6 ALWAYS_EXCEPT="15c $REPLAY_DUAL_EXCEPT"
9 PTLDEBUG=${PTLDEBUG:--1}
10 LUSTRE=${LUSTRE:-$(cd $(dirname $0)/..; echo $PWD)}
12 CLEANUP=${CLEANUP:-""}
13 MOUNT_2=${MOUNT_2:-"yes"}
14 . $LUSTRE/tests/test-framework.sh
16 if [ "$FAILURE_MODE" = "HARD" ] && mixed_ost_devs; then
17 CONFIG_EXCEPTIONS="17"
18 echo -n "Several ost services on one ost node are used with FAILURE_MODE=$FAILURE_MODE. "
19 echo "Except the tests: $CONFIG_EXCEPTIONS"
20 ALWAYS_EXCEPT="$ALWAYS_EXCEPT $CONFIG_EXCEPTIONS"
25 . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
27 remote_mds_nodsh && skip "remote MDS with nodsh" && exit 0
29 [ "$SLOW" = "no" ] && EXCEPT_SLOW="1 2 3 4 5 14"
33 check_and_setup_lustre
34 MOUNTED=$(mounted_lustre_filesystems)
35 if ! $(echo $MOUNTED | grep -w -q $MOUNT2); then
36 zconf_mount $HOSTNAME $MOUNT2
41 rm -rf $DIR/[df][0-9]*
43 [ "$DAEMONFILE" ] && $LCTL debug_daemon start $DAEMONFILE $DAEMONSIZE
51 checkstat $MOUNT2/a || return 1
52 checkstat $MOUNT1/b || return 2
53 rm $MOUNT2/a $MOUNT1/b
54 checkstat $MOUNT1/a && return 3
55 checkstat $MOUNT2/b && return 4
59 run_test 1 "|X| simple create"
67 checkstat $MOUNT2/adir || return 1
69 checkstat $MOUNT2/adir && return 2
72 run_test 2 "|X| mkdir adir"
77 mkdir $MOUNT2/adir/bdir
80 checkstat $MOUNT2/adir || return 1
81 checkstat $MOUNT1/adir/bdir || return 2
82 rmdir $MOUNT2/adir/bdir $MOUNT1/adir
83 checkstat $MOUNT1/adir && return 3
84 checkstat $MOUNT2/adir/bdir && return 4
87 run_test 3 "|X| mkdir adir, mkdir adir/bdir "
92 mkdir $MOUNT1/adir && return 1
93 mkdir $MOUNT2/adir/bdir
96 checkstat $MOUNT2/adir || return 2
97 checkstat $MOUNT1/adir/bdir || return 3
99 rmdir $MOUNT2/adir/bdir $MOUNT1/adir
100 checkstat $MOUNT1/adir && return 4
101 checkstat $MOUNT2/adir/bdir && return 5
104 run_test 4 "|X| mkdir adir (-EEXIST), mkdir adir/bdir "
108 # multiclient version of replay_single.sh/test_8
110 multiop_bg_pause $MOUNT2/a o_tSc || return 1
115 wait $pid || return 1
118 [ -e $MOUNT2/a ] && return 2
121 run_test 5 "open, unlink |X| close"
126 multiop_bg_pause $MOUNT2/a o_c || return 1
128 multiop_bg_pause $MOUNT1/a o_c || return 1
133 wait $pid1 || return 1
137 wait $pid2 || return 1
138 [ -e $MOUNT2/a ] && return 2
141 run_test 6 "open1, open2, unlink |X| close1 [fail mds] close2"
145 drop_reint_reply "mcreate $MOUNT1/$tfile" || return 1
147 checkstat $MOUNT2/$tfile || return 2
148 rm $MOUNT1/$tfile || return 3
152 run_test 8 "replay of resent request"
156 mcreate $MOUNT1/$tfile-1
157 mcreate $MOUNT2/$tfile-2
158 # drop first reint reply
159 do_facet mds lctl set_param fail_loc=0x80000119
161 do_facet mds lctl set_param fail_loc=0
163 rm $MOUNT1/$tfile-[1,2] || return 1
167 run_test 9 "resending a replayed create"
170 mcreate $MOUNT1/$tfile-1
172 munlink $MOUNT1/$tfile-1
173 mcreate $MOUNT2/$tfile-2
174 # drop first reint reply
175 do_facet mds lctl set_param fail_loc=0x80000119
177 do_facet mds lctl set_param fail_loc=0
179 checkstat $MOUNT1/$tfile-1 && return 1
180 checkstat $MOUNT1/$tfile-2 || return 2
185 run_test 10 "resending a replayed unlink"
189 mcreate $MOUNT1/$tfile-1
190 mcreate $MOUNT2/$tfile-2
191 mcreate $MOUNT1/$tfile-3
192 mcreate $MOUNT2/$tfile-4
193 mcreate $MOUNT1/$tfile-5
194 # drop all reint replies for a while
195 do_facet mds lctl set_param fail_loc=0x0119
196 # note that with this fail_loc set, facet_failover df will fail
198 #sleep for while, let both clients reconnect and timeout
199 sleep $((TIMEOUT * 2))
200 do_facet mds lctl set_param fail_loc=0
202 while [ -z "$(ls $MOUNT1/$tfile-[1-5] 2>/dev/null)" ]; do
206 ls $MOUNT1/$tfile-[1-5]
207 rm $MOUNT1/$tfile-[1-5] || return 1
211 run_test 11 "both clients timeout during replay"
216 multiop_bg_pause $DIR/$tfile mo_c || return 1
219 #define OBD_FAIL_LDLM_ENQUEUE 0x302
220 do_facet mds lctl set_param fail_loc=0x80000302
222 do_facet mds lctl set_param fail_loc=0
223 df $MOUNT || { kill -USR1 $MULTIPID && return 1; }
226 kill -USR1 $MULTIPID || return 3
227 wait $MULTIPID || return 4
228 $CHECKSTAT -t file $DIR/$tfile || return 2
233 run_test 12 "open resend timeout"
236 multiop_bg_pause $DIR/$tfile mo_c || return 1
241 kill -USR1 $MULTIPID || return 3
242 wait $MULTIPID || return 4
245 do_facet mds lctl set_param fail_loc=0x80000115
247 do_facet mds lctl set_param fail_loc=0
248 df $MOUNT || return 1
251 $CHECKSTAT -t file $DIR/$tfile || return 2
256 run_test 13 "close resend timeout"
260 createmany -o $MOUNT1/$tfile- 25
261 createmany -o $MOUNT2/$tfile-2- 1
262 createmany -o $MOUNT1/$tfile-3- 25
266 # expect recovery to fail due to missing client 2
267 df $MOUNT1 && return 1
270 # first 25 files should have been replayed
271 unlinkmany $MOUNT1/$tfile- 25 || return 2
273 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
276 run_test 14a "timeouts waiting for lost client during replay"
279 BEFOREUSED=`df -P $DIR | tail -1 | awk '{ print $3 }'`
280 #lfs setstripe --index=0 --count=1 $MOUNT1
281 mkdir -p $MOUNT1/$tdir
282 #lfs setstripe --index=0 --count=1 $MOUNT1/$tdir
284 createmany -o $MOUNT1/$tfile- 5
285 echo "data" > $MOUNT2/$tdir/$tfile-2
286 createmany -o $MOUNT1/$tfile-3- 5
290 # expect recovery don't fail due to VBR
291 df $MOUNT1 || return 1
293 # first 25 files should have been replayed
294 unlinkmany $MOUNT1/$tfile- 5 || return 2
295 unlinkmany $MOUNT1/$tfile-3- 5 || return 3
297 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
298 # give ost time to process llogs
300 AFTERUSED=`df -P $DIR | tail -1 | awk '{ print $3 }'`
301 log "before $BEFOREUSED, after $AFTERUSED"
302 [ $AFTERUSED -ne $BEFOREUSED ] && \
303 error "after $AFTERUSED > before $BEFOREUSED" && return 4
306 run_test 14b "delete ost orphans if gap occured in objids due to VBR"
308 test_15a() { # was test_15
310 createmany -o $MOUNT1/$tfile- 25
311 createmany -o $MOUNT2/$tfile-2- 1
315 df $MOUNT || return 1
317 unlinkmany $MOUNT1/$tfile- 25 || return 2
318 [ -e $MOUNT1/$tfile-2-0 ] && error "$tfile-2-0 exists"
320 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
323 run_test 15a "timeout waiting for lost client during replay, 1 client completes"
327 for ((i = 0; i < 2000; i++)); do
328 echo "data" > "$MOUNT2/${tfile}-$i" || error "create ${tfile}-$i failed"
334 df $MOUNT || return 1
336 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
339 run_test 15c "remove multiple OST orphans"
343 createmany -o $MOUNT1/$tfile- 25
344 createmany -o $MOUNT2/$tfile-2- 1
350 df $MOUNT || return 1
352 unlinkmany $MOUNT1/$tfile- 25 || return 2
354 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
358 run_test 16 "fail MDS during recovery (3571)"
361 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
363 createmany -o $MOUNT1/$tfile- 25
364 createmany -o $MOUNT2/$tfile-2- 1
366 # Make sure the disconnect is lost
373 df $MOUNT || return 1
375 unlinkmany $MOUNT1/$tfile- 25 || return 2
377 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
381 run_test 17 "fail OST during recovery (3571)"
383 # cleanup with blocked enqueue fails until timer elapses (MDS busy), wait for it
386 test_18() { # bug 3822 - evicting client with enqueued lock
388 mkdir -p $MOUNT1/$tdir
389 touch $MOUNT1/$tdir/f0
390 #define OBD_FAIL_LDLM_ENQUEUE_BLOCKED 0x30b
391 statmany -s $MOUNT1/$tdir/f 1 500 &
394 do_facet mds lctl set_param fail_loc=0x8000030b # hold enqueue
396 #define OBD_FAIL_LDLM_BL_CALLBACK 0x305
397 do_facet client lctl set_param fail_loc=0x80000305 # drop cb, evict
399 sleep 0.500s # wait to ensure first client is one that will be evicted
400 openfile -f O_RDONLY $MOUNT2/$tdir/f0
402 dmesg | grep "entering recovery in server" && \
403 error "client not evicted" || true
405 run_test 18 "ldlm_handle_enqueue succeeds on evicted export (3822)"
407 test_19() { # Bug 10991 - resend of open request does not fail assertion.
409 drop_ldlm_reply "createmany -o $DIR/$tfile 1" || return 1
411 checkstat $DIR2/${tfile}0 || return 2
412 rm $DIR/${tfile}0 || return 3
416 run_test 19 "resend of open request"
425 df $MOUNT1 || return 1
427 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
428 TIER1=$((`date +%s` - BEFORE))
435 df $MOUNT1 || return 1
437 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
438 TIER2=$((`date +%s` - BEFORE))
439 [ $TIER2 -ge $((TIER1 * 2)) ] && \
440 error "recovery time is growing $TIER2 > $TIER1"
443 run_test 20 "recovery time is not increasing"
445 equals_msg `basename $0`: test complete, cleaning up
446 SLEEP=$((`date +%s` - $NOW))
447 [ $SLEEP -lt $TIMEOUT ] && sleep $SLEEP
448 [ "$MOUNTED2" = yes ] && zconf_umount $HOSTNAME $MOUNT2 || true
449 check_and_cleanup_lustre
450 [ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG && grep -q FAIL $TESTSUITELOG && exit 1 || true