. $LUSTRE/tests/test-framework.sh
init_test_env $@
. ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
+init_logging
CHECK_GRANT=${CHECK_GRANT:-"yes"}
GRANT_CHECK_LIST=${GRANT_CHECK_LIST:-""}
-remote_mds_nodsh && log "SKIP: remote MDS with nodsh" && exit 0
+require_dsh_mds || exit 0
# Skip these tests
-# bug number: 17466 15962
-ALWAYS_EXCEPT="61d 33b $REPLAY_SINGLE_EXCEPT"
+# bug number: 17466 18857
+ALWAYS_EXCEPT="61d 33a 33b $REPLAY_SINGLE_EXCEPT"
if [ "$FAILURE_MODE" = "HARD" ] && mixed_ost_devs; then
CONFIG_EXCEPTIONS="0b 42 47 61a 61c"
touch $DIR/$tfile || return 3
rm $DIR/$tfile || return 4
}
+start_full_debug_logging
run_test 0c "fld create"
+stop_full_debug_logging
test_1() {
replay_barrier $SINGLEMDS
wait $pid || return 1
$CHECKSTAT -s 1 -p 0 $DIR/$tfile || return 2
+ rm $DIR/$tfile || return 4
return 0
}
run_test 13 "open chmod 0 |x| write close"
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
+ client_up || client_up || true # reconnect
fail $SINGLEMDS # start orphan recovery
- df -P $DIR || df -P $DIR || true # reconnect
wait_recovery_complete $SINGLEMDS || error "MDS recovery not done"
-
- # FIXME just because recovery is done doesn't mean we've finished
- # orphan cleanup. Fake it with a sleep for now...
- sleep 10
+ wait_mds_ost_sync || return 3
AFTERUSED=`df -P $DIR | tail -1 | awk '{ print $3 }'`
log "before $BEFOREUSED, after $AFTERUSED"
[ $AFTERUSED -gt $((BEFOREUSED + 20)) ] && \
ls -la $DIR/$tfile
mds_evict_client
-
- df -P $DIR || df -P $DIR || true # reconnect
+ client_up || client_up || true # reconnect
kill -USR1 $pid
- test -s $DIR/$tfile || error "File was truncated"
-
wait $pid || return 1
+ [ -s $DIR/$tfile ] || error "File was truncated"
+
return 0
}
run_test 20c "check that client eviction does not affect file content"
multiop_bg_pause $DIR/$tfile O_c || return 3
pid2=$!
mds_evict_client
- df $MOUNT || sleep 1 && df $MOUNT || return 1
+ client_up || client_up || return 1
kill -USR1 $pid1
kill -USR1 $pid2
wait $pid1 || return 4
sync
return 0
}
+start_full_debug_logging
run_test 37 "abort recovery before client does replay (test mds_cleanup_orphans for directories)"
+stop_full_debug_logging
test_38() {
createmany -o $DIR/$tfile-%d 800
# assert on trying to unlock the unlocked page.
test_41() {
[ $OSTCOUNT -lt 2 ] && \
- skip "skipping test 41: we don't have a second OST to test with" && \
+ skip_env "skipping test 41: we don't have a second OST to test with" && \
return
local f=$MOUNT/$tfile
}
run_test 43 "mds osc import failure during recovery; don't LBUG"
-test_44a() { # was test_44
+test_44a() { # was test_44
local at_max_saved=0
mdcdev=`lctl get_param -n devices | awk '/MDT0000-mdc-/ {print $1}'`
- [ "$mdcdev" ] || exit 2
+ [ "$mdcdev" ] || return 2
+ [ $(echo $mdcdev | wc -w) -eq 1 ] || { echo $mdcdev=$mdcdev && return 3; }
# adaptive timeouts slow this way down
- if at_is_valid && at_is_enabled; then
+ if at_is_enabled; then
at_max_saved=$(at_max_get mds)
at_max_set 40 mds
fi
for i in `seq 1 10`; do
- echo "$i of 10 ($(date +%s))"
- do_facet $SINGLEMDS "lctl get_param -n mdt.*.mdt.timeouts | grep service"
- #define OBD_FAIL_TGT_CONN_RACE 0x701
- do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000701"
- $LCTL --device $mdcdev recover
- df $MOUNT
+ echo "$i of 10 ($(date +%s))"
+ do_facet $SINGLEMDS "lctl get_param -n mdt.*.mdt.timeouts | grep service"
+ #define OBD_FAIL_TGT_CONN_RACE 0x701
+ do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000701"
+ # lctl below may fail, it is valid case
+ $LCTL --device $mdcdev recover
+ df $MOUNT
done
do_facet $SINGLEMDS "lctl set_param fail_loc=0"
[ $at_max_saved -ne 0 ] && at_max_set $at_max_saved mds
run_test 44a "race in target handle connect"
test_44b() {
- mdcdev=`lctl get_param -n devices | awk '/MDT0000-mdc-/ {print $1}'`
- [ "$mdcdev" ] || exit 2
+ local mdcdev=`lctl get_param -n devices | awk '/MDT0000-mdc-/ {print $1}'`
+ [ "$mdcdev" ] || return 2
+ [ $(echo $mdcdev | wc -w) -eq 1 ] || { echo $mdcdev=$mdcdev && return 3; }
+
for i in `seq 1 10`; do
echo "$i of 10 ($(date +%s))"
- do_facet $SINGLEMDS "lctl get_param -n mdt.*.mdt.timeouts | grep service"
- #define OBD_FAIL_TGT_DELAY_RECONNECT 0x704
- do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000704"
- $LCTL --device $mdcdev recover
- df $MOUNT
+ do_facet $SINGLEMDS "lctl get_param -n mdt.*.mdt.timeouts | grep service"
+ #define OBD_FAIL_TGT_DELAY_RECONNECT 0x704
+ do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000704"
+ # lctl below may fail, it is valid case
+ $LCTL --device $mdcdev recover
+ df $MOUNT
done
do_facet $SINGLEMDS "lctl set_param fail_loc=0"
return 0
# Handle failed close
test_45() {
mdcdev=`lctl get_param -n devices | awk '/MDT0000-mdc-/ {print $1}'`
- [ "$mdcdev" ] || exit 2
- $LCTL --device $mdcdev recover
+ [ "$mdcdev" ] || return 2
+ [ $(echo $mdcdev | wc -w) -eq 1 ] || { echo $mdcdev=$mdcdev && return 3; }
+
+ $LCTL --device $mdcdev recover || return 6
multiop_bg_pause $DIR/$tfile O_c || return 1
pid=$!
# OBD_FAIL_OST_CREATE_NET 0x204
fail ost1
do_facet ost1 "lctl set_param fail_loc=0x80000204"
- df $MOUNT || return 2
+ client_up || return 2
# let the MDS discover the OST failure, attempt to recover, fail
# and recover again.
test_48() {
remote_ost_nodsh && skip "remote OST with nodsh" && return 0
- [ "$OSTCOUNT" -lt "2" ] && skip "$OSTCOUNT < 2 OSTs -- skipping" && return
+ [ "$OSTCOUNT" -lt "2" ] && skip_env "$OSTCOUNT < 2 OSTs -- skipping" && return
replay_barrier $SINGLEMDS
createmany -o $DIR/$tfile 20 || return 1
# OBD_FAIL_OST_EROFS 0x216
facet_failover $SINGLEMDS
do_facet ost1 "lctl set_param fail_loc=0x80000216"
- df $MOUNT || return 2
+ client_up || return 2
createmany -o $DIR/$tfile 20 20 || return 2
unlinkmany $DIR/$tfile 40 || return 3
mkdir -p $DIR/${tdir}-1
mkdir -p $DIR/${tdir}-2
- multiop $DIR/${tdir}-1/f O_c &
+ multiop_bg_pause $DIR/${tdir}-1/f O_c || return 6
close_pid=$!
#define OBD_FAIL_MDS_REINT_NET 0x107
kill -USR1 $close_pid
cancel_lru_locks mdc # force the close
+ #bz20647: make sure all pids are exists before failover
+ [ -d /proc/$close_pid ] || error "close_pid doesn't exist"
+ [ -d /proc/$open_pid ] || error "open_pid doesn't exists"
replay_barrier_nodf $SINGLEMDS
fail_nodf $SINGLEMDS
wait $open_pid || return 1
run_test 53c "|X| open request and close request while two MDC requests in flight"
test_53d() {
+ cancel_lru_locks mdc # cleanup locks from former test cases
rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
mkdir -p $DIR/${tdir}-1
kill -USR1 $close_pid
cancel_lru_locks mdc # force the close
+ #bz20647: make sure all pids are exists before failover
+ [ -d /proc/$close_pid ] || error "close_pid doesn't exist"
+ [ -d /proc/$open_pid ] || error "open_pid doesn't exists"
replay_barrier_nodf $SINGLEMDS
fail_nodf $SINGLEMDS
wait $open_pid || return 1
do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
kill -USR1 $close_pid
cancel_lru_locks mdc # force the close
-
do_facet $SINGLEMDS "lctl set_param fail_loc=0"
+
+ #bz20647: make sure all pids are exists before failover
+ [ -d /proc/$close_pid ] || error "close_pid doesn't exist"
+ [ -d /proc/$open_pid ] || error "open_pid doesn't exists"
replay_barrier_nodf $SINGLEMDS
fail_nodf $SINGLEMDS
wait $open_pid || return 1
cancel_lru_locks mdc # force the close
sleep 1
+ #bz20647: make sure all pids are exists before failover
+ [ -d /proc/$close_pid ] || error "close_pid doesn't exist"
+ [ -d /proc/$open_pid ] || error "open_pid doesn't exists"
replay_barrier_nodf $SINGLEMDS
fail_nodf $SINGLEMDS
wait $open_pid || return 1
createmany -o $DIR/$tdir/$tfile- 25
#define OBD_FAIL_TGT_REPLAY_DROP 0x707
do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000707"
- facet_failover $SINGLEMDS
- df $MOUNT || return 1
+ fail $SINGLEMDS
do_facet $SINGLEMDS "lctl set_param fail_loc=0"
unlinkmany $DIR/$tdir/$tfile- 25 || return 2
return 0
echo "Cleaning up AT ..."
if [ -n "$ATOLDBASE" ]; then
- local at_history=$(do_facet mds "find /sys/ -name at_history")
- do_facet mds "echo $ATOLDBASE >> $at_history" || true
- do_facet ost1 "echo $ATOLDBASE >> $at_history" || true
+ local at_history=$($LCTL get_param -n at_history)
+ do_facet mds "lctl set_param at_history=$at_history" || true
+ do_facet ost1 "lctl set_param at_history=$at_history" || true
fi
if [ $AT_MAX_SET -ne 0 ]; then
at_start()
{
local at_max_new=600
- if ! at_is_valid; then
- skip "AT env is invalid"
- return 1
- fi
# Save at_max original values
local facet
done
if [ -z "$ATOLDBASE" ]; then
- local at_history=$(do_facet mds "find /sys/ -name at_history")
- [ -z "$at_history" ] && skip "missing /sys/.../at_history " && return 1
- ATOLDBASE=$(do_facet mds "cat $at_history")
+ ATOLDBASE=$(do_facet mds "lctl get_param -n at_history")
# speed up the timebase so we can check decreasing AT
- do_facet mds "echo 8 >> $at_history"
- do_facet ost1 "echo 8 >> $at_history"
+ do_facet mds "lctl set_param at_history=8" || true
+ do_facet ost1 "lctl set_param at_history=8" || true
# sleep for a while to cool down, should be > 8s and also allow
# at least one ping to be sent. simply use TIMEOUT to be safe.
do_facet ost1 "sysctl -w lustre.fail_loc=0"
CONN2=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
ATTEMPTS=$(($CONN2 - $CONN1))
- echo "$ATTEMPTS osc reconnect attemps on gradual slow"
+ echo "$ATTEMPTS osc reconnect attempts on gradual slow"
[ $ATTEMPTS -gt 0 ] && error_ignore 13721 "AT should have prevented reconnect"
return 0
}
at_start || return 0
CONN1=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
+
+ # exhaust precreations on ost1
+ local OST=$(lfs osts | grep 0": " | awk '{print $2}' | sed -e 's/_UUID$//')
+ local mdtosc=$(get_mdtosc_proc_path $OST)
+ local last_id=$(do_facet mds lctl get_param -n osc.$mdtosc.prealloc_last_id)
+ local next_id=$(do_facet mds lctl get_param -n osc.$mdtosc.prealloc_next_id)
+
+ mkdir -p $DIR/$tdir/${OST}
+ lfs setstripe $DIR/$tdir/${OST} -o 0 -c 1 || error "setstripe"
+ echo "Creating to objid $last_id on ost $OST..."
#define OBD_FAIL_OST_PAUSE_CREATE 0x223
do_facet ost1 "sysctl -w lustre.fail_val=20000"
do_facet ost1 "sysctl -w lustre.fail_loc=0x80000223"
- cp /etc/profile $DIR/$tfile || error "cp failed"
+ createmany -o $DIR/$tdir/${OST}/f $next_id $((last_id - next_id + 2))
+
client_reconnect
do_facet ost1 "lctl get_param -n ost.OSS.ost_create.timeouts"
log "phase 2"
CONN2=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
ATTEMPTS=$(($CONN2 - $CONN1))
- echo "$ATTEMPTS osc reconnect attemps on instant slow"
+ echo "$ATTEMPTS osc reconnect attempts on instant slow"
# do it again; should not timeout
do_facet ost1 "sysctl -w lustre.fail_loc=0x80000223"
cp /etc/profile $DIR/$tfile || error "cp failed"
do_facet ost1 "lctl get_param -n ost.OSS.ost_create.timeouts"
CONN3=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
ATTEMPTS=$(($CONN3 - $CONN2))
- echo "$ATTEMPTS osc reconnect attemps on 2nd slow"
+ echo "$ATTEMPTS osc reconnect attempts on 2nd slow"
[ $ATTEMPTS -gt 0 ] && error "AT should have prevented reconnect"
return 0
}
zconf_mount_clients $clients $DIR
- local duration=120
+ local duration=300
[ "$SLOW" = "no" ] && duration=60
local cmd="rundbench 1 -t $duration"
local PID=""
- do_nodes $clients "set -x; MISSING_DBENCH_OK=$MISSING_DBENCH_OK \
+ do_nodesv $clients "set -x; MISSING_DBENCH_OK=$MISSING_DBENCH_OK \
PATH=:$PATH:$LUSTRE/utils:$LUSTRE/tests/:$DBENCH_LIB \
DBENCH_LIB=$DBENCH_LIB TESTSUITE=$TESTSUITE TESTNAME=$TESTNAME \
LCTL=$LCTL $cmd" &
PID=$!
log "Started rundbench load PID=$PID ..."
-
- sleep $((duration / 4))
- replay_barrier $SINGLEMDS
- sleep 3 # give clients a time to do operations
-
- log "$TESTNAME fail mds 1"
- fail $SINGLEMDS
-
+ ELAPSED=0
+ NUM_FAILOVERS=0
+ START_TS=$(date +%s)
+ CURRENT_TS=$START_TS
+ while [ $ELAPSED -lt $duration ]; do
+ sleep 1
+ replay_barrier $SINGLEMDS
+ sleep 1 # give clients a time to do operations
+ # Increment the number of failovers
+ NUM_FAILOVERS=$((NUM_FAILOVERS+1))
+ log "$TESTNAME fail mds1 $NUM_FAILOVERS times"
+ fail $SINGLEMDS
+ CURRENT_TS=$(date +%s)
+ ELAPSED=$((CURRENT_TS - START_TS))
+ done
wait $PID || error "rundbench load on $CLIENTS failed!"
}
run_test 70b "mds recovery; $CLIENTCOUNT clients"
# bug 18554
test_74() {
+ local clients=${CLIENTS:-$HOSTNAME}
+
stop ost1
- zconf_umount $(hostname) $MOUNT
- fail $SINGLEMDS
- zconf_mount $(hostname) $MOUNT
+ zconf_umount_clients $clients $MOUNT
+ facet_failover $SINGLEMDS
+ zconf_mount_clients $clients $MOUNT
mount_facet ost1
touch $DIR/$tfile || return 1
rm $DIR/$tfile || return 2
- df $MOUNT || error "df failed: $?"
+ clients_up || error "client evicted: $?"
return 0
}
-run_test 74 "Ensure applications don't fail waiting for OST reocvery"
+run_test 74 "Ensure applications don't fail waiting for OST recovery"
test_80a() {
[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
}
run_test 82b "CMD: mkdir cross-node dir (fail mds with name)"
+test_83a() {
+ mkdir -p $DIR/$tdir
+ createmany -o $DIR/$tdir/$tfile- 10 || return 1
+#define OBD_FAIL_MDS_FAIL_LOV_LOG_ADD 0x140
+ do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000140"
+ unlinkmany $DIR/$tdir/$tfile- 10 || return 2
+}
+run_test 83a "fail log_add during unlink recovery"
+
+test_83b() {
+ mkdir -p $DIR/$tdir
+ createmany -o $DIR/$tdir/$tfile- 10 || return 1
+ replay_barrier $SINGLEMDS
+ unlinkmany $DIR/$tdir/$tfile- 10 || return 2
+#define OBD_FAIL_MDS_FAIL_LOV_LOG_ADD 0x140
+ do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000140"
+ fail $SINGLEMDS
+}
+run_test 83b "fail log_add during unlink recovery"
+
+test_84a() {
+#define OBD_FAIL_MDS_OPEN_WAIT_CREATE 0x144
+ do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000144"
+ createmany -o $DIR/$tfile- 1 &
+ PID=$!
+ mds_evict_client
+ wait $PID
+ client_up || client_up || true # reconnect
+}
+run_test 84a "stale open during export disconnect"
+
+test_85() { # bug 22190
+ local fail=0
+ do_facet ost1 "lctl set_param -n obdfilter.${ost1_svc}.sync_journal 1"
+
+ replay_barrier ost1
+ lfs setstripe -i 0 -c 1 $DIR/$tfile
+ dd oflag=dsync if=/dev/urandom of=$DIR/$tfile bs=4k count=100 || fail=1
+ fail_abort ost1
+ echo "FAIL $fail"
+ [ $fail -ne 0 ] || error "Write was successful"
+}
+run_test 85 "ensure there is no reply on bulk write if obd is in rdonly mode"
+
equals_msg `basename $0`: test complete, cleaning up
check_and_cleanup_lustre
[ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG && grep -q FAIL $TESTSUITELOG && exit 1 || true
git://git.whamcloud.com / fs / lustre-release.git / blobdiff