6 ALWAYS_EXCEPT="15c $REPLAY_DUAL_EXCEPT"
8 PTLDEBUG=${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
33 [ "$REFORMAT" ] && formatall
39 [ "$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
68 run_test 2 "|X| mkdir adir"
73 mkdir $MOUNT2/adir/bdir
76 checkstat $MOUNT2/adir || return 1
77 checkstat $MOUNT1/adir/bdir || return 2
78 rmdir $MOUNT2/adir/bdir $MOUNT1/adir
79 checkstat $MOUNT1/adir && return 3
80 checkstat $MOUNT2/adir/bdir && return 4
83 run_test 3 "|X| mkdir adir, mkdir adir/bdir "
88 mkdir $MOUNT1/adir && return 1
89 mkdir $MOUNT2/adir/bdir
92 checkstat $MOUNT2/adir || return 2
93 checkstat $MOUNT1/adir/bdir || return 3
95 rmdir $MOUNT2/adir/bdir $MOUNT1/adir
96 checkstat $MOUNT1/adir && return 4
97 checkstat $MOUNT2/adir/bdir && return 5
100 run_test 4 "|X| mkdir adir (-EEXIST), mkdir adir/bdir "
104 # multiclient version of replay_single.sh/test_8
106 multiop $MOUNT2/a o_tSc &
108 # give multiop a chance to open
113 wait $pid || return 1
116 [ -e $MOUNT2/a ] && return 2
119 run_test 5 "open, unlink |X| close"
124 multiop $MOUNT2/a o_c &
126 multiop $MOUNT1/a o_c &
128 # give multiop a chance to open
133 wait $pid1 || return 1
137 wait $pid2 || return 1
138 [ -e $MOUNT2/a ] && return 2
141 run_test 6 "open1, open2, unlink |X| close1 [fail mds] close2"
145 drop_reint_reply "mcreate $MOUNT1/$tfile" || return 1
147 checkstat $MOUNT2/$tfile || return 2
148 rm $MOUNT1/$tfile || return 3
152 run_test 8 "replay of resent request"
156 mcreate $MOUNT1/$tfile-1
157 mcreate $MOUNT2/$tfile-2
158 # drop first reint reply
159 do_facet mds sysctl -w lustre.fail_loc=0x80000119
161 do_facet mds sysctl -w lustre.fail_loc=0
163 rm $MOUNT1/$tfile-[1,2] || return 1
167 run_test 9 "resending a replayed create"
170 mcreate $MOUNT1/$tfile-1
172 munlink $MOUNT1/$tfile-1
173 mcreate $MOUNT2/$tfile-2
174 # drop first reint reply
175 do_facet mds sysctl -w lustre.fail_loc=0x80000119
177 do_facet mds sysctl -w lustre.fail_loc=0
179 checkstat $MOUNT1/$tfile-1 && return 1
180 checkstat $MOUNT1/$tfile-2 || return 2
185 run_test 10 "resending a replayed unlink"
189 mcreate $MOUNT1/$tfile-1
190 mcreate $MOUNT2/$tfile-2
191 mcreate $MOUNT1/$tfile-3
192 mcreate $MOUNT2/$tfile-4
193 mcreate $MOUNT1/$tfile-5
194 # drop all reint replies for a while
195 do_facet mds sysctl -w lustre.fail_loc=0x0119
197 #sleep for while, let both clients reconnect and timeout
198 sleep $((TIMEOUT * 2))
199 do_facet mds sysctl -w lustre.fail_loc=0
200 while [ -z "$(ls $MOUNT1/$tfile-[1-5] 2>/dev/null)" ]; do
204 ls $MOUNT1/$tfile-[1-5]
205 rm $MOUNT1/$tfile-[1-5] || return 1
209 run_test 11 "both clients timeout during replay"
214 multiop $DIR/$tfile mo_c &
218 #define OBD_FAIL_LDLM_ENQUEUE 0x302
219 do_facet mds sysctl -w lustre.fail_loc=0x80000302
221 df $MOUNT || return 1
222 do_facet mds sysctl -w lustre.fail_loc=0
225 kill -USR1 $MULTIPID || return 3
226 wait $MULTIPID || return 4
227 $CHECKSTAT -t file $DIR/$tfile || return 2
232 run_test 12 "open resend timeout"
235 multiop $DIR/$tfile mo_c &
241 kill -USR1 $MULTIPID || return 3
242 wait $MULTIPID || return 4
245 do_facet mds sysctl -w lustre.fail_loc=0x80000115
247 df $MOUNT || return 1
248 do_facet mds sysctl -w lustre.fail_loc=0
251 $CHECKSTAT -t file $DIR/$tfile || return 2
256 run_test 13 "close resend timeout"
260 createmany -o $MOUNT1/$tfile- 25
261 createmany -o $MOUNT2/$tfile-2- 1
262 createmany -o $MOUNT1/$tfile-3- 25
266 # expect failover to fail
267 df $MOUNT && return 1
270 # first 25 files should have been replayed
271 unlinkmany $MOUNT1/$tfile- 25 || return 2
273 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
276 run_test 14 "timeouts waiting for lost client during replay"
280 createmany -o $MOUNT1/$tfile- 25
281 createmany -o $MOUNT2/$tfile-2- 1
285 df $MOUNT || return 1
287 unlinkmany $MOUNT1/$tfile- 25 || return 2
288 [ -e $MOUNT1/$tfile-2-0 ] && error "$tfile-2-0 exists"
290 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
293 run_test 15 "timeout waiting for lost client during replay, 1 client completes"
300 echo "data" > "$MOUNT2/${tfile}-m2"
304 df $MOUNT || return 1
306 ost_last_id=`cat /proc/fs/lustre/obdfilter/*/last_id`
307 mds_last_id=`cat /proc/fs/lustre/osc/*mds*/last_id`
309 echo "Ids after MDS<->OST synchonizing"
310 echo "--------------------------------"
317 echo $ost_last_id | while read id; do
323 echo $mds_last_id | while read id; do
328 local arr_len=${#mds_ids[*]}
329 for ((i=0;i<$arr_len;i++)); do
333 test $mds_id -ge $ost_id || {
334 echo "MDS last id ($mds_id) is smaller than OST one ($ost_id)"
339 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
342 #CROW run_test 15a "OST clear orphans - synchronize ids on MDS and OST"
346 echo "data" > "$MOUNT2/${tfile}-m2"
349 do_facet ost1 "sysctl -w lustre.fail_loc=0x80000802"
352 df $MOUNT || return 1
353 do_facet ost1 "sysctl -w lustre.fail_loc=0"
355 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
358 #CROW run_test 15b "multiple delayed OST clear orphans"
362 for ((i = 0; i < 2000; i++)); do
363 echo "data" > "$MOUNT2/${tfile}-$i" || error "create ${tfile}-$i failed"
369 df $MOUNT || return 1
371 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
374 run_test 15c "remove multiple OST orphans"
378 createmany -o $MOUNT1/$tfile- 25
379 createmany -o $MOUNT2/$tfile-2- 1
385 df $MOUNT || return 1
387 unlinkmany $MOUNT1/$tfile- 25 || return 2
389 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
393 run_test 16 "fail MDS during recovery (3571)"
396 createmany -o $MOUNT1/$tfile- 25
397 createmany -o $MOUNT2/$tfile-2- 1
399 # Make sure the disconnect is lost
406 df $MOUNT || return 1
408 unlinkmany $MOUNT1/$tfile- 25 || return 2
410 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
414 run_test 17 "fail OST during recovery (3571)"
416 # cleanup with blocked enqueue fails until timer elapses (MDS busy), wait for it
419 test_18() { # bug 3822 - evicting client with enqueued lock
421 mkdir -p $MOUNT1/$tdir
422 touch $MOUNT1/$tdir/f0
423 #define OBD_FAIL_LDLM_ENQUEUE_BLOCKED 0x30b
424 statmany -s $MOUNT1/$tdir/f 1 500 &
427 do_facet mds sysctl -w lustre.fail_loc=0x8000030b # hold enqueue
429 #define OBD_FAIL_LDLM_BL_CALLBACK 0x305
430 do_facet client sysctl -w lustre.fail_loc=0x80000305 # drop cb, evict
432 sleep 0.500s # wait to ensure first client is one that will be evicted
433 openfile -f O_RDONLY $MOUNT2/$tdir/f0
435 dmesg | grep "entering recovery in server" && \
436 error "client not evicted" || true
438 run_test 18 "ldlm_handle_enqueue succeeds on evicted export (3822)"
440 test_19() { # Bug 10991 - resend of open request does not fail assertion.
442 drop_ldlm_reply "createmany -o $DIR/$tfile 1" || return 1
444 checkstat $DIR2/${tfile}0 || return 2
445 rm $DIR/${tfile}0 || return 3
449 run_test 19 "resend of open request"
451 if [ "$ONLY" != "setup" ]; then
452 equals_msg test complete, cleaning up
453 SLEEP=$((`date +%s` - $NOW))
454 [ $SLEEP -lt $TIMEOUT ] && sleep $SLEEP