6 ALWAYS_EXCEPT="15c $REPLAY_DUAL_EXCEPT"
9 SETSTRIPE=${SETSTRIPE:-"$LFS setstripe"}
10 GETSTRIPE=${GETSTRIPE:-"$LFS getstripe"}
12 PTLDEBUG=${PTLDEBUG:--1}
13 LUSTRE=${LUSTRE:-$(cd $(dirname $0)/..; echo $PWD)}
15 CLEANUP=${CLEANUP:-""}
16 MOUNT_2=${MOUNT_2:-"yes"}
17 . $LUSTRE/tests/test-framework.sh
20 . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
23 remote_mds_nodsh && skip "remote MDS with nodsh" && exit 0
25 [ "$SLOW" = "no" ] && EXCEPT_SLOW="21b"
29 check_and_setup_lustre
30 MOUNTED=$(mounted_lustre_filesystems)
31 if ! $(echo $MOUNTED | grep -w -q $MOUNT2); then
32 zconf_mount $HOSTNAME $MOUNT2
37 rm -rf $DIR/[df][0-9]*
39 [ "$DAEMONFILE" ] && $LCTL debug_daemon start $DAEMONFILE $DAEMONSIZE
43 replay_barrier $SINGLEMDS
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"
59 replay_barrier $SINGLEMDS
63 checkstat $MOUNT2/adir || return 1
65 checkstat $MOUNT2/adir && return 2
68 run_test 2 "|X| mkdir adir"
71 replay_barrier $SINGLEMDS
73 mkdir $MOUNT2/adir/bdir
76 checkstat $MOUNT2/adir || return 1
77 checkstat $MOUNT1/adir/bdir || return 2
78 rmdir $MOUNT2/adir/bdir $MOUNT1/adir
79 checkstat $MOUNT1/adir && return 3
80 checkstat $MOUNT2/adir/bdir && return 4
83 run_test 3 "|X| mkdir adir, mkdir adir/bdir "
87 replay_barrier $SINGLEMDS
88 mkdir $MOUNT1/adir && return 1
89 mkdir $MOUNT2/adir/bdir
92 checkstat $MOUNT2/adir || return 2
93 checkstat $MOUNT1/adir/bdir || return 3
95 rmdir $MOUNT2/adir/bdir $MOUNT1/adir
96 checkstat $MOUNT1/adir && return 4
97 checkstat $MOUNT2/adir/bdir && return 5
100 run_test 4 "|X| mkdir adir (-EEXIST), mkdir adir/bdir "
104 # multiclient version of replay_single.sh/test_8
106 multiop_bg_pause $MOUNT2/a o_tSc || return 1
109 replay_barrier $SINGLEMDS
111 wait $pid || return 1
114 [ -e $MOUNT2/a ] && return 2
117 run_test 5 "open, unlink |X| close"
122 multiop_bg_pause $MOUNT2/a o_c || return 1
124 multiop_bg_pause $MOUNT1/a o_c || return 1
127 replay_barrier $SINGLEMDS
129 wait $pid1 || return 1
133 wait $pid2 || return 1
134 [ -e $MOUNT2/a ] && return 2
137 run_test 6 "open1, open2, unlink |X| close1 [fail $SINGLEMDS] close2"
140 replay_barrier $SINGLEMDS
141 drop_reint_reply "mcreate $MOUNT1/$tfile" || return 1
143 checkstat $MOUNT2/$tfile || return 2
144 rm $MOUNT1/$tfile || return 3
148 run_test 8 "replay of resent request"
151 replay_barrier $SINGLEMDS
152 mcreate $MOUNT1/$tfile-1
153 mcreate $MOUNT2/$tfile-2
154 # drop first reint reply
155 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000119
157 do_facet $SINGLEMDS lctl set_param fail_loc=0
159 rm $MOUNT1/$tfile-[1,2] || return 1
163 run_test 9 "resending a replayed create"
166 mcreate $MOUNT1/$tfile-1
167 replay_barrier $SINGLEMDS
168 munlink $MOUNT1/$tfile-1
169 mcreate $MOUNT2/$tfile-2
170 # drop first reint reply
171 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000119
173 do_facet $SINGLEMDS lctl set_param fail_loc=0
175 checkstat $MOUNT1/$tfile-1 && return 1
176 checkstat $MOUNT1/$tfile-2 || return 2
181 run_test 10 "resending a replayed unlink"
184 replay_barrier $SINGLEMDS
185 mcreate $MOUNT1/$tfile-1
186 mcreate $MOUNT2/$tfile-2
187 mcreate $MOUNT1/$tfile-3
188 mcreate $MOUNT2/$tfile-4
189 mcreate $MOUNT1/$tfile-5
190 # drop all reint replies for a while
191 do_facet $SINGLEMDS lctl set_param fail_loc=0x0119
192 # note that with this fail_loc set, facet_failover df will fail
193 facet_failover $SINGLEMDS
194 #sleep for while, let both clients reconnect and timeout
195 sleep $((TIMEOUT * 2))
196 do_facet $SINGLEMDS lctl set_param fail_loc=0
198 rm $MOUNT1/$tfile-[1-5] || return 1
202 run_test 11 "both clients timeout during replay"
205 replay_barrier $SINGLEMDS
207 multiop_bg_pause $DIR/$tfile mo_c || return 1
210 #define OBD_FAIL_LDLM_ENQUEUE 0x302
211 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000302
212 facet_failover $SINGLEMDS
213 do_facet $SINGLEMDS lctl set_param fail_loc=0
214 clients_up || return 1
217 kill -USR1 $MULTIPID || return 3
218 wait $MULTIPID || return 4
219 $CHECKSTAT -t file $DIR/$tfile || return 2
224 run_test 12 "open resend timeout"
227 multiop_bg_pause $DIR/$tfile mo_c || return 1
230 replay_barrier $SINGLEMDS
232 kill -USR1 $MULTIPID || return 3
233 wait $MULTIPID || return 4
236 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000115
237 facet_failover $SINGLEMDS
238 do_facet $SINGLEMDS lctl set_param fail_loc=0
239 clients_up || return 1
242 $CHECKSTAT -t file $DIR/$tfile || return 2
247 run_test 13 "close resend timeout"
249 # test 14a removed after 18143 because it shouldn't fail anymore and do the same
254 wait_destroy_complete
255 BEFOREUSED=`df -P $DIR | tail -1 | awk '{ print $3 }'`
256 mkdir -p $MOUNT1/$tdir
257 $SETSTRIPE -o 0 $MOUNT1/$tdir
258 replay_barrier $SINGLEMDS
259 createmany -o $MOUNT1/$tdir/$tfile- 5
261 $SETSTRIPE -o 0 $MOUNT2/f14b-3
262 echo "data" > $MOUNT2/f14b-3
263 createmany -o $MOUNT1/$tdir/$tfile-3- 5
267 wait_recovery_complete $SINGLEMDS || error "MDS recovery not done"
269 # first 25 files should have been replayed
270 unlinkmany $MOUNT1/$tdir/$tfile- 5 || return 2
271 unlinkmany $MOUNT1/$tdir/$tfile-3- 5 || return 3
273 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
275 wait_mds_ost_sync || return 4
276 wait_destroy_complete || return 5
278 AFTERUSED=`df -P $DIR | tail -1 | awk '{ print $3 }'`
279 log "before $BEFOREUSED, after $AFTERUSED"
280 [ $AFTERUSED -ne $BEFOREUSED ] && \
281 error "after $AFTERUSED > before $BEFOREUSED" && return 4
284 run_test 14b "delete ost orphans if gap occured in objids due to VBR"
286 test_15a() { # was test_15
287 replay_barrier $SINGLEMDS
288 createmany -o $MOUNT1/$tfile- 25
289 createmany -o $MOUNT2/$tfile-2- 1
294 unlinkmany $MOUNT1/$tfile- 25 || return 2
295 [ -e $MOUNT1/$tfile-2-0 ] && error "$tfile-2-0 exists"
297 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
300 run_test 15a "timeout waiting for lost client during replay, 1 client completes"
303 replay_barrier $SINGLEMDS
304 for ((i = 0; i < 2000; i++)); do
305 echo "data" > "$MOUNT2/${tfile}-$i" || error "create ${tfile}-$i failed"
311 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
314 run_test 15c "remove multiple OST orphans"
317 replay_barrier $SINGLEMDS
318 createmany -o $MOUNT1/$tfile- 25
319 createmany -o $MOUNT2/$tfile-2- 1
322 facet_failover $SINGLEMDS
326 unlinkmany $MOUNT1/$tfile- 25 || return 2
328 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
332 run_test 16 "fail MDS during recovery (3571)"
335 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
337 createmany -o $MOUNT1/$tfile- 25
338 createmany -o $MOUNT2/$tfile-2- 1
340 # Make sure the disconnect is lost
348 unlinkmany $MOUNT1/$tfile- 25 || return 2
350 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
354 run_test 17 "fail OST during recovery (3571)"
356 # cleanup with blocked enqueue fails until timer elapses (MDS busy), wait for it
359 test_18() { # bug 3822 - evicting client with enqueued lock
361 mkdir -p $MOUNT1/$tdir
362 touch $MOUNT1/$tdir/f0
363 #define OBD_FAIL_LDLM_ENQUEUE_BLOCKED 0x30b
364 statmany -s $MOUNT1/$tdir/f 1 500 &
367 do_facet $SINGLEMDS lctl set_param fail_loc=0x8000030b # hold enqueue
369 #define OBD_FAIL_LDLM_BL_CALLBACK 0x305
370 do_facet client lctl set_param fail_loc=0x80000305 # drop cb, evict
372 usleep 500 # wait to ensure first client is one that will be evicted
373 openfile -f O_RDONLY $MOUNT2/$tdir/f0
375 dmesg | grep "entering recovery in server" && \
376 error "client not evicted" || true
377 do_facet client "lctl set_param fail_loc=0"
378 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
380 run_test 18 "ldlm_handle_enqueue succeeds on evicted export (3822)"
382 test_19() { # Bug 10991 - resend of open request does not fail assertion.
383 replay_barrier $SINGLEMDS
384 drop_ldlm_reply "createmany -o $DIR/$tfile 1" || return 1
386 checkstat $DIR2/${tfile}0 || return 2
387 rm $DIR/${tfile}0 || return 3
391 run_test 19 "resend of open request"
395 replay_barrier $SINGLEMDS
401 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
402 TIER1=$((`date +%s` - BEFORE))
404 replay_barrier $SINGLEMDS
410 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
411 TIER2=$((`date +%s` - BEFORE))
412 [ $TIER2 -ge $((TIER1 * 2)) ] && \
413 error "recovery time is growing $TIER2 > $TIER1"
416 run_test 20 "recovery time is not increasing"
418 # commit on sharing tests
420 local param_file=$TMP/$tfile-params
422 save_lustre_params $(facet_active_host $SINGLEMDS) "mdt.*.commit_on_sharing" > $param_file
423 do_facet $SINGLEMDS lctl set_param mdt.*.commit_on_sharing=1
424 touch $MOUNT1/$tfile-1
425 mv $MOUNT2/$tfile-1 $MOUNT2/$tfile-2
426 mv $MOUNT1/$tfile-2 $MOUNT1/$tfile-3
427 replay_barrier_nosync $SINGLEMDS
430 facet_failover $SINGLEMDS
432 # all renames are replayed
433 unlink $MOUNT1/$tfile-3 || return 2
435 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
437 do_facet $SINGLEMDS lctl set_param mdt.*.commit_on_sharing=0
438 rm -rf $MOUNT1/$tfile-*
439 restore_lustre_params < $param_file
443 run_test 21a "commit on sharing"
447 do_node $CLIENT1 rm -f $MOUNT1/$tfile-*
450 do_node $CLIENT1 touch $MOUNT1/$tfile-1
451 do_node $CLIENT2 mv $MOUNT1/$tfile-1 $MOUNT1/$tfile-2
452 do_node $CLIENT1 mv $MOUNT1/$tfile-2 $MOUNT1/$tfile-3
454 replay_barrier_nosync $mds
455 shutdown_client $CLIENT2 $MOUNT1
459 # were renames replayed?
461 echo UNLINK $MOUNT1/$tfile-3
462 do_node $CLIENT1 unlink $MOUNT1/$tfile-3 || \
463 { echo "unlink $tfile-3 fail!" && rc=1; }
466 zconf_mount_clients $CLIENT2 $MOUNT1 || error "mount $CLIENT2 $MOUNT1 fail"
472 [ -z "$CLIENTS" ] && skip "Need two or more clients." && return
473 [ $CLIENTCOUNT -lt 2 ] && \
474 { skip "Need two or more clients, have $CLIENTCOUNT" && return; }
476 if [ "$FAILURE_MODE" = "HARD" ] && mixed_mdt_devs; then
477 skip "Several mdt services on one mds node are used with FAILURE_MODE=$FAILURE_MODE. "
482 zconf_umount_clients $CLIENTS $MOUNT2
483 zconf_mount_clients $CLIENTS $MOUNT1
485 local param_file=$TMP/$tfile-params
487 local num=$(get_mds_dir $MOUNT1)
489 save_lustre_params $(facet_active_host mds$num) "mdt.*.commit_on_sharing" > $param_file
493 do_facet mds$num lctl set_param mdt.*.commit_on_sharing=$COS
495 test_21b_sub mds$num || error "Not all renames are replayed. COS=$COS"
497 # COS disabled (should fail)
499 do_facet mds$num lctl set_param mdt.*.commit_on_sharing=$COS
501 # there is still a window when transactions may be written to disk before
502 # the mds device is set R/O. To avoid such a rare test failure, the check
503 # is repeated several times.
506 test_21b_sub mds$num || break;
507 let n_attempts=n_attempts+1
508 [ $n_attempts -gt 3 ] &&
509 error "The test cannot check whether COS works or not: all renames are replied w/o COS"
511 restore_lustre_params < $param_file
515 run_test 21b "commit on sharing, two clients"
517 # end commit on sharing tests
519 equals_msg `basename $0`: test complete, cleaning up
520 SLEEP=$((`date +%s` - $NOW))
521 [ $SLEEP -lt $TIMEOUT ] && sleep $SLEEP
522 [ "$MOUNTED2" = yes ] && zconf_umount $HOSTNAME $MOUNT2 || true
523 check_and_cleanup_lustre
524 [ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG && grep -q FAIL $TESTSUITELOG && exit 1 || true