6 ALWAYS_EXCEPT="8 $REPLAY_DUAL_EXCEPT"
8 LUSTRE=${LUSTRE:-`dirname $0`/..}
9 . $LUSTRE/tests/test-framework.sh
13 . ${CONFIG:=$LUSTRE/tests/cfg/local.sh}
15 SETUP=${SETUP:-"setup"}
16 CLEANUP=${CLEANUP:-"cleanup"}
17 FORCE=${FORCE:-"--force"}
21 add_mds mds --dev $MDSDEV --size $MDSSIZE
22 if [ ! -z "$mdsfailover_HOST" ]; then
23 add_mdsfailover mds --dev $MDSDEV --size $MDSSIZE
26 add_lov lov1 mds --stripe_sz $STRIPE_BYTES \
27 --stripe_cnt $STRIPES_PER_OBJ --stripe_pattern 0
28 add_ost ost --lov lov1 --dev $OSTDEV --size $OSTSIZE --failover
29 add_ost ost2 --lov lov1 --dev ${OSTDEV}-2 --size $OSTSIZE --failover
30 add_client client mds --lov lov1 --path $MOUNT
38 # make sure we are using the primary MDS, so the config log will
39 # be able to clean up properly.
40 activemds=`facet_active mds`
41 if [ $activemds != "mds" ]; then
45 umount $MOUNT2 || true
50 stop ost ${FORCE} --dump $TMP/replay-dual-`hostname`.log
53 if [ "$ONLY" == "cleanup" ]; then
54 sysctl -w lnet.debug=0
61 start ost --reformat $OSTLCONFARGS
62 start ost2 --reformat $OSTLCONFARGS
63 start mds $MDSLCONFARGS --reformat
64 grep " $MOUNT " /proc/mounts || zconf_mount `hostname` $MOUNT
65 grep " $MOUNT2 " /proc/mounts || zconf_mount `hostname` $MOUNT2
67 # echo $TIMEOUT > /proc/sys/lustre/timeout
71 [ "$DAEMONFILE" ] && $LCTL debug_daemon start $DAEMONFILE $DAEMONSIZE
79 checkstat $MOUNT2/a || return 1
80 checkstat $MOUNT1/b || return 2
81 rm $MOUNT2/a $MOUNT1/b
82 checkstat $MOUNT1/a && return 3
83 checkstat $MOUNT2/b && return 4
87 run_test 1 "|X| simple create"
95 checkstat $MOUNT2/adir || return 1
97 checkstat $MOUNT2/adir && return 2
101 run_test 2 "|X| mkdir adir"
106 mkdir $MOUNT2/adir/bdir
109 checkstat $MOUNT2/adir || return 1
110 checkstat $MOUNT1/adir/bdir || return 2
111 rmdir $MOUNT2/adir/bdir $MOUNT1/adir
112 checkstat $MOUNT1/adir && return 3
113 checkstat $MOUNT2/adir/bdir && return 4
117 run_test 3 "|X| mkdir adir, mkdir adir/bdir "
122 mkdir $MOUNT1/adir && return 1
123 mkdir $MOUNT2/adir/bdir
126 checkstat $MOUNT2/adir || return 2
127 checkstat $MOUNT1/adir/bdir || return 3
129 rmdir $MOUNT2/adir/bdir $MOUNT1/adir
130 checkstat $MOUNT1/adir && return 4
131 checkstat $MOUNT2/adir/bdir && return 5
135 run_test 4 "|X| mkdir adir (-EEXIST), mkdir adir/bdir "
139 # multiclient version of replay_single.sh/test_8
141 multiop $MOUNT2/a o_tSc &
143 # give multiop a chance to open
148 wait $pid || return 1
151 [ -e $MOUNT2/a ] && return 2
154 run_test 5 "open, unlink |X| close"
159 multiop $MOUNT2/a o_c &
161 multiop $MOUNT1/a o_c &
163 # give multiop a chance to open
168 wait $pid1 || return 1
172 wait $pid2 || return 1
173 [ -e $MOUNT2/a ] && return 2
176 run_test 6 "open1, open2, unlink |X| close1 [fail mds] close2"
180 drop_reint_reply "mcreate $MOUNT1/$tfile" || return 1
182 checkstat $MOUNT2/$tfile || return 2
183 rm $MOUNT1/$tfile || return 3
187 run_test 8 "replay of resent request"
191 mcreate $MOUNT1/$tfile-1
192 mcreate $MOUNT2/$tfile-2
193 # drop first reint reply
194 sysctl -w lustre.fail_loc=0x80000119
196 sysctl -w lustre.fail_loc=0
198 rm $MOUNT1/$tfile-[1,2] || return 1
202 run_test 9 "resending a replayed create"
205 mcreate $MOUNT1/$tfile-1
207 munlink $MOUNT1/$tfile-1
208 mcreate $MOUNT2/$tfile-2
209 # drop first reint reply
210 sysctl -w lustre.fail_loc=0x80000119
212 sysctl -w lustre.fail_loc=0
214 checkstat $MOUNT1/$tfile-1 && return 1
215 checkstat $MOUNT1/$tfile-2 || return 2
220 run_test 10 "resending a replayed unlink"
224 mcreate $MOUNT1/$tfile-1
225 mcreate $MOUNT2/$tfile-2
226 mcreate $MOUNT1/$tfile-3
227 mcreate $MOUNT2/$tfile-4
228 mcreate $MOUNT1/$tfile-5
229 # drop all reint replies for a while
230 sysctl -w lustre.fail_loc=0x0119
232 #sleep for while, let both clients reconnect and timeout
233 sleep $((TIMEOUT * 2))
234 sysctl -w lustre.fail_loc=0
236 rm $MOUNT1/$tfile-[1-5] || return 1
240 run_test 11 "both clients timeout during replay"
245 multiop $DIR/$tfile mo_c &
249 #define OBD_FAIL_LDLM_ENQUEUE 0x302
250 sysctl -w lustre.fail_loc=0x80000302
252 df $MOUNT || return 1
253 sysctl -w lustre.fail_loc=0
256 kill -USR1 $MULTIPID || return 3
257 wait $MULTIPID || return 4
258 $CHECKSTAT -t file $DIR/$tfile || return 2
263 run_test 12 "open resend timeout"
266 multiop $DIR/$tfile mo_c &
272 kill -USR1 $MULTIPID || return 3
273 wait $MULTIPID || return 4
276 sysctl -w lustre.fail_loc=0x80000115
278 df $MOUNT || return 1
279 sysctl -w lustre.fail_loc=0
282 $CHECKSTAT -t file $DIR/$tfile || return 2
287 run_test 13 "close resend timeout"
291 createmany -o $MOUNT1/$tfile- 25
292 createmany -o $MOUNT2/$tfile-2- 1
293 createmany -o $MOUNT1/$tfile-3- 25
297 # expect failover to fail
298 df $MOUNT && return 1
301 # first 25 files should have been replayed
302 unlinkmany $MOUNT1/$tfile- 25 || return 2
304 zconf_mount `hostname` $MOUNT2
307 run_test 14 "timeouts waiting for lost client during replay"
311 createmany -o $MOUNT1/$tfile- 25
312 createmany -o $MOUNT2/$tfile-2- 1
316 df $MOUNT || return 1
318 unlinkmany $MOUNT1/$tfile- 25 || return 2
319 [ -e $MOUNT1/$tfile-2-0 ] && error "$tfile-2-0 exists"
321 zconf_mount `hostname` $MOUNT2
324 run_test 15 "timeout waiting for lost client during replay, 1 client completes"
331 echo "data" > "$MOUNT2/${tfile}-m2"
335 df $MOUNT || return 1
337 ost_last_id=`cat /proc/fs/lustre/obdfilter/*/last_id`
338 mds_last_id=`cat /proc/fs/lustre/osc/*mds*/last_id`
340 echo "Ids after MDS<->OST synchonizing"
341 echo "--------------------------------"
348 echo $ost_last_id | while read id; do
354 echo $mds_last_id | while read id; do
359 local arr_len=${#mds_ids[*]}
360 for ((i=0;i<$arr_len;i++)); do
364 test $mds_id -ge $ost_id || {
365 echo "MDS last id ($mds_id) is smaller than OST one ($ost_id)"
370 zconf_mount `hostname` $MOUNT2
373 #CROW run_test 15a "OST clear orphans - synchronize ids on MDS and OST"
377 echo "data" > "$MOUNT2/${tfile}-m2"
380 do_facet ost "sysctl -w lustre.fail_loc=0x80000802"
383 df $MOUNT || return 1
384 do_facet ost "sysctl -w lustre.fail_loc=0"
386 zconf_mount `hostname` $MOUNT2
389 #CROW run_test 15b "multiple delayed OST clear orphans"
393 for ((i = 0; i < 2000; i++)); do
394 echo "data" > "$MOUNT2/${tfile}-$i" || error "create ${tfile}-$i failed"
400 df $MOUNT || return 1
402 zconf_mount `hostname` $MOUNT2
405 run_test 15c "remove multiple OST orphans"
409 createmany -o $MOUNT1/$tfile- 25
410 createmany -o $MOUNT2/$tfile-2- 1
416 df $MOUNT || return 1
418 unlinkmany $MOUNT1/$tfile- 25 || return 2
420 zconf_mount `hostname` $MOUNT2
424 run_test 16 "fail MDS during recovery (3571)"
427 createmany -o $MOUNT1/$tfile- 25
428 createmany -o $MOUNT2/$tfile-2- 1
430 # Make sure the disconnect is lost
437 df $MOUNT || return 1
439 unlinkmany $MOUNT1/$tfile- 25 || return 2
441 zconf_mount `hostname` $MOUNT2
445 run_test 17 "fail OST during recovery (3571)"
447 # cleanup with blocked enqueue fails until timer elapses (MDS busy), wait for it
450 test_18() { # bug 3822 - evicting client with enqueued lock
452 mkdir -p $MOUNT1/$tdir
453 touch $MOUNT1/$tdir/f0
454 #define OBD_FAIL_LDLM_ENQUEUE_BLOCKED 0x30b
455 statmany -s $MOUNT1/$tdir/f 1 500 &
458 do_facet mds sysctl -w lustre.fail_loc=0x8000030b # hold enqueue
460 #define OBD_FAIL_LDLM_BL_CALLBACK 0x305
461 do_facet client sysctl -w lustre.fail_loc=0x80000305 # drop cb, evict
463 usleep 500 # wait to ensure first client is one that will be evicted
464 openfile -f O_RDONLY $MOUNT2/$tdir/f0
466 dmesg | grep "entering recovery in server" && \
467 error "client not evicted" || true
469 run_test 18 "ldlm_handle_enqueue succeeds on evicted export (3822)"
471 if [ "$ONLY" != "setup" ]; then
472 equals_msg test complete, cleaning up
473 SLEEP=$((`date +%s` - $NOW))
474 [ $SLEEP -lt $TIMEOUT ] && sleep $SLEEP
475 FORCE=--force $CLEANUP