# bug number: 6088 10124 10800
ALWAYS_EXCEPT="8 15c 17 $REPLAY_DUAL_EXCEPT"
+#
+[ "$SLOW" = "no" ] && EXCEPT_SLOW="1 2 3 4 5 14"
+
+SAVE_PWD=$PWD
PTLDEBUG=${PTLDEBUG:--1}
LUSTRE=${LUSTRE:-`dirname $0`/..}
+SETUP=${SETUP:-""}
+CLEANUP=${CLEANUP:-""}
+MOUNT_2=${MOUNT_2:-"yes"}
. $LUSTRE/tests/test-framework.sh
init_test_env $@
. ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
-SETUP=${SETUP:-"setup"}
-CLEANUP=${CLEANUP:-"cleanup"}
build_test_filter
-cleanup() {
- stopall
-}
-
-if [ "$ONLY" == "cleanup" ]; then
- sysctl -w lnet.debug=0
- cleanup
- exit
-fi
-
-setup() {
- cleanup
- formatall
- setupall
- mount_client $MOUNT2
-}
+cleanup_and_setup_lustre
+rm -rf $DIR/${TESTSUITE}/[df][0-9]* # bug 13798 new t-f tdir staff
+rm -rf $DIR/[df][0-9]*
-$SETUP
[ "$DAEMONFILE" ] && $LCTL debug_daemon start $DAEMONFILE $DAEMONSIZE
test_1() {
touch $MOUNT1/a
- replay_barrier mds
+ replay_barrier $SINGLEMDS
touch $MOUNT2/b
- fail mds
+ fail $SINGLEMDS
checkstat $MOUNT2/a || return 1
checkstat $MOUNT1/b || return 2
rm $MOUNT2/a $MOUNT1/b
test_2() {
- replay_barrier mds
+ replay_barrier $SINGLEMDS
mkdir $MOUNT1/adir
- fail mds
+ fail $SINGLEMDS
checkstat $MOUNT2/adir || return 1
rmdir $MOUNT2/adir
checkstat $MOUNT2/adir && return 2
return 0
}
-
run_test 2 "|X| mkdir adir"
test_3() {
- replay_barrier mds
+ replay_barrier $SINGLEMDS
mkdir $MOUNT1/adir
mkdir $MOUNT2/adir/bdir
- fail mds
+ fail $SINGLEMDS
checkstat $MOUNT2/adir || return 1
checkstat $MOUNT1/adir/bdir || return 2
rmdir $MOUNT2/adir/bdir $MOUNT1/adir
checkstat $MOUNT2/adir/bdir && return 4
return 0
}
-
run_test 3 "|X| mkdir adir, mkdir adir/bdir "
test_4() {
mkdir $MOUNT1/adir
- replay_barrier mds
+ replay_barrier $SINGLEMDS
mkdir $MOUNT1/adir && return 1
mkdir $MOUNT2/adir/bdir
- fail mds
+ fail $SINGLEMDS
checkstat $MOUNT2/adir || return 2
checkstat $MOUNT1/adir/bdir || return 3
checkstat $MOUNT2/adir/bdir && return 5
return 0
}
-
run_test 4 "|X| mkdir adir (-EEXIST), mkdir adir/bdir "
# give multiop a chance to open
sleep 1
rm -f $MOUNT1/a
- replay_barrier mds
+ replay_barrier $SINGLEMDS
kill -USR1 $pid
wait $pid || return 1
- fail mds
+ fail $SINGLEMDS
[ -e $MOUNT2/a ] && return 2
return 0
}
# give multiop a chance to open
sleep 1
rm -f $MOUNT1/a
- replay_barrier mds
+ replay_barrier $SINGLEMDS
kill -USR1 $pid1
wait $pid1 || return 1
- fail mds
+ fail $SINGLEMDS
kill -USR1 $pid2
wait $pid2 || return 1
[ -e $MOUNT2/a ] && return 2
return 0
}
-run_test 6 "open1, open2, unlink |X| close1 [fail mds] close2"
+run_test 6 "open1, open2, unlink |X| close1 [fail $SINGLEMDS] close2"
test_8() {
- replay_barrier mds
+ replay_barrier $SINGLEMDS
drop_reint_reply "mcreate $MOUNT1/$tfile" || return 1
- fail mds
+ fail $SINGLEMDS
checkstat $MOUNT2/$tfile || return 2
rm $MOUNT1/$tfile || return 3
run_test 8 "replay of resent request"
test_9() {
- replay_barrier mds
+ replay_barrier $SINGLEMDS
mcreate $MOUNT1/$tfile-1
mcreate $MOUNT2/$tfile-2
# drop first reint reply
- sysctl -w lustre.fail_loc=0x80000119
- fail mds
- sysctl -w lustre.fail_loc=0
+ do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0x80000119
+ fail $SINGLEMDS
+ do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0
rm $MOUNT1/$tfile-[1,2] || return 1
test_10() {
mcreate $MOUNT1/$tfile-1
- replay_barrier mds
+ replay_barrier $SINGLEMDS
munlink $MOUNT1/$tfile-1
mcreate $MOUNT2/$tfile-2
# drop first reint reply
- sysctl -w lustre.fail_loc=0x80000119
- fail mds
- sysctl -w lustre.fail_loc=0
+ do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0x80000119
+ fail $SINGLEMDS
+ do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0
checkstat $MOUNT1/$tfile-1 && return 1
checkstat $MOUNT1/$tfile-2 || return 2
run_test 10 "resending a replayed unlink"
test_11() {
- replay_barrier mds
+ replay_barrier $SINGLEMDS
mcreate $MOUNT1/$tfile-1
mcreate $MOUNT2/$tfile-2
mcreate $MOUNT1/$tfile-3
mcreate $MOUNT2/$tfile-4
mcreate $MOUNT1/$tfile-5
# drop all reint replies for a while
- sysctl -w lustre.fail_loc=0x0119
- facet_failover mds
+ do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0x0119
+ # note that with this fail_loc set, facet_failover df will fail
+ facet_failover $SINGLEMDS
#sleep for while, let both clients reconnect and timeout
sleep $((TIMEOUT * 2))
- sysctl -w lustre.fail_loc=0
+ do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0
rm $MOUNT1/$tfile-[1-5] || return 1
run_test 11 "both clients timeout during replay"
test_12() {
- replay_barrier mds
+ replay_barrier $SINGLEMDS
multiop $DIR/$tfile mo_c &
MULTIPID=$!
sleep 5
#define OBD_FAIL_LDLM_ENQUEUE 0x302
- sysctl -w lustre.fail_loc=0x80000302
- facet_failover mds
+ do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0x80000302
+ facet_failover $SINGLEMDS
+ do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0
df $MOUNT || return 1
- sysctl -w lustre.fail_loc=0
ls $DIR/$tfile
kill -USR1 $MULTIPID || return 3
MULTIPID=$!
sleep 5
- replay_barrier mds
+ replay_barrier $SINGLEMDS
kill -USR1 $MULTIPID || return 3
wait $MULTIPID || return 4
# drop close
- sysctl -w lustre.fail_loc=0x80000115
- facet_failover mds
+ do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0x80000115
+ facet_failover $SINGLEMDS
+ do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0
df $MOUNT || return 1
- sysctl -w lustre.fail_loc=0
ls $DIR/$tfile
$CHECKSTAT -t file $DIR/$tfile || return 2
run_test 13 "close resend timeout"
test_14() {
- replay_barrier mds
+ replay_barrier $SINGLEMDS
createmany -o $MOUNT1/$tfile- 25
createmany -o $MOUNT2/$tfile-2- 1
createmany -o $MOUNT1/$tfile-3- 25
umount $MOUNT2
- facet_failover mds
- # expect failover to fail
+ facet_failover $SINGLEMDS
+ # expect failover to fail due to missing client 2
df $MOUNT && return 1
sleep 1
# first 25 files should have been replayed
unlinkmany $MOUNT1/$tfile- 25 || return 2
- zconf_mount `hostname` $MOUNT2
+ zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
return 0
}
run_test 14 "timeouts waiting for lost client during replay"
test_15() {
- replay_barrier mds
+ replay_barrier $SINGLEMDS
createmany -o $MOUNT1/$tfile- 25
createmany -o $MOUNT2/$tfile-2- 1
umount $MOUNT2
- facet_failover mds
+ facet_failover $SINGLEMDS
df $MOUNT || return 1
unlinkmany $MOUNT1/$tfile- 25 || return 2
[ -e $MOUNT1/$tfile-2-0 ] && error "$tfile-2-0 exists"
- zconf_mount `hostname` $MOUNT2
+ zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
return 0
}
run_test 15 "timeout waiting for lost client during replay, 1 client completes"
local ost_last_id=""
local osc_last_id=""
- replay_barrier mds
+ replay_barrier $SINGLEMDS
echo "data" > "$MOUNT2/${tfile}-m2"
umount $MOUNT2
- facet_failover mds
+ facet_failover $SINGLEMDS
df $MOUNT || return 1
ost_last_id=`cat /proc/fs/lustre/obdfilter/*/last_id`
}
done
- zconf_mount `hostname` $MOUNT2
+ zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
return 0
}
#CROW run_test 15a "OST clear orphans - synchronize ids on MDS and OST"
test_15b() {
- replay_barrier mds
+ replay_barrier $SINGLEMDS
echo "data" > "$MOUNT2/${tfile}-m2"
umount $MOUNT2
do_facet ost1 "sysctl -w lustre.fail_loc=0x80000802"
- facet_failover mds
+ facet_failover $SINGLEMDS
df $MOUNT || return 1
do_facet ost1 "sysctl -w lustre.fail_loc=0"
- zconf_mount `hostname` $MOUNT2
+ zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
return 0
}
#CROW run_test 15b "multiple delayed OST clear orphans"
test_15c() {
- replay_barrier mds
+ replay_barrier $SINGLEMDS
for ((i = 0; i < 2000; i++)); do
echo "data" > "$MOUNT2/${tfile}-$i" || error "create ${tfile}-$i failed"
done
umount $MOUNT2
- facet_failover mds
+ facet_failover $SINGLEMDS
df $MOUNT || return 1
- zconf_mount `hostname` $MOUNT2
+ zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
return 0
}
run_test 15c "remove multiple OST orphans"
test_16() {
- replay_barrier mds
+ replay_barrier $SINGLEMDS
createmany -o $MOUNT1/$tfile- 25
createmany -o $MOUNT2/$tfile-2- 1
umount $MOUNT2
- facet_failover mds
+ facet_failover $SINGLEMDS
sleep $TIMEOUT
- facet_failover mds
+ facet_failover $SINGLEMDS
df $MOUNT || return 1
unlinkmany $MOUNT1/$tfile- 25 || return 2
- zconf_mount `hostname` $MOUNT2
+ zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
return 0
}
unlinkmany $MOUNT1/$tfile- 25 || return 2
- zconf_mount `hostname` $MOUNT2
+ zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
return 0
}
statmany -s $MOUNT1/$tdir/f 1 500 &
OPENPID=$!
NOW=`date +%s`
- do_facet mds sysctl -w lustre.fail_loc=0x8000030b # hold enqueue
+ do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0x8000030b # hold enqueue
sleep 1
#define OBD_FAIL_LDLM_BL_CALLBACK 0x305
do_facet client sysctl -w lustre.fail_loc=0x80000305 # drop cb, evict
run_test 18 "ldlm_handle_enqueue succeeds on evicted export (3822)"
test_19() { # Bug 10991 - resend of open request does not fail assertion.
- replay_barrier mds
+ replay_barrier $SINGLEMDS
drop_ldlm_reply "createmany -o $DIR/$tfile 1" || return 1
- fail mds
+ fail $SINGLEMDS
checkstat $DIR2/${tfile}0 || return 2
rm $DIR/${tfile}0 || return 3
}
run_test 19 "resend of open request"
-if [ "$ONLY" != "setup" ]; then
- equals_msg test complete, cleaning up
- SLEEP=$((`date +%s` - $NOW))
- [ $SLEEP -lt $TIMEOUT ] && sleep $SLEEP
- $CLEANUP
-fi
-echo "$0: completed"
+equals_msg `basename $0`: test complete, cleaning up
+SLEEP=$((`date +%s` - $NOW))
+[ $SLEEP -lt $TIMEOUT ] && sleep $SLEEP
+check_and_cleanup_lustre
+[ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG || true
+