remote_mds_nodsh && log "SKIP: remote MDS with nodsh" && exit 0
# Skip these tests
-# bug number: 17466
-ALWAYS_EXCEPT="61d $REPLAY_SINGLE_EXCEPT"
+# bug number: 17466 15962
+ALWAYS_EXCEPT="61d 33b $REPLAY_SINGLE_EXCEPT"
if [ "$FAILURE_MODE" = "HARD" ] && mixed_ost_devs; then
CONFIG_EXCEPTIONS="0b 42 47 61a 61c"
run_test 2b "touch"
test_3a() {
+ local file=$DIR/$tfile
replay_barrier $SINGLEMDS
- mcreate $DIR/$tfile
- o_directory $DIR/$tfile
+ mcreate $file
+ openfile -f O_DIRECTORY $file
fail $SINGLEMDS
- $CHECKSTAT -t file $DIR/$tfile || return 2
- rm $DIR/$tfile
+ $CHECKSTAT -t file $file || return 2
+ rm $file
}
run_test 3a "replay failed open(O_DIRECTORY)"
done
fail $SINGLEMDS
for i in `seq 220`; do
- grep -q "tag-$i" $DIR/$tfile-$i || error "f1c-$i"
+ grep -q "tag-$i" $DIR/$tfile-$i || error "$tfile-$i"
done
rm -rf $DIR/$tfile-*
sleep 3
run_test 20a "|X| open(O_CREAT), unlink, replay, close (test mds_cleanup_orphans)"
test_20b() { # bug 10480
+ # XXX increase the debug level temporary
+ do_nodes $(comma_list $(nodes_list)) "$LCTL set_param debug=0x33f0406; $LCTL set_param debug_mb=150"
BEFOREUSED=`df -P $DIR | tail -1 | awk '{ print $3 }'`
dd if=/dev/zero of=$DIR/$tfile bs=4k count=10000 &
fail $SINGLEMDS # start orphan recovery
df -P $DIR || df -P $DIR || true # reconnect
- wait_mds_recovery_done || error "MDS recovery not done"
+ 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...
log "before $BEFOREUSED, after $AFTERUSED"
[ $AFTERUSED -gt $((BEFOREUSED + 20)) ] && \
error "after $AFTERUSED > before $BEFOREUSED"
+ # XXX decrease it back
+ do_nodes $(comma_list $(nodes_list)) "$LCTL set_param debug=$PTLDEBUG; $LCTL set_param debug_mb=$DEBUG_SIZE"
return 0
}
run_test 20b "write, unlink, eviction, replay, (test mds_cleanup_orphans)"
}
run_test 33a "abort recovery before client does replay"
-# Stale FID sequence
+# Stale FID sequence bug 15962
test_33b() { # was test_33a
replay_barrier $SINGLEMDS
createmany -o $DIR/$tfile-%d 10
run_test 53a "|X| close request while two MDC requests in flight"
test_53b() {
+ rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
+
mkdir -p $DIR/${tdir}-1
mkdir -p $DIR/${tdir}-2
multiop $DIR/${tdir}-1/f O_c &
run_test 53b "|X| open request while two MDC requests in flight"
test_53c() {
+ rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
+
mkdir -p $DIR/${tdir}-1
mkdir -p $DIR/${tdir}-2
multiop $DIR/${tdir}-1/f O_c &
run_test 53c "|X| open request and close request while two MDC requests in flight"
test_53d() {
+ rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
+
mkdir -p $DIR/${tdir}-1
mkdir -p $DIR/${tdir}-2
multiop $DIR/${tdir}-1/f O_c &
# give multiop a chance to open
sleep 1
- #define OBD_FAIL_MDS_CLOSE_NET_REP 0x13f
- do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000013f"
+ #define OBD_FAIL_MDS_CLOSE_NET_REP 0x13b
+ do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000013b"
kill -USR1 $close_pid
cancel_lru_locks mdc # force the close
do_facet $SINGLEMDS "lctl set_param fail_loc=0"
run_test 53d "|X| close reply while two MDC requests in flight"
test_53e() {
+ rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
+
mkdir -p $DIR/${tdir}-1
mkdir -p $DIR/${tdir}-2
multiop $DIR/${tdir}-1/f O_c &
run_test 53e "|X| open reply while two MDC requests in flight"
test_53f() {
+ rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
+
mkdir -p $DIR/${tdir}-1
mkdir -p $DIR/${tdir}-2
multiop $DIR/${tdir}-1/f O_c &
open_pid=$!
sleep 1
- #define OBD_FAIL_MDS_CLOSE_NET_REP 0x13f
- do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000013f"
+ #define OBD_FAIL_MDS_CLOSE_NET_REP 0x13b
+ do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000013b"
kill -USR1 $close_pid
cancel_lru_locks mdc # force the close
run_test 53f "|X| open reply and close reply while two MDC requests in flight"
test_53g() {
+ rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
+
mkdir -p $DIR/${tdir}-1
mkdir -p $DIR/${tdir}-2
multiop $DIR/${tdir}-1/f O_c &
run_test 53g "|X| drop open reply and close request while close and open are both in flight"
test_53h() {
+ rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
+
mkdir -p $DIR/${tdir}-1
mkdir -p $DIR/${tdir}-2
multiop $DIR/${tdir}-1/f O_c &
open_pid=$!
sleep 1
- #define OBD_FAIL_MDS_CLOSE_NET_REP 0x13f
- do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000013f"
+ #define OBD_FAIL_MDS_CLOSE_NET_REP 0x13b
+ do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000013b"
kill -USR1 $close_pid
cancel_lru_locks mdc # force the close
sleep 1
}
run_test 59 "test log_commit_thread vs filter_destroy race"
-# bug 17323
-test_59b() {
- do_facet $SINGLEMDS "lctl set_param debug=+rpctrace"
- mkdir -p $DIR/$tdir
- createmany -o $DIR/$tdir/$tfile-%d 2000
- sync
-#define OBD_FAIL_OBD_LOG_CANCEL_REP 0x606
- do_facet $SINGLEMDS "lctl set_param fail_loc=0x606"
- unlinkmany $DIR/$tdir/$tfile-%d 2000
-
- # make sure that all llcds left ost and nothing left cached
- sync
- sleep 10
- do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
-
- # sleep 2 obd_timeouts from ost to make sure that we get resents.
- local timeout=$(do_facet ost1 lctl get_param -n timeout)
- timeout=$((timeout * 2))
- log "Sleep $timeout"
- sleep $timeout
- do_facet $SINGLEMDS $LCTL dk | grep -q "RESENT cancel req"
- local res=$?
- rmdir $DIR/$tdir
- return $res
-}
-run_test 59b "resent handle in llog_origin_handle_cancel"
-
# race between add unlink llog vs cat log init in post_recovery (only for b1_6)
# bug 12086: should no oops and No ctxt error for this test
test_60() {
test_61a() { # was test_61
remote_ost_nodsh && skip "remote OST with nodsh" && return 0
- mkdir $DIR/$tdir
+ mkdir -p $DIR/$tdir
createmany -o $DIR/$tdir/$tfile-%d 800
- replay_barrier ost1
-# OBD_FAIL_OST_LLOG_RECOVERY_TIMEOUT 0x221
- unlinkmany $DIR/$tdir/$tfile-%d 800
+ replay_barrier ost1
+# OBD_FAIL_OST_LLOG_RECOVERY_TIMEOUT 0x221
+ unlinkmany $DIR/$tdir/$tfile-%d 800
set_nodes_failloc "$(osts_nodes)" 0x80000221
facet_failover ost1
- sleep 10
+ sleep 10
fail ost1
sleep 30
set_nodes_failloc "$(osts_nodes)" 0x0
-
+
$CHECKSTAT -t file $DIR/$tdir/$tfile-* && return 1
rmdir $DIR/$tdir
}
#test race mds llog sync vs llog cleanup
test_61b() {
-# OBD_FAIL_MDS_LLOG_SYNC_TIMEOUT 0x140
- do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000140"
- facet_failover $SINGLEMDS
+# OBD_FAIL_MDS_LLOG_SYNC_TIMEOUT 0x13a
+ do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000013a"
+ facet_failover $SINGLEMDS
sleep 10
fail $SINGLEMDS
do_facet client dd if=/dev/zero of=$DIR/$tfile bs=4k count=1 || return 1
test_61c() {
remote_ost_nodsh && skip "remote OST with nodsh" && return 0
-# OBD_FAIL_OST_CANCEL_COOKIE_TIMEOUT 0x222
- touch $DIR/$tfile
+# OBD_FAIL_OST_CANCEL_COOKIE_TIMEOUT 0x222
+ touch $DIR/$tfile
set_nodes_failloc "$(osts_nodes)" 0x80000222
- rm $DIR/$tfile
+ rm $DIR/$tfile
sleep 10
fail ost1
set_nodes_failloc "$(osts_nodes)" 0x0
}
run_test 61c "test race mds llog sync vs llog cleanup"
+test_61d() { # bug 16002 # bug 17466
+ shutdown_facet $SINGLEMDS
+#define OBD_FAIL_OBD_LLOG_SETUP 0x605
+ do_facet $SINGLEMDS "lctl set_param fail_loc=0x605"
+ start $SINGLEMDS `mdsdevname 1` $MDS_MOUNT_OPTS && error "mds start should have failed"
+ do_facet $SINGLEMDS "lctl set_param fail_loc=0"
+ start $SINGLEMDS `mdsdevname 1` $MDS_MOUNT_OPTS || error "cannot restart mds"
+}
+run_test 61d "error in llog_setup should cleanup the llog context correctly"
+
test_62() { # Bug 15756 - don't mis-drop resent replay
mkdir -p $DIR/$tdir
replay_barrier $SINGLEMDS
createmany -o $DIR/$tdir/$tfile- 25
-#define OBD_FAIL_TGT_REPLAY_DROP 0x706
+#define OBD_FAIL_TGT_REPLAY_DROP 0x707
do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000707"
facet_failover $SINGLEMDS
df $MOUNT || return 1
#Adaptive Timeouts (bug 3055)
AT_MAX_SET=0
-# Suppose that all osts have the same at_max
-for facet in mds client ost; do
- eval AT_MAX_SAVE_${facet}=$(at_max_get $facet)
-done
+
+at_cleanup () {
+ local var
+ local facet
+ local at_new
+
+ echo "Cleaning up AT ..."
+ if [ -n "$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
+ fi
+
+ if [ $AT_MAX_SET -ne 0 ]; then
+ for facet in mds client ost; do
+ var=AT_MAX_SAVE_${facet}
+ echo restore AT on $facet to saved value ${!var}
+ at_max_set ${!var} $facet
+ at_new=$(at_max_get $facet)
+ echo Restored AT value on $facet $at_new
+ [ $at_new -eq ${!var} ] || \
+ error "$facet : AT value was not restored SAVED ${!var} NEW $at_new"
+ done
+ fi
+}
at_start()
{
return 1
fi
+ # Save at_max original values
+ local facet
+ if [ $AT_MAX_SET -eq 0 ]; then
+ # Suppose that all osts have the same at_max
+ for facet in mds client ost; do
+ eval AT_MAX_SAVE_${facet}=$(at_max_get $facet)
+ done
+ fi
local at_max
-
for facet in mds client ost; do
at_max=$(at_max_get $facet)
if [ $at_max -ne $at_max_new ]; then
# because previous tests may have caused this value to increase.
REQ_DELAY=`lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts |
awk '/portal 12/ {print $5}'`
- REQ_DELAY=$((${REQ_DELAY} + 5))
+ REQ_DELAY=$((${REQ_DELAY} + ${REQ_DELAY} / 4 + 5))
do_facet mds lctl set_param fail_val=$((${REQ_DELAY} * 1000))
#define OBD_FAIL_PTLRPC_PAUSE_REQ 0x50a
# because previous tests may have caused this value to increase.
REQ_DELAY=`lctl get_param -n osc.${FSNAME}-OST0000-osc-*.timeouts |
awk '/portal 6/ {print $5}'`
- REQ_DELAY=$((${REQ_DELAY} + 5))
+ REQ_DELAY=$((${REQ_DELAY} + ${REQ_DELAY} / 4 + 5))
do_facet ost1 lctl set_param fail_val=${REQ_DELAY}
#define OBD_FAIL_OST_BRW_PAUSE_PACK 0x224
at_start || return 0
local ldlm_enqueue_min=$(find /sys -name ldlm_enqueue_min)
[ -z "$ldlm_enqueue_min" ] && skip "missing /sys/.../ldlm_enqueue_min" && return 0
+ local ldlm_enqueue_min_r=$(do_facet ost1 "find /sys -name ldlm_enqueue_min")
+ [ -z "$ldlm_enqueue_min_r" ] && skip "missing /sys/.../ldlm_enqueue_min in the ost1" && return 0
local ENQ_MIN=$(cat $ldlm_enqueue_min)
+ local ENQ_MIN_R=$(do_facet ost1 "cat $ldlm_enqueue_min_r")
echo $TIMEOUT >> $ldlm_enqueue_min
- rm -f $DIR/${tfile}_[1-2]
- lfs setstripe $DIR/$tfile --index=0 --count=1
+ do_facet ost1 "echo $TIMEOUT >> $ldlm_enqueue_min_r"
+
+ rm -rf $DIR/$tdir
+ mkdir -p $DIR/$tdir
+ lfs setstripe $DIR/$tdir --index=0 --count=1
#define OBD_FAIL_LDLM_PAUSE_CANCEL 0x312
sysctl -w lustre.fail_val=$(($TIMEOUT - 1))
sysctl -w lustre.fail_loc=0x80000312
- cp /etc/profile $DIR/${tfile}_1 || error "1st cp failed $?"
- sysctl -w lustre.fail_val=$((TIMEOUT * 3 / 2))
+ cp /etc/profile $DIR/$tdir/${tfile}_1 || error "1st cp failed $?"
+ sysctl -w lustre.fail_val=$((TIMEOUT * 5 / 4))
sysctl -w lustre.fail_loc=0x80000312
- cp /etc/profile $DIR/${tfile}_2 || error "2nd cp failed $?"
+ cp /etc/profile $DIR/$tdir/${tfile}_2 || error "2nd cp failed $?"
sysctl -w lustre.fail_loc=0
+
echo $ENQ_MIN >> $ldlm_enqueue_min
+ do_facet ost1 "echo $ENQ_MIN_R >> $ldlm_enqueue_min_r"
+ rm -rf $DIR/$tdir
return 0
}
run_test 68 "AT: verify slowing locks"
-if [ -n "$ATOLDBASE" ]; then
- at_history=$(do_facet mds "find /sys/ -name at_history")
- do_facet mds "echo $ATOLDBASE >> $at_history" || true
- do_facet ost1 "echo $ATOLDBASE >> $at_history" || true
-fi
-
-if [ $AT_MAX_SET -ne 0 ]; then
- for facet in mds client ost; do
- var=AT_MAX_SAVE_${facet}
- echo restore AT on $facet to saved value ${!var}
- at_max_set ${!var} $facet
- AT_NEW=$(at_max_get $facet)
- echo Restored AT value on $facet $AT_NEW
- [ $AT_NEW -ne ${!var} ] && \
- error "$facet : AT value was not restored SAVED ${!var} NEW $AT_NEW"
- done
-fi
-
+at_cleanup
# end of AT tests includes above lines
error "dd failed on $CLIENT"
done
- local prev_client=$(echo $clients | sed 's/^.* \(.\+\)$/\1/')
+ local prev_client=$(echo $clients | sed 's/^.* \(.\+\)$/\1/')
for C in ${CLIENTS//,/ }; do
do_node $prev_client dd if=$DIR/${tfile}_${C} of=/dev/null 2>/dev/null || \
error "dd if=$DIR/${tfile}_${C} failed on $prev_client"
run_test 70a "check multi client t-f"
test_70b () {
- [ -z "$CLIENTS" ] && \
- { skip "Need two or more clients." && return; }
- [ $CLIENTCOUNT -lt 2 ] && \
- { skip "Need two or more clients, have $CLIENTCOUNT" && return; }
+ local clients=${CLIENTS:-$HOSTNAME}
- zconf_mount_clients $CLIENTS $DIR
+ zconf_mount_clients $clients $DIR
- local duration="-t 60"
- local cmd="rundbench 1 $duration "
+ local duration=300
+ [ "$SLOW" = "no" ] && duration=60
+ local cmd="rundbench 1 -t $duration"
local PID=""
- for CLIENT in ${CLIENTS//,/ }; do
- $PDSH $CLIENT "set -x; PATH=:$PATH:$LUSTRE/utils:$LUSTRE/tests/:${DBENCH_LIB} DBENCH_LIB=${DBENCH_LIB} $cmd" &
- PID=$!
- echo $PID >pid.$CLIENT
- echo "Started load PID=`cat pid.$CLIENT`"
+ do_nodes $clients "set -x; MISSING_DBENCH_OK=$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=$PID ..."
+ ELAPSED=0
+ NUM_FAILOVERS=0
+ START_TS=$(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"
+ facet_failover $SINGLEMDS
+ CURRENT_TS=$(date +%s)
+ ELAPSED=$((CURRENT_TS - START_TS))
done
+ wait $PID || error "rundbench load on $CLIENTS failed!"
+}
+run_test 70b "mds recovery; $CLIENTCOUNT clients"
+# end multi-client tests
- replay_barrier $SINGLEMDS
- sleep 3 # give clients a time to do operations
+test_73a() {
+ multiop_bg_pause $DIR/$tfile O_tSc || return 3
+ pid=$!
+ rm -f $DIR/$tfile
- log "$TESTNAME fail mds 1"
- fail $SINGLEMDS
+ replay_barrier $SINGLEMDS
+#define OBD_FAIL_LDLM_ENQUEUE 0x302
+ do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000302"
+ fail $SINGLEMDS
+ kill -USR1 $pid
+ wait $pid || return 1
+ [ -e $DIR/$tfile ] && return 2
+ return 0
+}
+run_test 73a "open(O_CREAT), unlink, replay, reconnect before open replay , close"
-# wait for client to reconnect to MDS
- sleep $TIMEOUT
+test_73b() {
+ multiop_bg_pause $DIR/$tfile O_tSc || return 3
+ pid=$!
+ rm -f $DIR/$tfile
- for CLIENT in ${CLIENTS//,/ }; do
- PID=`cat pid.$CLIENT`
- wait $PID
- rc=$?
- echo "load on ${CLIENT} returned $rc"
- done
+ replay_barrier $SINGLEMDS
+#define OBD_FAIL_LDLM_REPLY 0x30c
+ do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000030c"
+ fail $SINGLEMDS
+ kill -USR1 $pid
+ wait $pid || return 1
+ [ -e $DIR/$tfile ] && return 2
+ return 0
+}
+run_test 73b "open(O_CREAT), unlink, replay, reconnect at open_replay reply, close"
+test_73c() {
+ multiop_bg_pause $DIR/$tfile O_tSc || return 3
+ pid=$!
+ rm -f $DIR/$tfile
+
+ replay_barrier $SINGLEMDS
+#define OBD_FAIL_TGT_LAST_REPLAY 0x710
+ do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000710"
+ fail $SINGLEMDS
+ kill -USR1 $pid
+ wait $pid || return 1
+ [ -e $DIR/$tfile ] && return 2
+ return 0
}
-run_test 70b "mds recovery; $CLIENTCOUNT clients"
-# end multi-client tests
+run_test 73c "open(O_CREAT), unlink, replay, reconnect at last_replay, close"
+
+# bug 18554
+test_74() {
+ stop ost1
+ zconf_umount $(hostname) $MOUNT
+ fail $SINGLEMDS
+ zconf_mount $(hostname) $MOUNT
+ mount_facet ost1
+ touch $DIR/$tfile || return 1
+ rm $DIR/$tfile || return 2
+ df $MOUNT || error "df failed: $?"
+ return 0
+}
+run_test 74 "Ensure applications don't fail waiting for OST reocvery"
test_80a() {
[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
$CHECKSTAT -t dir $DIR/$tdir || error "$CHECKSTAT -t dir $DIR/$tdir failed"
rmdir $DIR/$tdir || error "rmdir $DIR/$tdir failed"
fail mds2
- stat $DIR/$tdir
+ stat $DIR/$tdir 2&>/dev/null && error "$DIR/$tdir still exist after recovery!"
+ return 0
}
run_test 80a "CMD: unlink cross-node dir (fail mds with inode)"
$CHECKSTAT -t dir $DIR/$tdir || error "$CHECKSTAT -t dir $DIR/$tdir failed"
rmdir $DIR/$tdir || error "rmdir $DIR/$tdir failed"
fail mds1
- stat $DIR/$tdir
+ stat $DIR/$tdir 2&>/dev/null && error "$DIR/$tdir still exist after recovery!"
+ return 0
}
run_test 80b "CMD: unlink cross-node dir (fail mds with name)"
equals_msg `basename $0`: test complete, cleaning up
check_and_cleanup_lustre
-[ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG || true
+[ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG && grep -q FAIL $TESTSUITELOG && exit 1 || true
git://git.whamcloud.com / fs / lustre-release.git / blobdiff