fsync error (should return error)"
+
+
+test_25a() {
+ mkdir -p $DIR/$tdir
+ # put a load of file creates/writes/deletes for 10 min.
+ do_facet client "writemany -q -a $DIR/$tdir/$tfile 600 5" &
+ CLIENT_PID=$!
+ echo writemany pid $CLIENT_PID
+ sleep 10
+ FAILURE_MODE="SOFT"
+ fail mds
+ # wait for client to reconnect to MDS
+ sleep 60
+ fail mds
+ sleep 60
+ fail mds
+ # client process should see no problems even though MDS went down
+ wait $CLIENT_PID
+ rc=$?
+ echo writemany returned $rc
+ return $rc
+}
+run_test 25a "failover MDS under load"
+
+test_25b() {
+ mkdir -p $DIR/$tdir
+ # put a load of file creates/writes/deletes
+ do_facet client "writemany -q -a $DIR/$tdir/$tfile 300 5" &
+ CLIENT_PID=$!
+ echo writemany pid $CLIENT_PID
+ sleep 1
+ FAILURE_MODE="SOFT"
+ facet_failover mds
+ # failover at various points during recovery
+ sleep 1
+ facet_failover mds
+ sleep 5
+ facet_failover mds
+ sleep 10
+ facet_failover mds
+ sleep 20
+ facet_failover mds
+ # client process should see no problems even though MDS went down
+ # and recovery was interrupted
+ wait $CLIENT_PID
+ rc=$?
+ echo writemany returned $rc
+ return $rc
+}
+run_test 25b "failover MDS during recovery"
+
+test_25c_guts() {
+ do_facet client "writemany -q $DIR/$tdir/$tfile 600 5" &
+ CLIENT_PID=$!
+ echo writemany pid $CLIENT_PID
+ sleep 10
+ FAILURE_MODE="SOFT"
+ fail ost
+ rc=0
+ wait $CLIENT_PID || rc=$?
+ # active client process should see an EIO for down OST
+ [ $rc -eq 5 ] && { echo "writemany correctly failed $rc" && return 0; }
+ # but timing or failover setup may allow success
+ [ $rc -eq 0 ] && { echo "writemany succeeded" && return 0; }
+ echo "writemany returned $rc"
+ return $rc
+}
+
+test_25c() {
+ mkdir -p $DIR/$tdir
+ test_25c_guts
+ rc=$?
+ [ $rc -ne 0 ] && { return $rc; }
+ # wait for client to reconnect to OST
+ sleep 30
+ test_25c_guts
+ rc=$?
+ [ $rc -ne 0 ] && { return $rc; }
+ sleep 30
+ test_25c_guts
+ rc=$?
+ client_reconnect
+ return $rc
+}
+run_test 25c "failover OST under load"
+
+FORCE=--force $CLEANUP
git://git.whamcloud.com / fs / lustre-release.git / blobdiff