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
24 rm -rf $DIR/${TESTSUITE}/[df][0-9]* # bug 13798 new t-f tdir staff
25 rm -rf $DIR/[df][0-9]*
27 [ "$DAEMONFILE" ] && $LCTL debug_daemon start $DAEMONFILE $DAEMONSIZE
31 replay_barrier $SINGLEMDS
35 checkstat $MOUNT2/a || return 1
36 checkstat $MOUNT1/b || return 2
37 rm $MOUNT2/a $MOUNT1/b
38 checkstat $MOUNT1/a && return 3
39 checkstat $MOUNT2/b && return 4
43 run_test 1 "|X| simple create"
47 replay_barrier $SINGLEMDS
51 checkstat $MOUNT2/adir || return 1
53 checkstat $MOUNT2/adir && return 2
56 run_test 2 "|X| mkdir adir"
59 replay_barrier $SINGLEMDS
61 mkdir $MOUNT2/adir/bdir
64 checkstat $MOUNT2/adir || return 1
65 checkstat $MOUNT1/adir/bdir || return 2
66 rmdir $MOUNT2/adir/bdir $MOUNT1/adir
67 checkstat $MOUNT1/adir && return 3
68 checkstat $MOUNT2/adir/bdir && return 4
71 run_test 3 "|X| mkdir adir, mkdir adir/bdir "
75 replay_barrier $SINGLEMDS
76 mkdir $MOUNT1/adir && return 1
77 mkdir $MOUNT2/adir/bdir
80 checkstat $MOUNT2/adir || return 2
81 checkstat $MOUNT1/adir/bdir || return 3
83 rmdir $MOUNT2/adir/bdir $MOUNT1/adir
84 checkstat $MOUNT1/adir && return 4
85 checkstat $MOUNT2/adir/bdir && return 5
88 run_test 4 "|X| mkdir adir (-EEXIST), mkdir adir/bdir "
92 # multiclient version of replay_single.sh/test_8
94 multiop $MOUNT2/a o_tSc &
96 # give multiop a chance to open
99 replay_barrier $SINGLEMDS
101 wait $pid || return 1
104 [ -e $MOUNT2/a ] && return 2
107 run_test 5 "open, unlink |X| close"
112 multiop $MOUNT2/a o_c &
114 multiop $MOUNT1/a o_c &
116 # give multiop a chance to open
119 replay_barrier $SINGLEMDS
121 wait $pid1 || return 1
125 wait $pid2 || return 1
126 [ -e $MOUNT2/a ] && return 2
129 run_test 6 "open1, open2, unlink |X| close1 [fail $SINGLEMDS] close2"
132 replay_barrier $SINGLEMDS
133 drop_reint_reply "mcreate $MOUNT1/$tfile" || return 1
135 checkstat $MOUNT2/$tfile || return 2
136 rm $MOUNT1/$tfile || return 3
140 run_test 8 "replay of resent request"
143 replay_barrier $SINGLEMDS
144 mcreate $MOUNT1/$tfile-1
145 mcreate $MOUNT2/$tfile-2
146 # drop first reint reply
147 do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0x80000119
149 do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0
151 rm $MOUNT1/$tfile-[1,2] || return 1
155 run_test 9 "resending a replayed create"
158 mcreate $MOUNT1/$tfile-1
159 replay_barrier $SINGLEMDS
160 munlink $MOUNT1/$tfile-1
161 mcreate $MOUNT2/$tfile-2
162 # drop first reint reply
163 do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0x80000119
165 do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0
167 checkstat $MOUNT1/$tfile-1 && return 1
168 checkstat $MOUNT1/$tfile-2 || return 2
173 run_test 10 "resending a replayed unlink"
176 replay_barrier $SINGLEMDS
177 mcreate $MOUNT1/$tfile-1
178 mcreate $MOUNT2/$tfile-2
179 mcreate $MOUNT1/$tfile-3
180 mcreate $MOUNT2/$tfile-4
181 mcreate $MOUNT1/$tfile-5
182 # drop all reint replies for a while
183 do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0x0119
184 # note that with this fail_loc set, facet_failover df will fail
185 facet_failover $SINGLEMDS
186 #sleep for while, let both clients reconnect and timeout
187 sleep $((TIMEOUT * 2))
188 do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0
190 rm $MOUNT1/$tfile-[1-5] || return 1
194 run_test 11 "both clients timeout during replay"
197 replay_barrier $SINGLEMDS
199 multiop $DIR/$tfile mo_c &
203 #define OBD_FAIL_LDLM_ENQUEUE 0x302
204 do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0x80000302
205 facet_failover $SINGLEMDS
206 do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0
207 df $MOUNT || return 1
210 kill -USR1 $MULTIPID || return 3
211 wait $MULTIPID || return 4
212 $CHECKSTAT -t file $DIR/$tfile || return 2
217 run_test 12 "open resend timeout"
220 multiop $DIR/$tfile mo_c &
224 replay_barrier $SINGLEMDS
226 kill -USR1 $MULTIPID || return 3
227 wait $MULTIPID || return 4
230 do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0x80000115
231 facet_failover $SINGLEMDS
232 do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0
233 df $MOUNT || return 1
236 $CHECKSTAT -t file $DIR/$tfile || return 2
241 run_test 13 "close resend timeout"
244 replay_barrier $SINGLEMDS
245 createmany -o $MOUNT1/$tfile- 25
246 createmany -o $MOUNT2/$tfile-2- 1
247 createmany -o $MOUNT1/$tfile-3- 25
250 facet_failover $SINGLEMDS
251 # expect failover to fail due to missing client 2
252 df $MOUNT && return 1
255 # first 25 files should have been replayed
256 unlinkmany $MOUNT1/$tfile- 25 || return 2
258 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
261 run_test 14 "timeouts waiting for lost client during replay"
264 replay_barrier $SINGLEMDS
265 createmany -o $MOUNT1/$tfile- 25
266 createmany -o $MOUNT2/$tfile-2- 1
269 facet_failover $SINGLEMDS
270 df $MOUNT || return 1
272 unlinkmany $MOUNT1/$tfile- 25 || return 2
273 [ -e $MOUNT1/$tfile-2-0 ] && error "$tfile-2-0 exists"
275 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
278 run_test 15 "timeout waiting for lost client during replay, 1 client completes"
284 replay_barrier $SINGLEMDS
285 echo "data" > "$MOUNT2/${tfile}-m2"
288 facet_failover $SINGLEMDS
289 df $MOUNT || return 1
291 ost_last_id=`cat /proc/fs/lustre/obdfilter/*/last_id`
292 mds_last_id=`cat /proc/fs/lustre/osc/*mds*/last_id`
294 echo "Ids after MDS<->OST synchonizing"
295 echo "--------------------------------"
302 echo $ost_last_id | while read id; do
308 echo $mds_last_id | while read id; do
313 local arr_len=${#mds_ids[*]}
314 for ((i=0;i<$arr_len;i++)); do
318 test $mds_id -ge $ost_id || {
319 echo "MDS last id ($mds_id) is smaller than OST one ($ost_id)"
324 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
327 #CROW run_test 15a "OST clear orphans - synchronize ids on MDS and OST"
330 replay_barrier $SINGLEMDS
331 echo "data" > "$MOUNT2/${tfile}-m2"
334 do_facet ost1 "sysctl -w lustre.fail_loc=0x80000802"
335 facet_failover $SINGLEMDS
337 df $MOUNT || return 1
338 do_facet ost1 "sysctl -w lustre.fail_loc=0"
340 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
343 #CROW run_test 15b "multiple delayed OST clear orphans"
346 replay_barrier $SINGLEMDS
347 for ((i = 0; i < 2000; i++)); do
348 echo "data" > "$MOUNT2/${tfile}-$i" || error "create ${tfile}-$i failed"
352 facet_failover $SINGLEMDS
354 df $MOUNT || return 1
356 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
359 run_test 15c "remove multiple OST orphans"
362 replay_barrier $SINGLEMDS
363 createmany -o $MOUNT1/$tfile- 25
364 createmany -o $MOUNT2/$tfile-2- 1
367 facet_failover $SINGLEMDS
369 facet_failover $SINGLEMDS
370 df $MOUNT || return 1
372 unlinkmany $MOUNT1/$tfile- 25 || return 2
374 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
378 run_test 16 "fail MDS during recovery (3571)"
381 createmany -o $MOUNT1/$tfile- 25
382 createmany -o $MOUNT2/$tfile-2- 1
384 # Make sure the disconnect is lost
391 df $MOUNT || return 1
393 unlinkmany $MOUNT1/$tfile- 25 || return 2
395 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
399 run_test 17 "fail OST during recovery (3571)"
401 # cleanup with blocked enqueue fails until timer elapses (MDS busy), wait for it
404 test_18() { # bug 3822 - evicting client with enqueued lock
406 mkdir -p $MOUNT1/$tdir
407 touch $MOUNT1/$tdir/f0
408 #define OBD_FAIL_LDLM_ENQUEUE_BLOCKED 0x30b
409 statmany -s $MOUNT1/$tdir/f 1 500 &
412 do_facet $SINGLEMDS sysctl -w lustre.fail_loc=0x8000030b # hold enqueue
414 #define OBD_FAIL_LDLM_BL_CALLBACK 0x305
415 do_facet client sysctl -w lustre.fail_loc=0x80000305 # drop cb, evict
417 usleep 500 # wait to ensure first client is one that will be evicted
418 openfile -f O_RDONLY $MOUNT2/$tdir/f0
420 dmesg | grep "entering recovery in server" && \
421 error "client not evicted" || true
423 run_test 18 "ldlm_handle_enqueue succeeds on evicted export (3822)"
425 test_19() { # Bug 10991 - resend of open request does not fail assertion.
426 replay_barrier $SINGLEMDS
427 drop_ldlm_reply "createmany -o $DIR/$tfile 1" || return 1
429 checkstat $DIR2/${tfile}0 || return 2
430 rm $DIR/${tfile}0 || return 3
434 run_test 19 "resend of open request"
436 equals_msg `basename $0`: test complete, cleaning up
437 SLEEP=$((`date +%s` - $NOW))
438 [ $SLEEP -lt $TIMEOUT ] && sleep $SLEEP
439 check_and_cleanup_lustre
440 [ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG || true