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/$NAME.sh}
21 [ "$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 $MOUNT2/a o_tSc &
97 # give multiop a chance to open
100 replay_barrier $SINGLEMDS
102 wait $pid || return 1
105 [ -e $MOUNT2/a ] && return 2
108 run_test 5 "open, unlink |X| close"
113 multiop $MOUNT2/a o_c &
115 multiop $MOUNT1/a o_c &
117 # give multiop a chance to open
120 replay_barrier $SINGLEMDS
122 wait $pid1 || return 1
126 wait $pid2 || return 1
127 [ -e $MOUNT2/a ] && return 2
130 run_test 6 "open1, open2, unlink |X| close1 [fail $SINGLEMDS] close2"
133 replay_barrier $SINGLEMDS
134 drop_reint_reply "mcreate $MOUNT1/$tfile" || return 1
136 checkstat $MOUNT2/$tfile || return 2
137 rm $MOUNT1/$tfile || return 3
141 run_test 8 "replay of resent request"
144 replay_barrier $SINGLEMDS
145 mcreate $MOUNT1/$tfile-1
146 mcreate $MOUNT2/$tfile-2
147 # drop first reint reply
148 do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0x80000119
150 do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0
152 rm $MOUNT1/$tfile-[1,2] || return 1
156 run_test 9 "resending a replayed create"
159 mcreate $MOUNT1/$tfile-1
160 replay_barrier $SINGLEMDS
161 munlink $MOUNT1/$tfile-1
162 mcreate $MOUNT2/$tfile-2
163 # drop first reint reply
164 do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0x80000119
166 do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0
168 checkstat $MOUNT1/$tfile-1 && return 1
169 checkstat $MOUNT1/$tfile-2 || return 2
174 run_test 10 "resending a replayed unlink"
177 replay_barrier $SINGLEMDS
178 mcreate $MOUNT1/$tfile-1
179 mcreate $MOUNT2/$tfile-2
180 mcreate $MOUNT1/$tfile-3
181 mcreate $MOUNT2/$tfile-4
182 mcreate $MOUNT1/$tfile-5
183 # drop all reint replies for a while
184 do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0x0119
185 # note that with this fail_loc set, facet_failover df will fail
186 facet_failover $SINGLEMDS
187 #sleep for while, let both clients reconnect and timeout
188 sleep $((TIMEOUT * 2))
189 do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0
191 rm $MOUNT1/$tfile-[1-5] || return 1
195 run_test 11 "both clients timeout during replay"
198 replay_barrier $SINGLEMDS
200 multiop $DIR/$tfile mo_c &
204 #define OBD_FAIL_LDLM_ENQUEUE 0x302
205 do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0x80000302
206 facet_failover $SINGLEMDS
207 do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0
208 df $MOUNT || return 1
211 kill -USR1 $MULTIPID || return 3
212 wait $MULTIPID || return 4
213 $CHECKSTAT -t file $DIR/$tfile || return 2
218 run_test 12 "open resend timeout"
221 multiop $DIR/$tfile mo_c &
225 replay_barrier $SINGLEMDS
227 kill -USR1 $MULTIPID || return 3
228 wait $MULTIPID || return 4
231 do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0x80000115
232 facet_failover $SINGLEMDS
233 do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0
234 df $MOUNT || return 1
237 $CHECKSTAT -t file $DIR/$tfile || return 2
242 run_test 13 "close resend timeout"
245 replay_barrier $SINGLEMDS
246 createmany -o $MOUNT1/$tfile- 25
247 createmany -o $MOUNT2/$tfile-2- 1
248 createmany -o $MOUNT1/$tfile-3- 25
251 facet_failover $SINGLEMDS
252 # expect failover to fail due to missing client 2
253 df $MOUNT && return 1
256 # first 25 files should have been replayed
257 unlinkmany $MOUNT1/$tfile- 25 || return 2
259 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
262 run_test 14 "timeouts waiting for lost client during replay"
265 replay_barrier $SINGLEMDS
266 createmany -o $MOUNT1/$tfile- 25
267 createmany -o $MOUNT2/$tfile-2- 1
270 facet_failover $SINGLEMDS
271 df $MOUNT || return 1
273 unlinkmany $MOUNT1/$tfile- 25 || return 2
274 [ -e $MOUNT1/$tfile-2-0 ] && error "$tfile-2-0 exists"
276 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
279 run_test 15 "timeout waiting for lost client during replay, 1 client completes"
285 replay_barrier $SINGLEMDS
286 echo "data" > "$MOUNT2/${tfile}-m2"
289 facet_failover $SINGLEMDS
290 df $MOUNT || return 1
292 ost_last_id=`cat /proc/fs/lustre/obdfilter/*/last_id`
293 mds_last_id=`cat /proc/fs/lustre/osc/*mds*/last_id`
295 echo "Ids after MDS<->OST synchonizing"
296 echo "--------------------------------"
303 echo $ost_last_id | while read id; do
309 echo $mds_last_id | while read id; do
314 local arr_len=${#mds_ids[*]}
315 for ((i=0;i<$arr_len;i++)); do
319 test $mds_id -ge $ost_id || {
320 echo "MDS last id ($mds_id) is smaller than OST one ($ost_id)"
325 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
328 #CROW run_test 15a "OST clear orphans - synchronize ids on MDS and OST"
331 replay_barrier $SINGLEMDS
332 echo "data" > "$MOUNT2/${tfile}-m2"
335 do_facet ost1 "sysctl -w lustre.fail_loc=0x80000802"
336 facet_failover $SINGLEMDS
338 df $MOUNT || return 1
339 do_facet ost1 "sysctl -w lustre.fail_loc=0"
341 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
344 #CROW run_test 15b "multiple delayed OST clear orphans"
347 replay_barrier $SINGLEMDS
348 for ((i = 0; i < 2000; i++)); do
349 echo "data" > "$MOUNT2/${tfile}-$i" || error "create ${tfile}-$i failed"
353 facet_failover $SINGLEMDS
355 df $MOUNT || return 1
357 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
360 run_test 15c "remove multiple OST orphans"
363 replay_barrier $SINGLEMDS
364 createmany -o $MOUNT1/$tfile- 25
365 createmany -o $MOUNT2/$tfile-2- 1
368 facet_failover $SINGLEMDS
370 facet_failover $SINGLEMDS
371 df $MOUNT || return 1
373 unlinkmany $MOUNT1/$tfile- 25 || return 2
375 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
379 run_test 16 "fail MDS during recovery (3571)"
382 createmany -o $MOUNT1/$tfile- 25
383 createmany -o $MOUNT2/$tfile-2- 1
385 # Make sure the disconnect is lost
392 df $MOUNT || return 1
394 unlinkmany $MOUNT1/$tfile- 25 || return 2
396 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
400 run_test 17 "fail OST during recovery (3571)"
402 # cleanup with blocked enqueue fails until timer elapses (MDS busy), wait for it
405 test_18() { # bug 3822 - evicting client with enqueued lock
407 mkdir -p $MOUNT1/$tdir
408 touch $MOUNT1/$tdir/f0
409 #define OBD_FAIL_LDLM_ENQUEUE_BLOCKED 0x30b
410 statmany -s $MOUNT1/$tdir/f 1 500 &
413 do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0x8000030b # hold enqueue
415 #define OBD_FAIL_LDLM_BL_CALLBACK 0x305
416 do_facet client sysctl -w lustre.fail_loc=0x80000305 # drop cb, evict
418 usleep 500 # wait to ensure first client is one that will be evicted
419 openfile -f O_RDONLY $MOUNT2/$tdir/f0
421 dmesg | grep "entering recovery in server" && \
422 error "client not evicted" || true
423 do_facet client "sysctl -w lustre.fail_loc=0"
424 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
426 run_test 18 "ldlm_handle_enqueue succeeds on evicted export (3822)"
428 test_19() { # Bug 10991 - resend of open request does not fail assertion.
429 replay_barrier $SINGLEMDS
430 drop_ldlm_reply "createmany -o $DIR/$tfile 1" || return 1
432 checkstat $DIR2/${tfile}0 || return 2
433 rm $DIR/${tfile}0 || return 3
437 run_test 19 "resend of open request"
439 equals_msg `basename $0`: test complete, cleaning up
440 SLEEP=$((`date +%s` - $NOW))
441 [ $SLEEP -lt $TIMEOUT ] && sleep $SLEEP
442 check_and_cleanup_lustre
443 [ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG || true