6 ALWAYS_EXCEPT="15c $REPLAY_DUAL_EXCEPT"
9 [ "$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 $DIR/${TESTSUITE}/[df][0-9]* # bug 13798 new t-f tdir staff
28 rm -rf $DIR/[df][0-9]*
30 [ "$DAEMONFILE" ] && $LCTL debug_daemon start $DAEMONFILE $DAEMONSIZE
38 checkstat $MOUNT2/a || return 1
39 checkstat $MOUNT1/b || return 2
40 rm $MOUNT2/a $MOUNT1/b
41 checkstat $MOUNT1/a && return 3
42 checkstat $MOUNT2/b && return 4
46 run_test 1 "|X| simple create"
54 checkstat $MOUNT2/adir || return 1
56 checkstat $MOUNT2/adir && return 2
59 run_test 2 "|X| mkdir adir"
64 mkdir $MOUNT2/adir/bdir
67 checkstat $MOUNT2/adir || return 1
68 checkstat $MOUNT1/adir/bdir || return 2
69 rmdir $MOUNT2/adir/bdir $MOUNT1/adir
70 checkstat $MOUNT1/adir && return 3
71 checkstat $MOUNT2/adir/bdir && return 4
74 run_test 3 "|X| mkdir adir, mkdir adir/bdir "
79 mkdir $MOUNT1/adir && return 1
80 mkdir $MOUNT2/adir/bdir
83 checkstat $MOUNT2/adir || return 2
84 checkstat $MOUNT1/adir/bdir || return 3
86 rmdir $MOUNT2/adir/bdir $MOUNT1/adir
87 checkstat $MOUNT1/adir && return 4
88 checkstat $MOUNT2/adir/bdir && return 5
91 run_test 4 "|X| mkdir adir (-EEXIST), mkdir adir/bdir "
95 # multiclient version of replay_single.sh/test_8
97 multiop $MOUNT2/a o_tSc &
99 # give multiop a chance to open
104 wait $pid || return 1
107 [ -e $MOUNT2/a ] && return 2
110 run_test 5 "open, unlink |X| close"
115 multiop $MOUNT2/a o_c &
117 multiop $MOUNT1/a o_c &
119 # give multiop a chance to open
124 wait $pid1 || return 1
128 wait $pid2 || return 1
129 [ -e $MOUNT2/a ] && return 2
132 run_test 6 "open1, open2, unlink |X| close1 [fail mds] close2"
136 drop_reint_reply "mcreate $MOUNT1/$tfile" || return 1
138 checkstat $MOUNT2/$tfile || return 2
139 rm $MOUNT1/$tfile || return 3
143 run_test 8 "replay of resent request"
147 mcreate $MOUNT1/$tfile-1
148 mcreate $MOUNT2/$tfile-2
149 # drop first reint reply
150 do_facet mds sysctl -w lustre.fail_loc=0x80000119
152 do_facet mds sysctl -w lustre.fail_loc=0
154 rm $MOUNT1/$tfile-[1,2] || return 1
158 run_test 9 "resending a replayed create"
161 mcreate $MOUNT1/$tfile-1
163 munlink $MOUNT1/$tfile-1
164 mcreate $MOUNT2/$tfile-2
165 # drop first reint reply
166 do_facet mds sysctl -w lustre.fail_loc=0x80000119
168 do_facet mds sysctl -w lustre.fail_loc=0
170 checkstat $MOUNT1/$tfile-1 && return 1
171 checkstat $MOUNT1/$tfile-2 || return 2
176 run_test 10 "resending a replayed unlink"
180 mcreate $MOUNT1/$tfile-1
181 mcreate $MOUNT2/$tfile-2
182 mcreate $MOUNT1/$tfile-3
183 mcreate $MOUNT2/$tfile-4
184 mcreate $MOUNT1/$tfile-5
185 # drop all reint replies for a while
186 do_facet mds sysctl -w lustre.fail_loc=0x0119
187 # note that with this fail_loc set, facet_failover df will fail
189 #sleep for while, let both clients reconnect and timeout
190 sleep $((TIMEOUT * 2))
191 do_facet mds sysctl -w lustre.fail_loc=0
193 while [ -z "$(ls $MOUNT1/$tfile-[1-5] 2>/dev/null)" ]; do
197 ls $MOUNT1/$tfile-[1-5]
198 rm $MOUNT1/$tfile-[1-5] || return 1
202 run_test 11 "both clients timeout during replay"
207 multiop $DIR/$tfile mo_c &
211 #define OBD_FAIL_LDLM_ENQUEUE 0x302
212 do_facet mds sysctl -w lustre.fail_loc=0x80000302
214 do_facet mds sysctl -w lustre.fail_loc=0
215 df $MOUNT || { kill -USR1 $MULTIPID && return 1; }
218 kill -USR1 $MULTIPID || return 3
219 wait $MULTIPID || return 4
220 $CHECKSTAT -t file $DIR/$tfile || return 2
225 run_test 12 "open resend timeout"
228 multiop $DIR/$tfile mo_c &
234 kill -USR1 $MULTIPID || return 3
235 wait $MULTIPID || return 4
238 do_facet mds sysctl -w lustre.fail_loc=0x80000115
240 do_facet mds sysctl -w lustre.fail_loc=0
241 df $MOUNT || return 1
244 $CHECKSTAT -t file $DIR/$tfile || return 2
249 run_test 13 "close resend timeout"
253 createmany -o $MOUNT1/$tfile- 25
254 createmany -o $MOUNT2/$tfile-2- 1
255 createmany -o $MOUNT1/$tfile-3- 25
259 # expect recovery to fail due to missing client 2
260 df $MOUNT && return 1
263 # first 25 files should have been replayed
264 unlinkmany $MOUNT1/$tfile- 25 || return 2
266 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
269 run_test 14 "timeouts waiting for lost client during replay"
273 createmany -o $MOUNT1/$tfile- 25
274 createmany -o $MOUNT2/$tfile-2- 1
278 df $MOUNT || return 1
280 unlinkmany $MOUNT1/$tfile- 25 || return 2
281 [ -e $MOUNT1/$tfile-2-0 ] && error "$tfile-2-0 exists"
283 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
286 run_test 15 "timeout waiting for lost client during replay, 1 client completes"
293 echo "data" > "$MOUNT2/${tfile}-m2"
297 df $MOUNT || return 1
299 ost_last_id=`cat /proc/fs/lustre/obdfilter/*/last_id`
300 mds_last_id=`cat /proc/fs/lustre/osc/*mds*/last_id`
302 echo "Ids after MDS<->OST synchonizing"
303 echo "--------------------------------"
310 echo $ost_last_id | while read id; do
316 echo $mds_last_id | while read id; do
321 local arr_len=${#mds_ids[*]}
322 for ((i=0;i<$arr_len;i++)); do
326 test $mds_id -ge $ost_id || {
327 echo "MDS last id ($mds_id) is smaller than OST one ($ost_id)"
332 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
335 #CROW run_test 15a "OST clear orphans - synchronize ids on MDS and OST"
339 echo "data" > "$MOUNT2/${tfile}-m2"
342 do_facet ost1 "sysctl -w lustre.fail_loc=0x80000802"
345 df $MOUNT || return 1
346 do_facet ost1 "sysctl -w lustre.fail_loc=0"
348 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
351 #CROW run_test 15b "multiple delayed OST clear orphans"
355 for ((i = 0; i < 2000; i++)); do
356 echo "data" > "$MOUNT2/${tfile}-$i" || error "create ${tfile}-$i failed"
362 df $MOUNT || return 1
364 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
367 run_test 15c "remove multiple OST orphans"
371 createmany -o $MOUNT1/$tfile- 25
372 createmany -o $MOUNT2/$tfile-2- 1
378 df $MOUNT || return 1
380 unlinkmany $MOUNT1/$tfile- 25 || return 2
382 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
386 run_test 16 "fail MDS during recovery (3571)"
389 createmany -o $MOUNT1/$tfile- 25
390 createmany -o $MOUNT2/$tfile-2- 1
392 # Make sure the disconnect is lost
399 df $MOUNT || return 1
401 unlinkmany $MOUNT1/$tfile- 25 || return 2
403 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
407 run_test 17 "fail OST during recovery (3571)"
409 # cleanup with blocked enqueue fails until timer elapses (MDS busy), wait for it
412 test_18() { # bug 3822 - evicting client with enqueued lock
414 mkdir -p $MOUNT1/$tdir
415 touch $MOUNT1/$tdir/f0
416 #define OBD_FAIL_LDLM_ENQUEUE_BLOCKED 0x30b
417 statmany -s $MOUNT1/$tdir/f 1 500 &
420 do_facet mds sysctl -w lustre.fail_loc=0x8000030b # hold enqueue
422 #define OBD_FAIL_LDLM_BL_CALLBACK 0x305
423 do_facet client sysctl -w lustre.fail_loc=0x80000305 # drop cb, evict
425 sleep 0.500s # wait to ensure first client is one that will be evicted
426 openfile -f O_RDONLY $MOUNT2/$tdir/f0
428 dmesg | grep "entering recovery in server" && \
429 error "client not evicted" || true
431 run_test 18 "ldlm_handle_enqueue succeeds on evicted export (3822)"
433 test_19() { # Bug 10991 - resend of open request does not fail assertion.
435 drop_ldlm_reply "createmany -o $DIR/$tfile 1" || return 1
437 checkstat $DIR2/${tfile}0 || return 2
438 rm $DIR/${tfile}0 || return 3
442 run_test 19 "resend of open request"
444 equals_msg `basename $0`: test complete, cleaning up
445 SLEEP=$((`date +%s` - $NOW))
446 [ $SLEEP -lt $TIMEOUT ] && sleep $SLEEP
447 check_and_cleanup_lustre
448 [ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG || true