5 # bug number: 13129 13129 6088 10124
6 ALWAYS_EXCEPT="2 3 8 15c $REPLAY_DUAL_EXCEPT"
9 PTLDEBUG=${PTLDEBUG:--1}
10 LUSTRE=${LUSTRE:-`dirname $0`/..}
12 CLEANUP=${CLEANUP:-""}
13 MOUNT_2=${MOUNT_2:-"yes"}
14 . $LUSTRE/tests/test-framework.sh
18 . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
21 [ "$SLOW" = "no" ] && EXCEPT_SLOW="1 2 3 4 5 14"
25 cleanup_and_setup_lustre
26 rm -rf $DIR/[df][0-9]*
28 [ "$DAEMONFILE" ] && $LCTL debug_daemon start $DAEMONFILE $DAEMONSIZE
32 replay_barrier $SINGLEMDS
36 checkstat $MOUNT2/a || return 1
37 checkstat $MOUNT1/b || return 2
38 rm $MOUNT2/a $MOUNT1/b
39 checkstat $MOUNT1/a && return 3
40 checkstat $MOUNT2/b && return 4
44 run_test 1 "|X| simple create"
48 replay_barrier $SINGLEMDS
52 checkstat $MOUNT2/adir || return 1
54 checkstat $MOUNT2/adir && return 2
57 run_test 2 "|X| mkdir adir"
60 replay_barrier $SINGLEMDS
62 mkdir $MOUNT2/adir/bdir
65 checkstat $MOUNT2/adir || return 1
66 checkstat $MOUNT1/adir/bdir || return 2
67 rmdir $MOUNT2/adir/bdir $MOUNT1/adir
68 checkstat $MOUNT1/adir && return 3
69 checkstat $MOUNT2/adir/bdir && return 4
72 run_test 3 "|X| mkdir adir, mkdir adir/bdir "
76 replay_barrier $SINGLEMDS
77 mkdir $MOUNT1/adir && return 1
78 mkdir $MOUNT2/adir/bdir
81 checkstat $MOUNT2/adir || return 2
82 checkstat $MOUNT1/adir/bdir || return 3
84 rmdir $MOUNT2/adir/bdir $MOUNT1/adir
85 checkstat $MOUNT1/adir && return 4
86 checkstat $MOUNT2/adir/bdir && return 5
89 run_test 4 "|X| mkdir adir (-EEXIST), mkdir adir/bdir "
93 # multiclient version of replay_single.sh/test_8
95 multiop_bg_pause $MOUNT2/a o_tSc || return 1
98 replay_barrier $SINGLEMDS
100 wait $pid || return 1
103 [ -e $MOUNT2/a ] && return 2
106 run_test 5 "open, unlink |X| close"
111 multiop_bg_pause $MOUNT2/a o_c || return 1
113 multiop_bg_pause $MOUNT1/a o_c || return 1
116 replay_barrier $SINGLEMDS
118 wait $pid1 || return 1
122 wait $pid2 || return 1
123 [ -e $MOUNT2/a ] && return 2
126 run_test 6 "open1, open2, unlink |X| close1 [fail $SINGLEMDS] close2"
129 replay_barrier $SINGLEMDS
130 drop_reint_reply "mcreate $MOUNT1/$tfile" || return 1
132 checkstat $MOUNT2/$tfile || return 2
133 rm $MOUNT1/$tfile || return 3
137 run_test 8 "replay of resent request"
140 replay_barrier $SINGLEMDS
141 mcreate $MOUNT1/$tfile-1
142 mcreate $MOUNT2/$tfile-2
143 # drop first reint reply
144 do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0x80000119
146 do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0
148 rm $MOUNT1/$tfile-[1,2] || return 1
152 run_test 9 "resending a replayed create"
155 mcreate $MOUNT1/$tfile-1
156 replay_barrier $SINGLEMDS
157 munlink $MOUNT1/$tfile-1
158 mcreate $MOUNT2/$tfile-2
159 # drop first reint reply
160 do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0x80000119
162 do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0
164 checkstat $MOUNT1/$tfile-1 && return 1
165 checkstat $MOUNT1/$tfile-2 || return 2
170 run_test 10 "resending a replayed unlink"
173 replay_barrier $SINGLEMDS
174 mcreate $MOUNT1/$tfile-1
175 mcreate $MOUNT2/$tfile-2
176 mcreate $MOUNT1/$tfile-3
177 mcreate $MOUNT2/$tfile-4
178 mcreate $MOUNT1/$tfile-5
179 # drop all reint replies for a while
180 do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0x0119
181 # note that with this fail_loc set, facet_failover df will fail
182 facet_failover $SINGLEMDS
183 #sleep for while, let both clients reconnect and timeout
184 sleep $((TIMEOUT * 2))
185 do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0
187 rm $MOUNT1/$tfile-[1-5] || return 1
191 run_test 11 "both clients timeout during replay"
194 replay_barrier $SINGLEMDS
196 multiop_bg_pause $DIR/$tfile mo_c || return 1
199 #define OBD_FAIL_LDLM_ENQUEUE 0x302
200 do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0x80000302
201 facet_failover $SINGLEMDS
202 do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0
203 df $MOUNT || return 1
206 kill -USR1 $MULTIPID || return 3
207 wait $MULTIPID || return 4
208 $CHECKSTAT -t file $DIR/$tfile || return 2
213 run_test 12 "open resend timeout"
216 multiop_bg_pause $DIR/$tfile mo_c || return 1
219 replay_barrier $SINGLEMDS
221 kill -USR1 $MULTIPID || return 3
222 wait $MULTIPID || return 4
225 do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0x80000115
226 facet_failover $SINGLEMDS
227 do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0
228 df $MOUNT || return 1
231 $CHECKSTAT -t file $DIR/$tfile || return 2
236 run_test 13 "close resend timeout"
239 replay_barrier $SINGLEMDS
240 createmany -o $MOUNT1/$tfile- 25
241 createmany -o $MOUNT2/$tfile-2- 1
242 createmany -o $MOUNT1/$tfile-3- 25
245 facet_failover $SINGLEMDS
246 # expect failover to fail due to missing client 2
247 df $MOUNT && return 1
250 # first 25 files should have been replayed
251 unlinkmany $MOUNT1/$tfile- 25 || return 2
253 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
256 run_test 14 "timeouts waiting for lost client during replay"
259 replay_barrier $SINGLEMDS
260 createmany -o $MOUNT1/$tfile- 25
261 createmany -o $MOUNT2/$tfile-2- 1
264 facet_failover $SINGLEMDS
265 df $MOUNT || return 1
267 unlinkmany $MOUNT1/$tfile- 25 || return 2
268 [ -e $MOUNT1/$tfile-2-0 ] && error "$tfile-2-0 exists"
270 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
273 run_test 15 "timeout waiting for lost client during replay, 1 client completes"
279 replay_barrier $SINGLEMDS
280 echo "data" > "$MOUNT2/${tfile}-m2"
283 facet_failover $SINGLEMDS
284 df $MOUNT || return 1
286 ost_last_id=`cat /proc/fs/lustre/obdfilter/*/last_id`
287 mds_last_id=`cat /proc/fs/lustre/osc/*mds*/last_id`
289 echo "Ids after MDS<->OST synchonizing"
290 echo "--------------------------------"
297 echo $ost_last_id | while read id; do
303 echo $mds_last_id | while read id; do
308 local arr_len=${#mds_ids[*]}
309 for ((i=0;i<$arr_len;i++)); do
313 test $mds_id -ge $ost_id || {
314 echo "MDS last id ($mds_id) is smaller than OST one ($ost_id)"
319 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
322 #CROW run_test 15a "OST clear orphans - synchronize ids on MDS and OST"
325 replay_barrier $SINGLEMDS
326 echo "data" > "$MOUNT2/${tfile}-m2"
329 do_facet ost1 "sysctl -w lustre.fail_loc=0x80000802"
330 facet_failover $SINGLEMDS
332 df $MOUNT || return 1
333 do_facet ost1 "sysctl -w lustre.fail_loc=0"
335 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
338 #CROW run_test 15b "multiple delayed OST clear orphans"
341 replay_barrier $SINGLEMDS
342 for ((i = 0; i < 2000; i++)); do
343 echo "data" > "$MOUNT2/${tfile}-$i" || error "create ${tfile}-$i failed"
347 facet_failover $SINGLEMDS
349 df $MOUNT || return 1
351 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
354 run_test 15c "remove multiple OST orphans"
357 replay_barrier $SINGLEMDS
358 createmany -o $MOUNT1/$tfile- 25
359 createmany -o $MOUNT2/$tfile-2- 1
362 facet_failover $SINGLEMDS
364 facet_failover $SINGLEMDS
365 df $MOUNT || return 1
367 unlinkmany $MOUNT1/$tfile- 25 || return 2
369 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
373 run_test 16 "fail MDS during recovery (3571)"
376 createmany -o $MOUNT1/$tfile- 25
377 createmany -o $MOUNT2/$tfile-2- 1
379 # Make sure the disconnect is lost
386 df $MOUNT || return 1
388 unlinkmany $MOUNT1/$tfile- 25 || return 2
390 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
394 run_test 17 "fail OST during recovery (3571)"
396 # cleanup with blocked enqueue fails until timer elapses (MDS busy), wait for it
399 test_18() { # bug 3822 - evicting client with enqueued lock
401 mkdir -p $MOUNT1/$tdir
402 touch $MOUNT1/$tdir/f0
403 #define OBD_FAIL_LDLM_ENQUEUE_BLOCKED 0x30b
404 statmany -s $MOUNT1/$tdir/f 1 500 &
407 do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0x8000030b # hold enqueue
409 #define OBD_FAIL_LDLM_BL_CALLBACK 0x305
410 do_facet client sysctl -w lustre.fail_loc=0x80000305 # drop cb, evict
412 usleep 500 # wait to ensure first client is one that will be evicted
413 openfile -f O_RDONLY $MOUNT2/$tdir/f0
415 dmesg | grep "entering recovery in server" && \
416 error "client not evicted" || true
417 do_facet client "sysctl -w lustre.fail_loc=0"
418 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
420 run_test 18 "ldlm_handle_enqueue succeeds on evicted export (3822)"
422 test_19() { # Bug 10991 - resend of open request does not fail assertion.
423 replay_barrier $SINGLEMDS
424 drop_ldlm_reply "createmany -o $DIR/$tfile 1" || return 1
426 checkstat $DIR2/${tfile}0 || return 2
427 rm $DIR/${tfile}0 || return 3
431 run_test 19 "resend of open request"
433 equals_msg `basename $0`: test complete, cleaning up
434 SLEEP=$((`date +%s` - $NOW))
435 [ $SLEEP -lt $TIMEOUT ] && sleep $SLEEP
436 check_and_cleanup_lustre
437 [ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG || true