5 # bug number: 6088 10124 10800
6 ALWAYS_EXCEPT="8 15c 17 $REPLAY_DUAL_EXCEPT"
8 PTLDEBUG=${PTLDEBUG:--1}
9 LUSTRE=${LUSTRE:-`dirname $0`/..}
10 . $LUSTRE/tests/test-framework.sh
14 . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
16 SETUP=${SETUP:-"setup"}
17 CLEANUP=${CLEANUP:-"cleanup"}
25 if [ "$ONLY" == "cleanup" ]; then
26 sysctl -w lnet.debug=0
39 [ "$DAEMONFILE" ] && $LCTL debug_daemon start $DAEMONFILE $DAEMONSIZE
47 checkstat $MOUNT2/a || return 1
48 checkstat $MOUNT1/b || return 2
49 rm $MOUNT2/a $MOUNT1/b
50 checkstat $MOUNT1/a && return 3
51 checkstat $MOUNT2/b && return 4
55 run_test 1 "|X| simple create"
63 checkstat $MOUNT2/adir || return 1
65 checkstat $MOUNT2/adir && return 2
69 run_test 2 "|X| mkdir adir"
74 mkdir $MOUNT2/adir/bdir
77 checkstat $MOUNT2/adir || return 1
78 checkstat $MOUNT1/adir/bdir || return 2
79 rmdir $MOUNT2/adir/bdir $MOUNT1/adir
80 checkstat $MOUNT1/adir && return 3
81 checkstat $MOUNT2/adir/bdir && return 4
85 run_test 3 "|X| mkdir adir, mkdir adir/bdir "
90 mkdir $MOUNT1/adir && return 1
91 mkdir $MOUNT2/adir/bdir
94 checkstat $MOUNT2/adir || return 2
95 checkstat $MOUNT1/adir/bdir || return 3
97 rmdir $MOUNT2/adir/bdir $MOUNT1/adir
98 checkstat $MOUNT1/adir && return 4
99 checkstat $MOUNT2/adir/bdir && return 5
103 run_test 4 "|X| mkdir adir (-EEXIST), mkdir adir/bdir "
107 # multiclient version of replay_single.sh/test_8
109 multiop $MOUNT2/a o_tSc &
111 # give multiop a chance to open
116 wait $pid || return 1
119 [ -e $MOUNT2/a ] && return 2
122 run_test 5 "open, unlink |X| close"
127 multiop $MOUNT2/a o_c &
129 multiop $MOUNT1/a o_c &
131 # give multiop a chance to open
136 wait $pid1 || return 1
140 wait $pid2 || return 1
141 [ -e $MOUNT2/a ] && return 2
144 run_test 6 "open1, open2, unlink |X| close1 [fail mds] close2"
148 drop_reint_reply "mcreate $MOUNT1/$tfile" || return 1
150 checkstat $MOUNT2/$tfile || return 2
151 rm $MOUNT1/$tfile || return 3
155 run_test 8 "replay of resent request"
159 mcreate $MOUNT1/$tfile-1
160 mcreate $MOUNT2/$tfile-2
161 # drop first reint reply
162 sysctl -w lustre.fail_loc=0x80000119
164 sysctl -w lustre.fail_loc=0
166 rm $MOUNT1/$tfile-[1,2] || return 1
170 run_test 9 "resending a replayed create"
173 mcreate $MOUNT1/$tfile-1
175 munlink $MOUNT1/$tfile-1
176 mcreate $MOUNT2/$tfile-2
177 # drop first reint reply
178 sysctl -w lustre.fail_loc=0x80000119
180 sysctl -w lustre.fail_loc=0
182 checkstat $MOUNT1/$tfile-1 && return 1
183 checkstat $MOUNT1/$tfile-2 || return 2
188 run_test 10 "resending a replayed unlink"
192 mcreate $MOUNT1/$tfile-1
193 mcreate $MOUNT2/$tfile-2
194 mcreate $MOUNT1/$tfile-3
195 mcreate $MOUNT2/$tfile-4
196 mcreate $MOUNT1/$tfile-5
197 # drop all reint replies for a while
198 sysctl -w lustre.fail_loc=0x0119
200 #sleep for while, let both clients reconnect and timeout
201 sleep $((TIMEOUT * 2))
202 sysctl -w lustre.fail_loc=0
204 rm $MOUNT1/$tfile-[1-5] || return 1
208 run_test 11 "both clients timeout during replay"
213 multiop $DIR/$tfile mo_c &
217 #define OBD_FAIL_LDLM_ENQUEUE 0x302
218 sysctl -w lustre.fail_loc=0x80000302
220 df $MOUNT || return 1
221 sysctl -w lustre.fail_loc=0
224 kill -USR1 $MULTIPID || return 3
225 wait $MULTIPID || return 4
226 $CHECKSTAT -t file $DIR/$tfile || return 2
231 run_test 12 "open resend timeout"
234 multiop $DIR/$tfile mo_c &
240 kill -USR1 $MULTIPID || return 3
241 wait $MULTIPID || return 4
244 sysctl -w lustre.fail_loc=0x80000115
246 df $MOUNT || return 1
247 sysctl -w lustre.fail_loc=0
250 $CHECKSTAT -t file $DIR/$tfile || return 2
255 run_test 13 "close resend timeout"
259 createmany -o $MOUNT1/$tfile- 25
260 createmany -o $MOUNT2/$tfile-2- 1
261 createmany -o $MOUNT1/$tfile-3- 25
265 # expect failover to fail
266 df $MOUNT && return 1
269 # first 25 files should have been replayed
270 unlinkmany $MOUNT1/$tfile- 25 || return 2
272 zconf_mount `hostname` $MOUNT2
275 run_test 14 "timeouts waiting for lost client during replay"
279 createmany -o $MOUNT1/$tfile- 25
280 createmany -o $MOUNT2/$tfile-2- 1
284 df $MOUNT || return 1
286 unlinkmany $MOUNT1/$tfile- 25 || return 2
287 [ -e $MOUNT1/$tfile-2-0 ] && error "$tfile-2-0 exists"
289 zconf_mount `hostname` $MOUNT2
292 run_test 15 "timeout waiting for lost client during replay, 1 client completes"
299 echo "data" > "$MOUNT2/${tfile}-m2"
303 df $MOUNT || return 1
305 ost_last_id=`cat /proc/fs/lustre/obdfilter/*/last_id`
306 mds_last_id=`cat /proc/fs/lustre/osc/*mds*/last_id`
308 echo "Ids after MDS<->OST synchonizing"
309 echo "--------------------------------"
316 echo $ost_last_id | while read id; do
322 echo $mds_last_id | while read id; do
327 local arr_len=${#mds_ids[*]}
328 for ((i=0;i<$arr_len;i++)); do
332 test $mds_id -ge $ost_id || {
333 echo "MDS last id ($mds_id) is smaller than OST one ($ost_id)"
338 zconf_mount `hostname` $MOUNT2
341 #CROW run_test 15a "OST clear orphans - synchronize ids on MDS and OST"
345 echo "data" > "$MOUNT2/${tfile}-m2"
348 do_facet ost1 "sysctl -w lustre.fail_loc=0x80000802"
351 df $MOUNT || return 1
352 do_facet ost1 "sysctl -w lustre.fail_loc=0"
354 zconf_mount `hostname` $MOUNT2
357 #CROW run_test 15b "multiple delayed OST clear orphans"
361 for ((i = 0; i < 2000; i++)); do
362 echo "data" > "$MOUNT2/${tfile}-$i" || error "create ${tfile}-$i failed"
368 df $MOUNT || return 1
370 zconf_mount `hostname` $MOUNT2
373 run_test 15c "remove multiple OST orphans"
377 createmany -o $MOUNT1/$tfile- 25
378 createmany -o $MOUNT2/$tfile-2- 1
384 df $MOUNT || return 1
386 unlinkmany $MOUNT1/$tfile- 25 || return 2
388 zconf_mount `hostname` $MOUNT2
392 run_test 16 "fail MDS during recovery (3571)"
395 createmany -o $MOUNT1/$tfile- 25
396 createmany -o $MOUNT2/$tfile-2- 1
398 # Make sure the disconnect is lost
405 df $MOUNT || return 1
407 unlinkmany $MOUNT1/$tfile- 25 || return 2
409 zconf_mount `hostname` $MOUNT2
413 run_test 17 "fail OST during recovery (3571)"
415 # cleanup with blocked enqueue fails until timer elapses (MDS busy), wait for it
418 test_18() { # bug 3822 - evicting client with enqueued lock
420 mkdir -p $MOUNT1/$tdir
421 touch $MOUNT1/$tdir/f0
422 #define OBD_FAIL_LDLM_ENQUEUE_BLOCKED 0x30b
423 statmany -s $MOUNT1/$tdir/f 1 500 &
426 do_facet mds sysctl -w lustre.fail_loc=0x8000030b # hold enqueue
428 #define OBD_FAIL_LDLM_BL_CALLBACK 0x305
429 do_facet client sysctl -w lustre.fail_loc=0x80000305 # drop cb, evict
431 usleep 500 # wait to ensure first client is one that will be evicted
432 openfile -f O_RDONLY $MOUNT2/$tdir/f0
434 dmesg | grep "entering recovery in server" && \
435 error "client not evicted" || true
437 run_test 18 "ldlm_handle_enqueue succeeds on evicted export (3822)"
439 test_19() { # Bug 10991 - resend of open request does not fail assertion.
441 drop_ldlm_reply "createmany -o $DIR/$tfile 1" || return 1
443 checkstat $DIR2/${tfile}0 || return 2
444 rm $DIR/${tfile}0 || return 3
448 run_test 19 "resend of open request"
450 if [ "$ONLY" != "setup" ]; then
451 equals_msg test complete, cleaning up
452 SLEEP=$((`date +%s` - $NOW))
453 [ $SLEEP -lt $TIMEOUT ] && sleep $SLEEP