5 LUSTRE=${LUSTRE:-`dirname $0`/..}
6 . $LUSTRE/tests/test-framework.sh
10 . ${CONFIG:=$LUSTRE/tests/cfg/lmv.sh}
12 SETUP=${SETUP:-"setup"}
13 CLEANUP=${CLEANUP:-"cleanup"}
17 if [ "$MDSCOUNT" -gt 1 ]; then
19 for mds in `mds_list`; do
20 MDSDEV=$TMP/${mds}-`hostname`
21 add_mds $mds --dev $MDSDEV --size $MDSSIZE --lmv lmv1_svc
23 add_lov_to_lmv lov1 lmv1_svc --stripe_sz $STRIPE_BYTES \
24 --stripe_cnt $STRIPES_PER_OBJ --stripe_pattern 0
27 add_mds mds1 --dev $MDSDEV --size $MDSSIZE
28 add_lov lov1 mds1 --stripe_sz $STRIPE_BYTES \
29 --stripe_cnt $STRIPES_PER_OBJ --stripe_pattern 0
33 add_ost ost --lov lov1 --dev $OSTDEV --size $OSTSIZE --failover
34 add_ost ost2 --lov lov1 --dev ${OSTDEV}-2 --size $OSTSIZE --failover
35 add_client client ${MDS} --lov lov1 --path $MOUNT
40 SETUP=${SETUP:-"setup"}
41 CLEANUP=${CLEANUP:-"cleanup"}
44 # make sure we are using the primary MDS, so the config log will
45 # be able to clean up properly.
46 activemds=`facet_active mds1`
47 if [ $activemds != "mds1" ]; then
51 umount $MOUNT2 || true
54 for mds in `mds_list`; do
55 stop $mds ${FORCE} $MDSLCONFARGS
58 stop ost ${FORCE} --dump cleanup-dual.log
61 if [ "$ONLY" == "cleanup" ]; then
62 sysctl -w portals.debug=0
69 start ost --reformat $OSTLCONFARGS
70 PINGER=`cat /proc/fs/lustre/pinger`
72 if [ "$PINGER" != "on" ]; then
73 echo "ERROR: Lustre must be built with --enable-pinger for replay-dual"
78 start ost2 --reformat $OSTLCONFARGS
79 [ "$DAEMONFILE" ] && $LCTL debug_daemon start $DAEMONFILE $DAEMONSIZE
80 for mds in `mds_list`; do
81 start $mds --reformat $MDSLCONFARGS
83 grep " $MOUNT " /proc/mounts || zconf_mount `hostname` $MOUNT
84 grep " $MOUNT2 " /proc/mounts || zconf_mount `hostname` $MOUNT2
86 echo $TIMEOUT > /proc/sys/lustre/timeout
87 echo $UPCALL > /proc/sys/lustre/upcall
91 [ "$DAEMONFILE" ] && $LCTL debug_daemon start $DAEMONFILE $DAEMONSIZE
100 checkstat $MOUNT2/a || return 1
101 checkstat $MOUNT1/b || return 2
102 rm $MOUNT2/a $MOUNT1/b
103 checkstat $MOUNT1/a && return 3
104 checkstat $MOUNT2/b && return 4
108 run_test 1 "|X| simple create"
116 checkstat $MOUNT2/adir || return 1
118 checkstat $MOUNT2/adir && return 2
122 run_test 2 "|X| mkdir adir"
127 mkdir $MOUNT2/adir/bdir
130 checkstat $MOUNT2/adir || return 1
131 checkstat $MOUNT1/adir/bdir || return 2
132 rmdir $MOUNT2/adir/bdir $MOUNT1/adir
133 checkstat $MOUNT1/adir && return 3
134 checkstat $MOUNT2/adir/bdir && return 4
138 run_test 3 "|X| mkdir adir, mkdir adir/bdir "
143 mkdir $MOUNT1/adir && return 1
144 mkdir $MOUNT2/adir/bdir
147 checkstat $MOUNT2/adir || return 2
148 checkstat $MOUNT1/adir/bdir || return 3
150 rmdir $MOUNT2/adir/bdir $MOUNT1/adir
151 checkstat $MOUNT1/adir && return 4
152 checkstat $MOUNT2/adir/bdir && return 5
156 run_test 4 "|X| mkdir adir (-EEXIST), mkdir adir/bdir "
160 # multiclient version of replay_single.sh/test_8
162 multiop $MOUNT2/a o_tSc &
164 # give multiop a chance to open
169 wait $pid || return 1
172 [ -e $MOUNT2/a ] && return 2
175 run_test 5 "open, unlink |X| close"
180 multiop $MOUNT2/a o_c &
182 multiop $MOUNT1/a o_c &
184 # give multiop a chance to open
189 wait $pid1 || return 1
193 wait $pid2 || return 1
194 [ -e $MOUNT2/a ] && return 2
197 run_test 6 "open1, open2, unlink |X| close1 [fail mds] close2"
201 multiop $MOUNT2/a o_c &
203 multiop $MOUNT1/a o_c &
205 # give multiop a chance to open
210 wait $pid2 || return 1
214 wait $pid1 || return 1
215 [ -e $MOUNT2/a ] && return 2
218 run_test 6b "open1, open2, unlink |X| close2 [fail mds] close1"
222 drop_reint_reply "mcreate $MOUNT1/$tfile" || return 1
224 checkstat $MOUNT2/$tfile || return 2
225 rm $MOUNT1/$tfile || return 3
229 run_test 8 "replay of resent request"
233 mcreate $MOUNT1/$tfile-1
234 mcreate $MOUNT2/$tfile-2
235 # drop first reint reply
236 sysctl -w lustre.fail_loc=0x80000119
238 sysctl -w lustre.fail_loc=0
240 rm $MOUNT1/$tfile-[1,2] || return 1
244 run_test 9 "resending a replayed create"
247 mcreate $MOUNT1/$tfile-1
249 munlink $MOUNT1/$tfile-1
250 mcreate $MOUNT2/$tfile-2
251 # drop first reint reply
252 sysctl -w lustre.fail_loc=0x80000119
254 sysctl -w lustre.fail_loc=0
256 checkstat $MOUNT1/$tfile-1 && return 1
257 checkstat $MOUNT1/$tfile-2 || return 2
262 run_test 10 "resending a replayed unlink"
266 mcreate $MOUNT1/$tfile-1
267 mcreate $MOUNT2/$tfile-2
268 mcreate $MOUNT1/$tfile-3
269 mcreate $MOUNT2/$tfile-4
270 mcreate $MOUNT1/$tfile-5
271 # drop all reint replies for a while
272 sysctl -w lustre.fail_loc=0x0119
274 #sleep for while, let both clients reconnect and timeout
275 sleep $((TIMEOUT * 2))
276 sysctl -w lustre.fail_loc=0
278 rm $MOUNT1/$tfile-[1-5] || return 1
282 run_test 11 "both clients timeout during replay"
287 multiop $DIR/$tfile mo_c &
292 sysctl -w lustre.fail_loc=0x80000302
294 df $MOUNT || return 1
295 sysctl -w lustre.fail_loc=0
298 $CHECKSTAT -t file $DIR/$tfile || return 2
299 kill -USR1 $MULTIPID || return 3
300 wait $MULTIPID || return 4
305 run_test 12 "open resend timeout"
308 multiop $DIR/$tfile mo_c &
314 kill -USR1 $MULTIPID || return 3
315 wait $MULTIPID || return 4
318 sysctl -w lustre.fail_loc=0x80000115
320 df $MOUNT || return 1
321 sysctl -w lustre.fail_loc=0
324 $CHECKSTAT -t file $DIR/$tfile || return 2
329 run_test 13 "close resend timeout"
334 createmany -o $MOUNT1/$tfile- 25
335 createmany -o $MOUNT2/$tfile-2- 1
336 createmany -o $MOUNT1/$tfile-3- 25
340 # expect failover to fail
341 df $MOUNT && return 1
343 # first 25 files shouuld have been
346 unlinkmany $MOUNT1/$tfile- 25 || return 2
348 zconf_mount `hostname` $MOUNT2
351 run_test 14 "timeouts waiting for lost client during replay"
355 createmany -o $MOUNT1/$tfile- 25
356 createmany -o $MOUNT2/$tfile-2- 1
360 df $MOUNT || return 1
362 unlinkmany $MOUNT1/$tfile- 25 || return 2
364 zconf_mount `hostname` $MOUNT2
367 run_test 15 "timeout waiting for lost client during replay, 1 client completes"
370 createmany -o $MOUNT1/$tfile- 25
371 createmany -o $MOUNT2/$tfile-2- 1
377 df $MOUNT || return 1
379 unlinkmany $MOUNT1/$tfile- 25 || return 2
381 zconf_mount `hostname` $MOUNT2
385 #run_test 16 "fail MDS during recovery (3571)"
388 createmany -o $MOUNT1/$tfile- 25
389 createmany -o $MOUNT2/$tfile-2- 1
391 # Make sure the disconnect is lost
395 echo -1 > /proc/sys/portals/debug
399 df $MOUNT || return 1
401 unlinkmany $MOUNT1/$tfile- 25 || return 2
403 zconf_mount `hostname` $MOUNT2
407 #Still not support ost fail over
408 #run_test 17 "fail OST during recovery (3571)"
412 multiop $MOUNT2/$tfile O_c &
414 multiop $MOUNT1/$tfile O_c &
416 # give multiop a chance to open
424 zconf_mount `hostname` $MOUNT2
426 run_test 18 "replay open, Abort recovery, don't assert (3892)"
429 # cleanup with blocked enqueue fails until timer elapses (MDS busy), wait for
432 test_20() { # bug 3822 - evicting client with enqueued lock
433 mkdir -p $MOUNT1/$tdir
434 touch $MOUNT1/$tdir/f0
435 #define OBD_FAIL_LDLM_ENQUEUE_BLOCKED 0x30b
436 statmany -s $MOUNT1/$tdir/f 500 &
439 do_facet mds1 sysctl -w lustre.fail_loc=0x8000030b # hold enqueue
441 #define OBD_FAIL_LDLM_BL_CALLBACK 0x305
442 do_facet client sysctl -w lustre.fail_loc=0x80000305 # drop cb, evict
444 usleep 500 # wait to ensure first client is one that will be evicted
445 openfile -f O_RDONLY $MOUNT2/$tdir/f0
447 dmesg | grep "entering recovery in server" && \
448 error "client not evicted" || true
450 run_test 20 "ldlm_handle_enqueue succeeds on evicted export (3822)"
452 if [ "$ONLY" != "setup" ]; then
453 equals_msg test complete, cleaning up
455 SLEEP=$((`date +%s` - $NOW))
456 [ $SLEEP -lt $TIMEOUT ] && sleep $SLEEP