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
34 add_ost ost --lov lov1 --dev $OSTDEV --size $OSTSIZE --failover
35 add_ost ost2 --lov lov1 --dev ${OSTDEV}-2 --size $OSTSIZE --failover
36 add_client client ${MDS} --lov lov1 --path $MOUNT
43 SETUP=${SETUP:-"setup"}
44 CLEANUP=${CLEANUP:-"cleanup"}
47 # make sure we are using the primary MDS, so the config log will
48 # be able to clean up properly.
49 activemds=`facet_active mds1`
50 if [ $activemds != "mds1" ]; then
54 umount $MOUNT2 || true
57 for mds in `mds_list`; do
58 stop $mds ${FORCE} $MDSLCONFARGS
61 stop ost ${FORCE} --dump cleanup-dual.log
64 if [ "$ONLY" == "cleanup" ]; then
65 sysctl -w portals.debug=0
72 start ost --reformat $OSTLCONFARGS
73 PINGER=`cat /proc/fs/lustre/pinger`
75 if [ "$PINGER" != "on" ]; then
76 echo "ERROR: Lustre must be built with --enable-pinger for replay-dual"
81 start ost2 --reformat $OSTLCONFARGS
82 [ "$DAEMONFILE" ] && $LCTL debug_daemon start $DAEMONFILE $DAEMONSIZE
83 for mds in `mds_list`; do
84 start $mds --reformat $MDSLCONFARGS
86 grep " $MOUNT " /proc/mounts || zconf_mount `hostname` $MOUNT
87 grep " $MOUNT2 " /proc/mounts || zconf_mount `hostname` $MOUNT2
89 echo $TIMEOUT > /proc/sys/lustre/timeout
90 echo $UPCALL > /proc/sys/lustre/upcall
94 [ "$DAEMONFILE" ] && $LCTL debug_daemon start $DAEMONFILE $DAEMONSIZE
103 checkstat $MOUNT2/a || return 1
104 checkstat $MOUNT1/b || return 2
105 rm $MOUNT2/a $MOUNT1/b
106 checkstat $MOUNT1/a && return 3
107 checkstat $MOUNT2/b && return 4
111 run_test 1 "|X| simple create"
119 checkstat $MOUNT2/adir || return 1
121 checkstat $MOUNT2/adir && return 2
125 run_test 2 "|X| mkdir adir"
130 mkdir $MOUNT2/adir/bdir
133 checkstat $MOUNT2/adir || return 1
134 checkstat $MOUNT1/adir/bdir || return 2
135 rmdir $MOUNT2/adir/bdir $MOUNT1/adir
136 checkstat $MOUNT1/adir && return 3
137 checkstat $MOUNT2/adir/bdir && return 4
141 run_test 3 "|X| mkdir adir, mkdir adir/bdir "
146 mkdir $MOUNT1/adir && return 1
147 mkdir $MOUNT2/adir/bdir
150 checkstat $MOUNT2/adir || return 2
151 checkstat $MOUNT1/adir/bdir || return 3
153 rmdir $MOUNT2/adir/bdir $MOUNT1/adir
154 checkstat $MOUNT1/adir && return 4
155 checkstat $MOUNT2/adir/bdir && return 5
159 run_test 4 "|X| mkdir adir (-EEXIST), mkdir adir/bdir "
163 # multiclient version of replay_single.sh/test_8
165 multiop $MOUNT2/a o_tSc &
167 # give multiop a chance to open
172 wait $pid || return 1
175 [ -e $MOUNT2/a ] && return 2
178 run_test 5 "open, unlink |X| close"
183 multiop $MOUNT2/a o_c &
185 multiop $MOUNT1/a o_c &
187 # give multiop a chance to open
192 wait $pid1 || return 1
196 wait $pid2 || return 1
197 [ -e $MOUNT2/a ] && return 2
200 run_test 6 "open1, open2, unlink |X| close1 [fail mds] close2"
204 multiop $MOUNT2/a o_c &
206 multiop $MOUNT1/a o_c &
208 # give multiop a chance to open
213 wait $pid2 || return 1
217 wait $pid1 || return 1
218 [ -e $MOUNT2/a ] && return 2
221 run_test 6b "open1, open2, unlink |X| close2 [fail mds] close1"
225 drop_reint_reply "mcreate $MOUNT1/$tfile" || return 1
227 checkstat $MOUNT2/$tfile || return 2
228 rm $MOUNT1/$tfile || return 3
232 run_test 8 "replay of resent request"
236 mcreate $MOUNT1/$tfile-1
237 mcreate $MOUNT2/$tfile-2
238 # drop first reint reply
239 sysctl -w lustre.fail_loc=0x80000119
241 sysctl -w lustre.fail_loc=0
243 rm $MOUNT1/$tfile-[1,2] || return 1
247 run_test 9 "resending a replayed create"
250 mcreate $MOUNT1/$tfile-1
252 munlink $MOUNT1/$tfile-1
253 mcreate $MOUNT2/$tfile-2
254 # drop first reint reply
255 sysctl -w lustre.fail_loc=0x80000119
257 sysctl -w lustre.fail_loc=0
259 checkstat $MOUNT1/$tfile-1 && return 1
260 checkstat $MOUNT1/$tfile-2 || return 2
265 run_test 10 "resending a replayed unlink"
269 mcreate $MOUNT1/$tfile-1
270 mcreate $MOUNT2/$tfile-2
271 mcreate $MOUNT1/$tfile-3
272 mcreate $MOUNT2/$tfile-4
273 mcreate $MOUNT1/$tfile-5
274 # drop all reint replies for a while
275 sysctl -w lustre.fail_loc=0x0119
277 #sleep for while, let both clients reconnect and timeout
278 sleep $((TIMEOUT * 2))
279 sysctl -w lustre.fail_loc=0
281 rm $MOUNT1/$tfile-[1-5] || return 1
285 run_test 11 "both clients timeout during replay"
290 multiop $DIR/$tfile mo_c &
295 sysctl -w lustre.fail_loc=0x80000302
297 df $MOUNT || return 1
298 sysctl -w lustre.fail_loc=0
301 $CHECKSTAT -t file $DIR/$tfile || return 2
302 kill -USR1 $MULTIPID || return 3
303 wait $MULTIPID || return 4
308 run_test 12 "open resend timeout"
311 multiop $DIR/$tfile mo_c &
317 kill -USR1 $MULTIPID || return 3
318 wait $MULTIPID || return 4
321 sysctl -w lustre.fail_loc=0x80000115
323 df $MOUNT || return 1
324 sysctl -w lustre.fail_loc=0
327 $CHECKSTAT -t file $DIR/$tfile || return 2
332 run_test 13 "close resend timeout"
337 createmany -o $MOUNT1/$tfile- 25
338 createmany -o $MOUNT2/$tfile-2- 1
339 createmany -o $MOUNT1/$tfile-3- 25
343 # expect failover to fail
344 df $MOUNT && return 1
346 # first 25 files shouuld have been
349 unlinkmany $MOUNT1/$tfile- 25 || return 2
351 zconf_mount `hostname` $MOUNT2
354 run_test 14 "timeouts waiting for lost client during replay"
358 createmany -o $MOUNT1/$tfile- 25
359 createmany -o $MOUNT2/$tfile-2- 1
363 df $MOUNT || return 1
365 unlinkmany $MOUNT1/$tfile- 25 || return 2
367 zconf_mount `hostname` $MOUNT2
370 run_test 15 "timeout waiting for lost client during replay, 1 client completes"
373 createmany -o $MOUNT1/$tfile- 25
374 createmany -o $MOUNT2/$tfile-2- 1
380 df $MOUNT || return 1
382 unlinkmany $MOUNT1/$tfile- 25 || return 2
384 zconf_mount `hostname` $MOUNT2
388 #run_test 16 "fail MDS during recovery (3571)"
391 createmany -o $MOUNT1/$tfile- 25
392 createmany -o $MOUNT2/$tfile-2- 1
394 # Make sure the disconnect is lost
398 echo -1 > /proc/sys/portals/debug
402 df $MOUNT || return 1
404 unlinkmany $MOUNT1/$tfile- 25 || return 2
406 zconf_mount `hostname` $MOUNT2
410 #Still not support ost fail over
411 #run_test 17 "fail OST during recovery (3571)"
415 multiop $MOUNT2/$tfile O_c &
417 multiop $MOUNT1/$tfile O_c &
419 # give multiop a chance to open
427 zconf_mount `hostname` $MOUNT2
429 run_test 18 "replay open, Abort recovery, don't assert (3892)"
432 # cleanup with blocked enqueue fails until timer elapses (MDS busy), wait for
435 test_20() { # bug 3822 - evicting client with enqueued lock
436 mkdir -p $MOUNT1/$tdir
437 touch $MOUNT1/$tdir/f0
438 #define OBD_FAIL_LDLM_ENQUEUE_BLOCKED 0x30b
439 statmany -s $MOUNT1/$tdir/f 500 &
442 do_facet mds1 sysctl -w lustre.fail_loc=0x8000030b # hold enqueue
444 #define OBD_FAIL_LDLM_BL_CALLBACK 0x305
445 do_facet client sysctl -w lustre.fail_loc=0x80000305 # drop cb, evict
447 usleep 500 # wait to ensure first client is one that will be evicted
448 openfile -f O_RDONLY $MOUNT2/$tdir/f0
450 dmesg | grep "entering recovery in server" && \
451 error "client not evicted" || true
453 run_test 20 "ldlm_handle_enqueue succeeds on evicted export (3822)"
455 if [ "$ONLY" != "setup" ]; then
456 equals_msg test complete, cleaning up
458 SLEEP=$((`date +%s` - $NOW))
459 [ $SLEEP -lt $TIMEOUT ] && sleep $SLEEP