5 # bug number: 6088 10124 10800
6 ALWAYS_EXCEPT="8 15c 17 $REPLAY_DUAL_EXCEPT"
9 PTLDEBUG=${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/$NAME.sh}
23 cleanup_and_setup_lustre
25 [ "$DAEMONFILE" ] && $LCTL debug_daemon start $DAEMONFILE $DAEMONSIZE
29 replay_barrier $SINGLEMDS
33 checkstat $MOUNT2/a || return 1
34 checkstat $MOUNT1/b || return 2
35 rm $MOUNT2/a $MOUNT1/b
36 checkstat $MOUNT1/a && return 3
37 checkstat $MOUNT2/b && return 4
41 run_test 1 "|X| simple create"
45 replay_barrier $SINGLEMDS
49 checkstat $MOUNT2/adir || return 1
51 checkstat $MOUNT2/adir && return 2
54 run_test 2 "|X| mkdir adir"
57 replay_barrier $SINGLEMDS
59 mkdir $MOUNT2/adir/bdir
62 checkstat $MOUNT2/adir || return 1
63 checkstat $MOUNT1/adir/bdir || return 2
64 rmdir $MOUNT2/adir/bdir $MOUNT1/adir
65 checkstat $MOUNT1/adir && return 3
66 checkstat $MOUNT2/adir/bdir && return 4
69 run_test 3 "|X| mkdir adir, mkdir adir/bdir "
73 replay_barrier $SINGLEMDS
74 mkdir $MOUNT1/adir && return 1
75 mkdir $MOUNT2/adir/bdir
78 checkstat $MOUNT2/adir || return 2
79 checkstat $MOUNT1/adir/bdir || return 3
81 rmdir $MOUNT2/adir/bdir $MOUNT1/adir
82 checkstat $MOUNT1/adir && return 4
83 checkstat $MOUNT2/adir/bdir && return 5
86 run_test 4 "|X| mkdir adir (-EEXIST), mkdir adir/bdir "
90 # multiclient version of replay_single.sh/test_8
92 multiop $MOUNT2/a o_tSc &
94 # give multiop a chance to open
97 replay_barrier $SINGLEMDS
102 [ -e $MOUNT2/a ] && return 2
105 run_test 5 "open, unlink |X| close"
110 multiop $MOUNT2/a o_c &
112 multiop $MOUNT1/a o_c &
114 # give multiop a chance to open
117 replay_barrier $SINGLEMDS
119 wait $pid1 || return 1
123 wait $pid2 || return 1
124 [ -e $MOUNT2/a ] && return 2
127 run_test 6 "open1, open2, unlink |X| close1 [fail $SINGLEMDS] close2"
130 replay_barrier $SINGLEMDS
131 drop_reint_reply "mcreate $MOUNT1/$tfile" || return 1
133 checkstat $MOUNT2/$tfile || return 2
134 rm $MOUNT1/$tfile || return 3
138 run_test 8 "replay of resent request"
141 replay_barrier $SINGLEMDS
142 mcreate $MOUNT1/$tfile-1
143 mcreate $MOUNT2/$tfile-2
144 # drop first reint reply
145 do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0x80000119
147 do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0
149 rm $MOUNT1/$tfile-[1,2] || return 1
153 run_test 9 "resending a replayed create"
156 mcreate $MOUNT1/$tfile-1
157 replay_barrier $SINGLEMDS
158 munlink $MOUNT1/$tfile-1
159 mcreate $MOUNT2/$tfile-2
160 # drop first reint reply
161 do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0x80000119
163 do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0
165 checkstat $MOUNT1/$tfile-1 && return 1
166 checkstat $MOUNT1/$tfile-2 || return 2
171 run_test 10 "resending a replayed unlink"
174 replay_barrier $SINGLEMDS
175 mcreate $MOUNT1/$tfile-1
176 mcreate $MOUNT2/$tfile-2
177 mcreate $MOUNT1/$tfile-3
178 mcreate $MOUNT2/$tfile-4
179 mcreate $MOUNT1/$tfile-5
180 # drop all reint replies for a while
181 do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0x0119
182 # note that with this fail_loc set, facet_failover df will fail
183 facet_failover $SINGLEMDS
184 #sleep for while, let both clients reconnect and timeout
185 sleep $((TIMEOUT * 2))
186 do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0
188 rm $MOUNT1/$tfile-[1-5] || return 1
192 run_test 11 "both clients timeout during replay"
195 replay_barrier $SINGLEMDS
197 multiop $DIR/$tfile mo_c &
201 #define OBD_FAIL_LDLM_ENQUEUE 0x302
202 do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0x80000302
203 facet_failover $SINGLEMDS
204 do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0
205 df $MOUNT || return 1
208 kill -USR1 $MULTIPID || return 3
209 wait $MULTIPID || return 4
210 $CHECKSTAT -t file $DIR/$tfile || return 2
215 run_test 12 "open resend timeout"
218 multiop $DIR/$tfile mo_c &
222 replay_barrier $SINGLEMDS
224 kill -USR1 $MULTIPID || return 3
225 wait $MULTIPID || return 4
228 do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0x80000115
229 facet_failover $SINGLEMDS
230 do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0
231 df $MOUNT || return 1
234 $CHECKSTAT -t file $DIR/$tfile || return 2
239 run_test 13 "close resend timeout"
242 replay_barrier $SINGLEMDS
243 createmany -o $MOUNT1/$tfile- 25
244 createmany -o $MOUNT2/$tfile-2- 1
245 createmany -o $MOUNT1/$tfile-3- 25
248 facet_failover $SINGLEMDS
249 # expect failover to fail due to missing client 2
250 df $MOUNT && return 1
253 # first 25 files should have been replayed
254 unlinkmany $MOUNT1/$tfile- 25 || return 2
256 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
259 run_test 14 "timeouts waiting for lost client during replay"
262 replay_barrier $SINGLEMDS
263 createmany -o $MOUNT1/$tfile- 25
264 createmany -o $MOUNT2/$tfile-2- 1
267 facet_failover $SINGLEMDS
268 df $MOUNT || return 1
270 unlinkmany $MOUNT1/$tfile- 25 || return 2
271 [ -e $MOUNT1/$tfile-2-0 ] && error "$tfile-2-0 exists"
273 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
276 run_test 15 "timeout waiting for lost client during replay, 1 client completes"
282 replay_barrier $SINGLEMDS
283 echo "data" > "$MOUNT2/${tfile}-m2"
286 facet_failover $SINGLEMDS
287 df $MOUNT || return 1
289 ost_last_id=`cat /proc/fs/lustre/obdfilter/*/last_id`
290 mds_last_id=`cat /proc/fs/lustre/osc/*mds*/last_id`
292 echo "Ids after MDS<->OST synchonizing"
293 echo "--------------------------------"
300 echo $ost_last_id | while read id; do
306 echo $mds_last_id | while read id; do
311 local arr_len=${#mds_ids[*]}
312 for ((i=0;i<$arr_len;i++)); do
316 test $mds_id -ge $ost_id || {
317 echo "MDS last id ($mds_id) is smaller than OST one ($ost_id)"
322 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
325 #CROW run_test 15a "OST clear orphans - synchronize ids on MDS and OST"
328 replay_barrier $SINGLEMDS
329 echo "data" > "$MOUNT2/${tfile}-m2"
332 do_facet ost1 "sysctl -w lustre.fail_loc=0x80000802"
333 facet_failover $SINGLEMDS
335 df $MOUNT || return 1
336 do_facet ost1 "sysctl -w lustre.fail_loc=0"
338 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
341 #CROW run_test 15b "multiple delayed OST clear orphans"
344 replay_barrier $SINGLEMDS
345 for ((i = 0; i < 2000; i++)); do
346 echo "data" > "$MOUNT2/${tfile}-$i" || error "create ${tfile}-$i failed"
350 facet_failover $SINGLEMDS
352 df $MOUNT || return 1
354 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
357 run_test 15c "remove multiple OST orphans"
360 replay_barrier $SINGLEMDS
361 createmany -o $MOUNT1/$tfile- 25
362 createmany -o $MOUNT2/$tfile-2- 1
365 facet_failover $SINGLEMDS
367 facet_failover $SINGLEMDS
368 df $MOUNT || return 1
370 unlinkmany $MOUNT1/$tfile- 25 || return 2
372 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
376 run_test 16 "fail MDS during recovery (3571)"
379 createmany -o $MOUNT1/$tfile- 25
380 createmany -o $MOUNT2/$tfile-2- 1
382 # Make sure the disconnect is lost
389 df $MOUNT || return 1
391 unlinkmany $MOUNT1/$tfile- 25 || return 2
393 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
397 run_test 17 "fail OST during recovery (3571)"
399 # cleanup with blocked enqueue fails until timer elapses (MDS busy), wait for it
402 test_18() { # bug 3822 - evicting client with enqueued lock
404 mkdir -p $MOUNT1/$tdir
405 touch $MOUNT1/$tdir/f0
406 #define OBD_FAIL_LDLM_ENQUEUE_BLOCKED 0x30b
407 statmany -s $MOUNT1/$tdir/f 1 500 &
410 do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0x8000030b # hold enqueue
412 #define OBD_FAIL_LDLM_BL_CALLBACK 0x305
413 do_facet client sysctl -w lustre.fail_loc=0x80000305 # drop cb, evict
415 usleep 500 # wait to ensure first client is one that will be evicted
416 openfile -f O_RDONLY $MOUNT2/$tdir/f0
418 dmesg | grep "entering recovery in server" && \
419 error "client not evicted" || true
421 run_test 18 "ldlm_handle_enqueue succeeds on evicted export (3822)"
423 test_19() { # Bug 10991 - resend of open request does not fail assertion.
424 replay_barrier $SINGLEMDS
425 drop_ldlm_reply "createmany -o $DIR/$tfile 1" || return 1
427 checkstat $DIR2/${tfile}0 || return 2
428 rm $DIR/${tfile}0 || return 3
432 run_test 19 "resend of open request"
434 equals_msg `basename $0`: test complete, cleaning up
435 SLEEP=$((`date +%s` - $NOW))
436 [ $SLEEP -lt $TIMEOUT ] && sleep $SLEEP
437 check_and_cleanup_lustre
438 [ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG || true