5 # bug number: 6088 10124 10800
6 ALWAYS_EXCEPT="8 15c 17 $REPLAY_DUAL_EXCEPT"
9 [ "$SLOW" = "no" ] && EXCEPT_SLOW="1 2 3 4 5 14"
12 PTLDEBUG=${PTLDEBUG:--1}
13 LUSTRE=${LUSTRE:-`dirname $0`/..}
15 CLEANUP=${CLEANUP:-""}
16 MOUNT_2=${MOUNT_2:-"yes"}
17 . $LUSTRE/tests/test-framework.sh
21 . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
26 cleanup_and_setup_lustre
27 rm -rf $DIR/[df][0-9]*
29 [ "$DAEMONFILE" ] && $LCTL debug_daemon start $DAEMONFILE $DAEMONSIZE
33 replay_barrier $SINGLEMDS
37 checkstat $MOUNT2/a || return 1
38 checkstat $MOUNT1/b || return 2
39 rm $MOUNT2/a $MOUNT1/b
40 checkstat $MOUNT1/a && return 3
41 checkstat $MOUNT2/b && return 4
45 run_test 1 "|X| simple create"
49 replay_barrier $SINGLEMDS
53 checkstat $MOUNT2/adir || return 1
55 checkstat $MOUNT2/adir && return 2
58 run_test 2 "|X| mkdir adir"
61 replay_barrier $SINGLEMDS
63 mkdir $MOUNT2/adir/bdir
66 checkstat $MOUNT2/adir || return 1
67 checkstat $MOUNT1/adir/bdir || return 2
68 rmdir $MOUNT2/adir/bdir $MOUNT1/adir
69 checkstat $MOUNT1/adir && return 3
70 checkstat $MOUNT2/adir/bdir && return 4
73 run_test 3 "|X| mkdir adir, mkdir adir/bdir "
77 replay_barrier $SINGLEMDS
78 mkdir $MOUNT1/adir && return 1
79 mkdir $MOUNT2/adir/bdir
82 checkstat $MOUNT2/adir || return 2
83 checkstat $MOUNT1/adir/bdir || return 3
85 rmdir $MOUNT2/adir/bdir $MOUNT1/adir
86 checkstat $MOUNT1/adir && return 4
87 checkstat $MOUNT2/adir/bdir && return 5
90 run_test 4 "|X| mkdir adir (-EEXIST), mkdir adir/bdir "
94 # multiclient version of replay_single.sh/test_8
96 multiop $MOUNT2/a o_tSc &
98 # give multiop a chance to open
101 replay_barrier $SINGLEMDS
103 wait $pid || return 1
106 [ -e $MOUNT2/a ] && return 2
109 run_test 5 "open, unlink |X| close"
114 multiop $MOUNT2/a o_c &
116 multiop $MOUNT1/a o_c &
118 # give multiop a chance to open
121 replay_barrier $SINGLEMDS
123 wait $pid1 || return 1
127 wait $pid2 || return 1
128 [ -e $MOUNT2/a ] && return 2
131 run_test 6 "open1, open2, unlink |X| close1 [fail $SINGLEMDS] close2"
134 replay_barrier $SINGLEMDS
135 drop_reint_reply "mcreate $MOUNT1/$tfile" || return 1
137 checkstat $MOUNT2/$tfile || return 2
138 rm $MOUNT1/$tfile || return 3
142 run_test 8 "replay of resent request"
145 replay_barrier $SINGLEMDS
146 mcreate $MOUNT1/$tfile-1
147 mcreate $MOUNT2/$tfile-2
148 # drop first reint reply
149 do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0x80000119
151 do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0
153 rm $MOUNT1/$tfile-[1,2] || return 1
157 run_test 9 "resending a replayed create"
160 mcreate $MOUNT1/$tfile-1
161 replay_barrier $SINGLEMDS
162 munlink $MOUNT1/$tfile-1
163 mcreate $MOUNT2/$tfile-2
164 # drop first reint reply
165 do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0x80000119
167 do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0
169 checkstat $MOUNT1/$tfile-1 && return 1
170 checkstat $MOUNT1/$tfile-2 || return 2
175 run_test 10 "resending a replayed unlink"
178 replay_barrier $SINGLEMDS
179 mcreate $MOUNT1/$tfile-1
180 mcreate $MOUNT2/$tfile-2
181 mcreate $MOUNT1/$tfile-3
182 mcreate $MOUNT2/$tfile-4
183 mcreate $MOUNT1/$tfile-5
184 # drop all reint replies for a while
185 do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0x0119
186 # note that with this fail_loc set, facet_failover df will fail
187 facet_failover $SINGLEMDS
188 #sleep for while, let both clients reconnect and timeout
189 sleep $((TIMEOUT * 2))
190 do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0
192 rm $MOUNT1/$tfile-[1-5] || return 1
196 run_test 11 "both clients timeout during replay"
199 replay_barrier $SINGLEMDS
201 multiop $DIR/$tfile mo_c &
205 #define OBD_FAIL_LDLM_ENQUEUE 0x302
206 do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0x80000302
207 facet_failover $SINGLEMDS
208 do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0
209 df $MOUNT || return 1
212 kill -USR1 $MULTIPID || return 3
213 wait $MULTIPID || return 4
214 $CHECKSTAT -t file $DIR/$tfile || return 2
219 run_test 12 "open resend timeout"
222 multiop $DIR/$tfile mo_c &
226 replay_barrier $SINGLEMDS
228 kill -USR1 $MULTIPID || return 3
229 wait $MULTIPID || return 4
232 do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0x80000115
233 facet_failover $SINGLEMDS
234 do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0
235 df $MOUNT || return 1
238 $CHECKSTAT -t file $DIR/$tfile || return 2
243 run_test 13 "close resend timeout"
246 replay_barrier $SINGLEMDS
247 createmany -o $MOUNT1/$tfile- 25
248 createmany -o $MOUNT2/$tfile-2- 1
249 createmany -o $MOUNT1/$tfile-3- 25
252 facet_failover $SINGLEMDS
253 # expect failover to fail due to missing client 2
254 df $MOUNT && return 1
257 # first 25 files should have been replayed
258 unlinkmany $MOUNT1/$tfile- 25 || return 2
260 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
263 run_test 14 "timeouts waiting for lost client during replay"
266 replay_barrier $SINGLEMDS
267 createmany -o $MOUNT1/$tfile- 25
268 createmany -o $MOUNT2/$tfile-2- 1
271 facet_failover $SINGLEMDS
272 df $MOUNT || return 1
274 unlinkmany $MOUNT1/$tfile- 25 || return 2
275 [ -e $MOUNT1/$tfile-2-0 ] && error "$tfile-2-0 exists"
277 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
280 run_test 15 "timeout waiting for lost client during replay, 1 client completes"
286 replay_barrier $SINGLEMDS
287 echo "data" > "$MOUNT2/${tfile}-m2"
290 facet_failover $SINGLEMDS
291 df $MOUNT || return 1
293 ost_last_id=`cat /proc/fs/lustre/obdfilter/*/last_id`
294 mds_last_id=`cat /proc/fs/lustre/osc/*mds*/last_id`
296 echo "Ids after MDS<->OST synchonizing"
297 echo "--------------------------------"
304 echo $ost_last_id | while read id; do
310 echo $mds_last_id | while read id; do
315 local arr_len=${#mds_ids[*]}
316 for ((i=0;i<$arr_len;i++)); do
320 test $mds_id -ge $ost_id || {
321 echo "MDS last id ($mds_id) is smaller than OST one ($ost_id)"
326 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
329 #CROW run_test 15a "OST clear orphans - synchronize ids on MDS and OST"
332 replay_barrier $SINGLEMDS
333 echo "data" > "$MOUNT2/${tfile}-m2"
336 do_facet ost1 "sysctl -w lustre.fail_loc=0x80000802"
337 facet_failover $SINGLEMDS
339 df $MOUNT || return 1
340 do_facet ost1 "sysctl -w lustre.fail_loc=0"
342 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
345 #CROW run_test 15b "multiple delayed OST clear orphans"
348 replay_barrier $SINGLEMDS
349 for ((i = 0; i < 2000; i++)); do
350 echo "data" > "$MOUNT2/${tfile}-$i" || error "create ${tfile}-$i failed"
354 facet_failover $SINGLEMDS
356 df $MOUNT || return 1
358 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
361 run_test 15c "remove multiple OST orphans"
364 replay_barrier $SINGLEMDS
365 createmany -o $MOUNT1/$tfile- 25
366 createmany -o $MOUNT2/$tfile-2- 1
369 facet_failover $SINGLEMDS
371 facet_failover $SINGLEMDS
372 df $MOUNT || return 1
374 unlinkmany $MOUNT1/$tfile- 25 || return 2
376 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
380 run_test 16 "fail MDS during recovery (3571)"
383 createmany -o $MOUNT1/$tfile- 25
384 createmany -o $MOUNT2/$tfile-2- 1
386 # Make sure the disconnect is lost
393 df $MOUNT || return 1
395 unlinkmany $MOUNT1/$tfile- 25 || return 2
397 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
401 run_test 17 "fail OST during recovery (3571)"
403 # cleanup with blocked enqueue fails until timer elapses (MDS busy), wait for it
406 test_18() { # bug 3822 - evicting client with enqueued lock
408 mkdir -p $MOUNT1/$tdir
409 touch $MOUNT1/$tdir/f0
410 #define OBD_FAIL_LDLM_ENQUEUE_BLOCKED 0x30b
411 statmany -s $MOUNT1/$tdir/f 1 500 &
414 do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0x8000030b # hold enqueue
416 #define OBD_FAIL_LDLM_BL_CALLBACK 0x305
417 do_facet client sysctl -w lustre.fail_loc=0x80000305 # drop cb, evict
419 usleep 500 # wait to ensure first client is one that will be evicted
420 openfile -f O_RDONLY $MOUNT2/$tdir/f0
422 dmesg | grep "entering recovery in server" && \
423 error "client not evicted" || true
424 do_facet client "sysctl -w lustre.fail_loc=0"
425 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
427 run_test 18 "ldlm_handle_enqueue succeeds on evicted export (3822)"
429 test_19() { # Bug 10991 - resend of open request does not fail assertion.
430 replay_barrier $SINGLEMDS
431 drop_ldlm_reply "createmany -o $DIR/$tfile 1" || return 1
433 checkstat $DIR2/${tfile}0 || return 2
434 rm $DIR/${tfile}0 || return 3
438 run_test 19 "resend of open request"
440 equals_msg `basename $0`: test complete, cleaning up
441 SLEEP=$((`date +%s` - $NOW))
442 [ $SLEEP -lt $TIMEOUT ] && sleep $SLEEP
443 check_and_cleanup_lustre
444 [ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG || true