6 ALWAYS_EXCEPT="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
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"
52 checkstat $MOUNT2/adir || return 1
54 checkstat $MOUNT2/adir && return 2
57 run_test 2 "|X| mkdir adir"
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 "
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
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
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 mds] close2"
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"
145 mcreate $MOUNT1/$tfile-1
146 mcreate $MOUNT2/$tfile-2
147 # drop first reint reply
148 do_facet mds sysctl -w lustre.fail_loc=0x80000119
150 do_facet mds 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
161 munlink $MOUNT1/$tfile-1
162 mcreate $MOUNT2/$tfile-2
163 # drop first reint reply
164 do_facet mds sysctl -w lustre.fail_loc=0x80000119
166 do_facet mds 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"
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 mds sysctl -w lustre.fail_loc=0x0119
185 # note that with this fail_loc set, facet_failover df will fail
187 #sleep for while, let both clients reconnect and timeout
188 sleep $((TIMEOUT * 2))
189 do_facet mds sysctl -w lustre.fail_loc=0
191 while [ -z "$(ls $MOUNT1/$tfile-[1-5] 2>/dev/null)" ]; do
195 ls $MOUNT1/$tfile-[1-5]
196 rm $MOUNT1/$tfile-[1-5] || return 1
200 run_test 11 "both clients timeout during replay"
205 multiop $DIR/$tfile mo_c &
209 #define OBD_FAIL_LDLM_ENQUEUE 0x302
210 do_facet mds sysctl -w lustre.fail_loc=0x80000302
212 do_facet mds sysctl -w lustre.fail_loc=0
213 df $MOUNT || { kill -USR1 $MULTIPID && return 1; }
216 kill -USR1 $MULTIPID || return 3
217 wait $MULTIPID || return 4
218 $CHECKSTAT -t file $DIR/$tfile || return 2
223 run_test 12 "open resend timeout"
226 multiop $DIR/$tfile mo_c &
232 kill -USR1 $MULTIPID || return 3
233 wait $MULTIPID || return 4
236 do_facet mds sysctl -w lustre.fail_loc=0x80000115
238 do_facet mds sysctl -w lustre.fail_loc=0
239 df $MOUNT || return 1
242 $CHECKSTAT -t file $DIR/$tfile || return 2
247 run_test 13 "close resend timeout"
251 createmany -o $MOUNT1/$tfile- 25
252 createmany -o $MOUNT2/$tfile-2- 1
253 createmany -o $MOUNT1/$tfile-3- 25
257 # expect recovery to fail due to missing client 2
258 df $MOUNT && return 1
261 # first 25 files should have been replayed
262 unlinkmany $MOUNT1/$tfile- 25 || return 2
264 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
267 run_test 14 "timeouts waiting for lost client during replay"
271 createmany -o $MOUNT1/$tfile- 25
272 createmany -o $MOUNT2/$tfile-2- 1
276 df $MOUNT || return 1
278 unlinkmany $MOUNT1/$tfile- 25 || return 2
279 [ -e $MOUNT1/$tfile-2-0 ] && error "$tfile-2-0 exists"
281 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
284 run_test 15 "timeout waiting for lost client during replay, 1 client completes"
291 echo "data" > "$MOUNT2/${tfile}-m2"
295 df $MOUNT || return 1
297 ost_last_id=`cat /proc/fs/lustre/obdfilter/*/last_id`
298 mds_last_id=`cat /proc/fs/lustre/osc/*mds*/last_id`
300 echo "Ids after MDS<->OST synchonizing"
301 echo "--------------------------------"
308 echo $ost_last_id | while read id; do
314 echo $mds_last_id | while read id; do
319 local arr_len=${#mds_ids[*]}
320 for ((i=0;i<$arr_len;i++)); do
324 test $mds_id -ge $ost_id || {
325 echo "MDS last id ($mds_id) is smaller than OST one ($ost_id)"
330 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
333 #CROW run_test 15a "OST clear orphans - synchronize ids on MDS and OST"
337 echo "data" > "$MOUNT2/${tfile}-m2"
340 do_facet ost1 "sysctl -w lustre.fail_loc=0x80000802"
343 df $MOUNT || return 1
344 do_facet ost1 "sysctl -w lustre.fail_loc=0"
346 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
349 #CROW run_test 15b "multiple delayed OST clear orphans"
353 for ((i = 0; i < 2000; i++)); do
354 echo "data" > "$MOUNT2/${tfile}-$i" || error "create ${tfile}-$i failed"
360 df $MOUNT || return 1
362 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
365 run_test 15c "remove multiple OST orphans"
369 createmany -o $MOUNT1/$tfile- 25
370 createmany -o $MOUNT2/$tfile-2- 1
376 df $MOUNT || return 1
378 unlinkmany $MOUNT1/$tfile- 25 || return 2
380 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
384 run_test 16 "fail MDS during recovery (3571)"
387 createmany -o $MOUNT1/$tfile- 25
388 createmany -o $MOUNT2/$tfile-2- 1
390 # Make sure the disconnect is lost
397 df $MOUNT || return 1
399 unlinkmany $MOUNT1/$tfile- 25 || return 2
401 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
405 run_test 17 "fail OST during recovery (3571)"
407 # cleanup with blocked enqueue fails until timer elapses (MDS busy), wait for it
410 test_18() { # bug 3822 - evicting client with enqueued lock
412 mkdir -p $MOUNT1/$tdir
413 touch $MOUNT1/$tdir/f0
414 #define OBD_FAIL_LDLM_ENQUEUE_BLOCKED 0x30b
415 statmany -s $MOUNT1/$tdir/f 1 500 &
418 do_facet mds sysctl -w lustre.fail_loc=0x8000030b # hold enqueue
420 #define OBD_FAIL_LDLM_BL_CALLBACK 0x305
421 do_facet client sysctl -w lustre.fail_loc=0x80000305 # drop cb, evict
423 sleep 0.500s # wait to ensure first client is one that will be evicted
424 openfile -f O_RDONLY $MOUNT2/$tdir/f0
426 dmesg | grep "entering recovery in server" && \
427 error "client not evicted" || true
429 run_test 18 "ldlm_handle_enqueue succeeds on evicted export (3822)"
431 test_19() { # Bug 10991 - resend of open request does not fail assertion.
433 drop_ldlm_reply "createmany -o $DIR/$tfile 1" || return 1
435 checkstat $DIR2/${tfile}0 || return 2
436 rm $DIR/${tfile}0 || return 3
440 run_test 19 "resend of open request"
442 equals_msg `basename $0`: test complete, cleaning up
443 SLEEP=$((`date +%s` - $NOW))
444 [ $SLEEP -lt $TIMEOUT ] && sleep $SLEEP
445 check_and_cleanup_lustre
446 [ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG || true