5 # bug number: 6088 10124 10800
6 ALWAYS_EXCEPT="8 15c 17 $REPLAY_DUAL_EXCEPT"
8 PTLDEBUG=${PTLDEBUG:--1}
9 LUSTRE=${LUSTRE:-`dirname $0`/..}
10 . $LUSTRE/tests/test-framework.sh
14 . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
16 SETUP=${SETUP:-"setup"}
17 CLEANUP=${CLEANUP:-"cleanup"}
25 if [ "$ONLY" == "cleanup" ]; then
26 sysctl -w lnet.debug=0
39 [ "$DAEMONFILE" ] && $LCTL debug_daemon start $DAEMONFILE $DAEMONSIZE
47 checkstat $MOUNT2/a || return 1
48 checkstat $MOUNT1/b || return 2
49 rm $MOUNT2/a $MOUNT1/b
50 checkstat $MOUNT1/a && return 3
51 checkstat $MOUNT2/b && return 4
55 run_test 1 "|X| simple create"
63 checkstat $MOUNT2/adir || return 1
65 checkstat $MOUNT2/adir && return 2
69 run_test 2 "|X| mkdir adir"
74 mkdir $MOUNT2/adir/bdir
77 checkstat $MOUNT2/adir || return 1
78 checkstat $MOUNT1/adir/bdir || return 2
79 rmdir $MOUNT2/adir/bdir $MOUNT1/adir
80 checkstat $MOUNT1/adir && return 3
81 checkstat $MOUNT2/adir/bdir && return 4
85 run_test 3 "|X| mkdir adir, mkdir adir/bdir "
90 mkdir $MOUNT1/adir && return 1
91 mkdir $MOUNT2/adir/bdir
94 checkstat $MOUNT2/adir || return 2
95 checkstat $MOUNT1/adir/bdir || return 3
97 rmdir $MOUNT2/adir/bdir $MOUNT1/adir
98 checkstat $MOUNT1/adir && return 4
99 checkstat $MOUNT2/adir/bdir && return 5
103 run_test 4 "|X| mkdir adir (-EEXIST), mkdir adir/bdir "
107 # multiclient version of replay_single.sh/test_8
109 multiop $MOUNT2/a o_tSc &
111 # give multiop a chance to open
116 wait $pid || return 1
119 [ -e $MOUNT2/a ] && return 2
122 run_test 5 "open, unlink |X| close"
127 multiop $MOUNT2/a o_c &
129 multiop $MOUNT1/a o_c &
131 # give multiop a chance to open
136 wait $pid1 || return 1
140 wait $pid2 || return 1
141 [ -e $MOUNT2/a ] && return 2
144 run_test 6 "open1, open2, unlink |X| close1 [fail mds] close2"
148 drop_reint_reply "mcreate $MOUNT1/$tfile" || return 1
150 checkstat $MOUNT2/$tfile || return 2
151 rm $MOUNT1/$tfile || return 3
155 run_test 8 "replay of resent request"
159 mcreate $MOUNT1/$tfile-1
160 mcreate $MOUNT2/$tfile-2
161 # drop first reint reply
162 sysctl -w lustre.fail_loc=0x80000119
164 sysctl -w lustre.fail_loc=0
166 rm $MOUNT1/$tfile-[1,2] || return 1
170 run_test 9 "resending a replayed create"
173 mcreate $MOUNT1/$tfile-1
175 munlink $MOUNT1/$tfile-1
176 mcreate $MOUNT2/$tfile-2
177 # drop first reint reply
178 sysctl -w lustre.fail_loc=0x80000119
180 sysctl -w lustre.fail_loc=0
182 checkstat $MOUNT1/$tfile-1 && return 1
183 checkstat $MOUNT1/$tfile-2 || return 2
188 run_test 10 "resending a replayed unlink"
192 mcreate $MOUNT1/$tfile-1
193 mcreate $MOUNT2/$tfile-2
194 mcreate $MOUNT1/$tfile-3
195 mcreate $MOUNT2/$tfile-4
196 mcreate $MOUNT1/$tfile-5
197 # drop all reint replies for a while
198 sysctl -w lustre.fail_loc=0x0119
200 #sleep for while, let both clients reconnect and timeout
201 sleep $((TIMEOUT * 2))
202 sysctl -w lustre.fail_loc=0
203 while [ -z "$(ls $MOUNT1/$tfile-[1-5] 2>/dev/null)" ]; do
207 ls $MOUNT1/$tfile-[1-5]
208 rm $MOUNT1/$tfile-[1-5] || return 1
212 run_test 11 "both clients timeout during replay"
217 multiop $DIR/$tfile mo_c &
221 #define OBD_FAIL_LDLM_ENQUEUE 0x302
222 sysctl -w lustre.fail_loc=0x80000302
224 df $MOUNT || return 1
225 sysctl -w lustre.fail_loc=0
228 kill -USR1 $MULTIPID || return 3
229 wait $MULTIPID || return 4
230 $CHECKSTAT -t file $DIR/$tfile || return 2
235 run_test 12 "open resend timeout"
238 multiop $DIR/$tfile mo_c &
244 kill -USR1 $MULTIPID || return 3
245 wait $MULTIPID || return 4
248 sysctl -w lustre.fail_loc=0x80000115
250 df $MOUNT || return 1
251 sysctl -w lustre.fail_loc=0
254 $CHECKSTAT -t file $DIR/$tfile || return 2
259 run_test 13 "close resend timeout"
263 createmany -o $MOUNT1/$tfile- 25
264 createmany -o $MOUNT2/$tfile-2- 1
265 createmany -o $MOUNT1/$tfile-3- 25
269 # expect failover to fail
270 df $MOUNT && return 1
273 # first 25 files should have been replayed
274 unlinkmany $MOUNT1/$tfile- 25 || return 2
276 zconf_mount `hostname` $MOUNT2
279 run_test 14 "timeouts waiting for lost client during replay"
283 createmany -o $MOUNT1/$tfile- 25
284 createmany -o $MOUNT2/$tfile-2- 1
288 df $MOUNT || return 1
290 unlinkmany $MOUNT1/$tfile- 25 || return 2
291 [ -e $MOUNT1/$tfile-2-0 ] && error "$tfile-2-0 exists"
293 zconf_mount `hostname` $MOUNT2
296 run_test 15 "timeout waiting for lost client during replay, 1 client completes"
303 echo "data" > "$MOUNT2/${tfile}-m2"
307 df $MOUNT || return 1
309 ost_last_id=`cat /proc/fs/lustre/obdfilter/*/last_id`
310 mds_last_id=`cat /proc/fs/lustre/osc/*mds*/last_id`
312 echo "Ids after MDS<->OST synchonizing"
313 echo "--------------------------------"
320 echo $ost_last_id | while read id; do
326 echo $mds_last_id | while read id; do
331 local arr_len=${#mds_ids[*]}
332 for ((i=0;i<$arr_len;i++)); do
336 test $mds_id -ge $ost_id || {
337 echo "MDS last id ($mds_id) is smaller than OST one ($ost_id)"
342 zconf_mount `hostname` $MOUNT2
345 #CROW run_test 15a "OST clear orphans - synchronize ids on MDS and OST"
349 echo "data" > "$MOUNT2/${tfile}-m2"
352 do_facet ost1 "sysctl -w lustre.fail_loc=0x80000802"
355 df $MOUNT || return 1
356 do_facet ost1 "sysctl -w lustre.fail_loc=0"
358 zconf_mount `hostname` $MOUNT2
361 #CROW run_test 15b "multiple delayed OST clear orphans"
365 for ((i = 0; i < 2000; i++)); do
366 echo "data" > "$MOUNT2/${tfile}-$i" || error "create ${tfile}-$i failed"
372 df $MOUNT || return 1
374 zconf_mount `hostname` $MOUNT2
377 run_test 15c "remove multiple OST orphans"
381 createmany -o $MOUNT1/$tfile- 25
382 createmany -o $MOUNT2/$tfile-2- 1
388 df $MOUNT || return 1
390 unlinkmany $MOUNT1/$tfile- 25 || return 2
392 zconf_mount `hostname` $MOUNT2
396 run_test 16 "fail MDS during recovery (3571)"
399 createmany -o $MOUNT1/$tfile- 25
400 createmany -o $MOUNT2/$tfile-2- 1
402 # Make sure the disconnect is lost
409 df $MOUNT || return 1
411 unlinkmany $MOUNT1/$tfile- 25 || return 2
413 zconf_mount `hostname` $MOUNT2
417 run_test 17 "fail OST during recovery (3571)"
419 # cleanup with blocked enqueue fails until timer elapses (MDS busy), wait for it
422 test_18() { # bug 3822 - evicting client with enqueued lock
424 mkdir -p $MOUNT1/$tdir
425 touch $MOUNT1/$tdir/f0
426 #define OBD_FAIL_LDLM_ENQUEUE_BLOCKED 0x30b
427 statmany -s $MOUNT1/$tdir/f 1 500 &
430 do_facet mds sysctl -w lustre.fail_loc=0x8000030b # hold enqueue
432 #define OBD_FAIL_LDLM_BL_CALLBACK 0x305
433 do_facet client sysctl -w lustre.fail_loc=0x80000305 # drop cb, evict
435 usleep 500 # wait to ensure first client is one that will be evicted
436 openfile -f O_RDONLY $MOUNT2/$tdir/f0
438 dmesg | grep "entering recovery in server" && \
439 error "client not evicted" || true
441 run_test 18 "ldlm_handle_enqueue succeeds on evicted export (3822)"
443 test_19() { # Bug 10991 - resend of open request does not fail assertion.
445 drop_ldlm_reply "createmany -o $DIR/$tfile 1" || return 1
447 checkstat $DIR2/${tfile}0 || return 2
448 rm $DIR/${tfile}0 || return 3
452 run_test 19 "resend of open request"
454 if [ "$ONLY" != "setup" ]; then
455 equals_msg test complete, cleaning up
456 SLEEP=$((`date +%s` - $NOW))
457 [ $SLEEP -lt $TIMEOUT ] && sleep $SLEEP