7 # This test needs to be run on the client
10 LUSTRE=${LUSTRE:-$(cd $(dirname $0)/..; echo $PWD)}
13 . $LUSTRE/tests/test-framework.sh
15 . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
16 CHECK_GRANT=${CHECK_GRANT:-"yes"}
17 GRANT_CHECK_LIST=${GRANT_CHECK_LIST:-""}
22 ALWAYS_EXCEPT="$REPLAY_SINGLE_EXCEPT"
24 # 63 min 7 min AT AT AT AT"
25 [ "$SLOW" = "no" ] && EXCEPT_SLOW="1 2 3 4 6 12 16 44a 44b 65 66 67 68"
29 cleanup_and_setup_lustre
34 rm -rf $DIR/[df][0-9]*
36 test_0a() { # was test_0
39 replay_barrier $SINGLEMDS
43 run_test 0a "empty replay"
46 # this test attempts to trigger a race in the precreation code,
47 # and must run before any other objects are created on the filesystem
49 createmany -o $DIR/$tfile 20 || return 1
50 unlinkmany $DIR/$tfile 20 || return 2
52 run_test 0b "ensure object created after recover exists. (3284)"
58 lctl set_param -n seq.cli-srv-$mds-mdc-*.width=$width
64 lctl get_param -n seq.cli-srv-$mds-mdc-*.width
67 # This test should pass for single-mds and multi-mds configs.
68 # But for different configurations it tests different things.
72 # (1) fld_create replay should happen;
74 # (2) fld_create replay should not return -EEXISTS, if it does
75 # this means sequence manager recovery code is buggy and allocated
76 # same sequence two times after recovery.
80 # (1) fld_create replay may not happen, because its home MDS is
81 # MDS2 which is not involved to revovery;
83 # (2) as fld_create does not happen on MDS1, it does not make any
86 local label=`mdsdevlabel 1`
87 [ -z "$label" ] && echo "No label for mds1" && return 1
89 replay_barrier $SINGLEMDS
90 local sw=`seq_get_width $label`
92 # make seq manager switch to next sequence each
93 # time as new fid is needed.
94 seq_set_width $label 1
96 # make sure that fld has created at least one new
98 touch $DIR/$tfile || return 2
99 seq_set_width $label $sw
101 # fail $SINGLEMDS and start recovery, replay RPCs, etc.
104 # wait for recovery finish
108 # flush fld cache and dentry cache to make it lookup
109 # created entry instead of revalidating existent one
111 zconf_mount `hostname` $MOUNT
113 # issue lookup which should call fld lookup which
114 # should fail if client did not replay fld create
115 # correctly and server has no fld entry
116 touch $DIR/$tfile || return 3
117 rm $DIR/$tfile || return 4
119 run_test 0c "fld create"
122 replay_barrier $SINGLEMDS
125 $CHECKSTAT -t file $DIR/$tfile || return 1
128 run_test 1 "simple create"
131 replay_barrier $SINGLEMDS
134 $CHECKSTAT -t file $DIR/$tfile || return 1
141 replay_barrier $SINGLEMDS
144 $CHECKSTAT -t file $DIR/$tfile || return 1
150 replay_barrier $SINGLEMDS
152 o_directory $DIR/$tfile
154 $CHECKSTAT -t file $DIR/$tfile || return 2
157 run_test 3a "replay failed open(O_DIRECTORY)"
160 replay_barrier $SINGLEMDS
161 #define OBD_FAIL_MDS_OPEN_PACK | OBD_FAIL_ONCE
162 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000114"
164 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
166 $CHECKSTAT -t file $DIR/$tfile && return 2
169 run_test 3b "replay failed open -ENOMEM"
172 replay_barrier $SINGLEMDS
173 #define OBD_FAIL_MDS_ALLOC_OBDO | OBD_FAIL_ONCE
174 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000128"
176 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
179 $CHECKSTAT -t file $DIR/$tfile && return 2
182 run_test 3c "replay failed open -ENOMEM"
184 test_4a() { # was test_4
185 replay_barrier $SINGLEMDS
186 for i in `seq 10`; do
187 echo "tag-$i" > $DIR/$tfile-$i
190 for i in `seq 10`; do
191 grep -q "tag-$i" $DIR/$tfile-$i || error "$tfile-$i"
194 run_test 4a "|x| 10 open(O_CREAT)s"
197 replay_barrier $SINGLEMDS
200 $CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
202 run_test 4b "|x| rm 10 files"
204 # The idea is to get past the first block of precreated files on both
205 # osts, and then replay.
207 replay_barrier $SINGLEMDS
208 for i in `seq 220`; do
209 echo "tag-$i" > $DIR/$tfile-$i
212 for i in `seq 220`; do
213 grep -q "tag-$i" $DIR/$tfile-$i || error "f1c-$i"
217 # waiting for commitment of removal
219 run_test 5 "|x| 220 open(O_CREAT)"
222 test_6a() { # was test_6
224 replay_barrier $SINGLEMDS
225 mcreate $DIR/$tdir/$tfile
227 $CHECKSTAT -t dir $DIR/$tdir || return 1
228 $CHECKSTAT -t file $DIR/$tdir/$tfile || return 2
230 # waiting for log process thread
232 run_test 6a "mkdir + contained create"
236 replay_barrier $SINGLEMDS
239 $CHECKSTAT -t dir $DIR/$tdir && return 1 || true
241 run_test 6b "|X| rmdir"
245 replay_barrier $SINGLEMDS
246 mcreate $DIR/$tdir/$tfile
248 $CHECKSTAT -t dir $DIR/$tdir || return 1
249 $CHECKSTAT -t file $DIR/$tdir/$tfile || return 2
252 run_test 7 "mkdir |X| contained create"
255 # make sure no side-effect from previous test.
257 replay_barrier $SINGLEMDS
258 multiop_bg_pause $DIR/$tfile mo_c || return 4
262 $CHECKSTAT -t file $DIR/$tfile || return 1
263 kill -USR1 $MULTIPID || return 2
264 wait $MULTIPID || return 3
267 run_test 8 "creat open |X| close"
270 replay_barrier $SINGLEMDS
272 local old_inum=`ls -i $DIR/$tfile | awk '{print $1}'`
274 local new_inum=`ls -i $DIR/$tfile | awk '{print $1}'`
276 echo " old_inum == $old_inum, new_inum == $new_inum"
277 if [ $old_inum -eq $new_inum ] ;
279 echo " old_inum and new_inum match"
281 echo "!!!! old_inum and new_inum NOT match"
286 run_test 9 "|X| create (same inum/gen)"
290 replay_barrier $SINGLEMDS
291 mv $DIR/$tfile $DIR/$tfile-2
294 $CHECKSTAT $DIR/$tfile && return 1
295 $CHECKSTAT $DIR/$tfile-2 ||return 2
299 run_test 10 "create |X| rename unlink"
303 echo "old" > $DIR/$tfile
304 mv $DIR/$tfile $DIR/$tfile-2
305 replay_barrier $SINGLEMDS
306 echo "new" > $DIR/$tfile
308 grep old $DIR/$tfile-2
310 grep new $DIR/$tfile || return 1
311 grep old $DIR/$tfile-2 || return 2
313 run_test 11 "create open write rename |X| create-old-name read"
317 multiop_bg_pause $DIR/$tfile o_tSc || return 3
320 replay_barrier $SINGLEMDS
322 wait $pid || return 1
325 [ -e $DIR/$tfile ] && return 2
328 run_test 12 "open, unlink |X| close"
331 # 1777 - replay open after committed chmod that would make
332 # a regular open a failure
335 multiop_bg_pause $DIR/$tfile O_wc || return 3
338 $CHECKSTAT -p 0 $DIR/$tfile
339 replay_barrier $SINGLEMDS
342 wait $pid || return 1
344 $CHECKSTAT -s 1 -p 0 $DIR/$tfile || return 2
347 run_test 13 "open chmod 0 |x| write close"
350 multiop_bg_pause $DIR/$tfile O_tSc || return 4
353 replay_barrier $SINGLEMDS
354 kill -USR1 $pid || return 1
355 wait $pid || return 2
358 [ -e $DIR/$tfile ] && return 3
361 run_test 14 "open(O_CREAT), unlink |X| close"
364 multiop_bg_pause $DIR/$tfile O_tSc || return 5
367 replay_barrier $SINGLEMDS
368 touch $DIR/g11 || return 1
370 wait $pid || return 2
373 [ -e $DIR/$tfile ] && return 3
374 touch $DIR/h11 || return 4
377 run_test 15 "open(O_CREAT), unlink |X| touch new, close"
381 replay_barrier $SINGLEMDS
384 mcreate $DIR/$tfile-2
386 [ -e $DIR/$tfile ] && return 1
387 [ -e $DIR/$tfile-2 ] || return 2
388 munlink $DIR/$tfile-2 || return 3
390 run_test 16 "|X| open(O_CREAT), unlink, touch new, unlink new"
393 replay_barrier $SINGLEMDS
394 multiop_bg_pause $DIR/$tfile O_c || return 4
397 kill -USR1 $pid || return 1
398 wait $pid || return 2
399 $CHECKSTAT -t file $DIR/$tfile || return 3
402 run_test 17 "|X| open(O_CREAT), |replay| close"
405 replay_barrier $SINGLEMDS
406 multiop_bg_pause $DIR/$tfile O_tSc || return 8
409 touch $DIR/$tfile-2 || return 1
410 echo "pid: $pid will close"
412 wait $pid || return 2
415 [ -e $DIR/$tfile ] && return 3
416 [ -e $DIR/$tfile-2 ] || return 4
417 # this touch frequently fails
418 touch $DIR/$tfile-3 || return 5
419 munlink $DIR/$tfile-2 || return 6
420 munlink $DIR/$tfile-3 || return 7
423 run_test 18 "|X| open(O_CREAT), unlink, touch new, close, touch, unlink"
425 # bug 1855 (a simpler form of test_11 above)
427 replay_barrier $SINGLEMDS
429 echo "old" > $DIR/$tfile
430 mv $DIR/$tfile $DIR/$tfile-2
431 grep old $DIR/$tfile-2
433 grep old $DIR/$tfile-2 || return 2
435 run_test 19 "|X| mcreate, open, write, rename "
437 test_20a() { # was test_20
438 replay_barrier $SINGLEMDS
439 multiop_bg_pause $DIR/$tfile O_tSc || return 3
445 wait $pid || return 1
446 [ -e $DIR/$tfile ] && return 2
449 run_test 20a "|X| open(O_CREAT), unlink, replay, close (test mds_cleanup_orphans)"
451 test_20b() { # bug 10480
452 BEFOREUSED=`df -P $DIR | tail -1 | awk '{ print $3 }'`
454 dd if=/dev/zero of=$DIR/$tfile bs=4k count=10000 &
456 while [ ! -e $DIR/$tfile ] ; do
457 usleep 60 # give dd a chance to start
460 lfs getstripe $DIR/$tfile || return 1
461 rm -f $DIR/$tfile || return 2 # make it an orphan
463 df -P $DIR || df -P $DIR || true # reconnect
465 fail $SINGLEMDS # start orphan recovery
466 df -P $DIR || df -P $DIR || true # reconnect
467 wait_mds_recovery_done || error "MDS recovery not done"
469 # FIXME just because recovery is done doesn't mean we've finished
470 # orphan cleanup. Fake it with a sleep for now...
472 AFTERUSED=`df -P $DIR | tail -1 | awk '{ print $3 }'`
473 log "before $BEFOREUSED, after $AFTERUSED"
474 [ $AFTERUSED -gt $((BEFOREUSED + 20)) ] && \
475 error "after $AFTERUSED > before $BEFOREUSED"
478 run_test 20b "write, unlink, eviction, replay, (test mds_cleanup_orphans)"
480 test_20c() { # bug 10480
481 multiop_bg_pause $DIR/$tfile Ow_c || return 1
488 df -P $DIR || df -P $DIR || true # reconnect
491 test -s $DIR/$tfile || error "File was truncated"
495 run_test 20c "check that client eviction does not affect file content"
498 replay_barrier $SINGLEMDS
499 multiop_bg_pause $DIR/$tfile O_tSc || return 5
502 touch $DIR/g11 || return 1
506 wait $pid || return 2
507 [ -e $DIR/$tfile ] && return 3
508 touch $DIR/h11 || return 4
511 run_test 21 "|X| open(O_CREAT), unlink touch new, replay, close (test mds_cleanup_orphans)"
514 multiop_bg_pause $DIR/$tfile O_tSc || return 3
517 replay_barrier $SINGLEMDS
522 wait $pid || return 1
523 [ -e $DIR/$tfile ] && return 2
526 run_test 22 "open(O_CREAT), |X| unlink, replay, close (test mds_cleanup_orphans)"
529 multiop_bg_pause $DIR/$tfile O_tSc || return 5
532 replay_barrier $SINGLEMDS
534 touch $DIR/g11 || return 1
538 wait $pid || return 2
539 [ -e $DIR/$tfile ] && return 3
540 touch $DIR/h11 || return 4
543 run_test 23 "open(O_CREAT), |X| unlink touch new, replay, close (test mds_cleanup_orphans)"
546 multiop_bg_pause $DIR/$tfile O_tSc || return 3
549 replay_barrier $SINGLEMDS
553 wait $pid || return 1
554 [ -e $DIR/$tfile ] && return 2
557 run_test 24 "open(O_CREAT), replay, unlink, close (test mds_cleanup_orphans)"
560 multiop_bg_pause $DIR/$tfile O_tSc || return 3
564 replay_barrier $SINGLEMDS
567 wait $pid || return 1
568 [ -e $DIR/$tfile ] && return 2
571 run_test 25 "open(O_CREAT), unlink, replay, close (test mds_cleanup_orphans)"
574 replay_barrier $SINGLEMDS
575 multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
577 multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
582 wait $pid2 || return 1
586 wait $pid1 || return 2
587 [ -e $DIR/$tfile-1 ] && return 3
588 [ -e $DIR/$tfile-2 ] && return 4
591 run_test 26 "|X| open(O_CREAT), unlink two, close one, replay, close one (test mds_cleanup_orphans)"
594 replay_barrier $SINGLEMDS
595 multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
597 multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
604 wait $pid1 || return 1
606 wait $pid2 || return 2
607 [ -e $DIR/$tfile-1 ] && return 3
608 [ -e $DIR/$tfile-2 ] && return 4
611 run_test 27 "|X| open(O_CREAT), unlink two, replay, close two (test mds_cleanup_orphans)"
614 multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
616 multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
618 replay_barrier $SINGLEMDS
622 wait $pid2 || return 1
626 wait $pid1 || return 2
627 [ -e $DIR/$tfile-1 ] && return 3
628 [ -e $DIR/$tfile-2 ] && return 4
631 run_test 28 "open(O_CREAT), |X| unlink two, close one, replay, close one (test mds_cleanup_orphans)"
634 multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
636 multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
638 replay_barrier $SINGLEMDS
644 wait $pid1 || return 1
646 wait $pid2 || return 2
647 [ -e $DIR/$tfile-1 ] && return 3
648 [ -e $DIR/$tfile-2 ] && return 4
651 run_test 29 "open(O_CREAT), |X| unlink two, replay, close two (test mds_cleanup_orphans)"
654 multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
656 multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
661 replay_barrier $SINGLEMDS
664 wait $pid1 || return 1
666 wait $pid2 || return 2
667 [ -e $DIR/$tfile-1 ] && return 3
668 [ -e $DIR/$tfile-2 ] && return 4
671 run_test 30 "open(O_CREAT) two, unlink two, replay, close two (test mds_cleanup_orphans)"
674 multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
676 multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
680 replay_barrier $SINGLEMDS
684 wait $pid1 || return 1
686 wait $pid2 || return 2
687 [ -e $DIR/$tfile-1 ] && return 3
688 [ -e $DIR/$tfile-2 ] && return 4
691 run_test 31 "open(O_CREAT) two, unlink one, |X| unlink one, close two (test mds_cleanup_orphans)"
693 # tests for bug 2104; completion without crashing is success. The close is
694 # stale, but we always return 0 for close, so the app never sees it.
696 multiop_bg_pause $DIR/$tfile O_c || return 2
698 multiop_bg_pause $DIR/$tfile O_c || return 3
701 df $MOUNT || sleep 1 && df $MOUNT || return 1
707 run_test 32 "close() notices client eviction; close() after client eviction"
709 # Abort recovery before client complete
710 test_33a() { # was test_33
711 replay_barrier $SINGLEMDS
712 createmany -o $DIR/$tfile-%d 100
713 fail_abort $SINGLEMDS
714 # this file should be gone, because the replay was aborted
715 $CHECKSTAT -t file $DIR/$tfile-* && return 3
716 unlinkmany $DIR/$tfile-%d 0 100
719 run_test 33a "abort recovery before client does replay"
722 test_33b() { # was test_33a
723 replay_barrier $SINGLEMDS
724 createmany -o $DIR/$tfile-%d 10
725 fail_abort $SINGLEMDS
726 unlinkmany $DIR/$tfile-%d 0 10
727 # recreate shouldn't fail
728 createmany -o $DIR/$tfile-%d 10 || return 3
729 unlinkmany $DIR/$tfile-%d 0 10
732 run_test 33b "fid shouldn't be reused after abort recovery"
735 multiop_bg_pause $DIR/$tfile O_c || return 2
739 replay_barrier $SINGLEMDS
740 fail_abort $SINGLEMDS
742 [ -e $DIR/$tfile ] && return 1
746 run_test 34 "abort recovery before client does replay (test mds_cleanup_orphans)"
748 # bug 2278 - generate one orphan on OST, then destroy it during recovery from llog
752 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
753 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000119"
758 # give a chance to remove from MDS
759 fail_abort $SINGLEMDS
760 $CHECKSTAT -t file $DIR/$tfile && return 1 || true
762 run_test 35 "test recovery from llog for unlink op"
764 # b=2432 resent cancel after replay uses wrong cookie,
765 # so don't resend cancels
767 replay_barrier $SINGLEMDS
769 checkstat $DIR/$tfile
770 facet_failover $SINGLEMDS
772 if dmesg | grep "unknown lock cookie"; then
773 echo "cancel after replay failed"
777 run_test 36 "don't resend cancel"
780 # directory orphans can't be unlinked from PENDING directory
782 rmdir $DIR/$tfile 2>/dev/null
783 multiop_bg_pause $DIR/$tfile dD_c || return 2
787 replay_barrier $SINGLEMDS
788 # clear the dmesg buffer so we only see errors from this recovery
790 fail_abort $SINGLEMDS
792 dmesg | grep "mds_unlink_orphan.*error .* unlinking orphan" && return 1
796 run_test 37 "abort recovery before client does replay (test mds_cleanup_orphans for directories)"
799 createmany -o $DIR/$tfile-%d 800
800 unlinkmany $DIR/$tfile-%d 0 400
801 replay_barrier $SINGLEMDS
803 unlinkmany $DIR/$tfile-%d 400 400
805 $CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
807 run_test 38 "test recovery from unlink llog (test llog_gen_rec) "
809 test_39() { # bug 4176
810 createmany -o $DIR/$tfile-%d 800
811 replay_barrier $SINGLEMDS
812 unlinkmany $DIR/$tfile-%d 0 400
814 unlinkmany $DIR/$tfile-%d 400 400
816 $CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
818 run_test 39 "test recovery from unlink llog (test llog_gen_rec) "
821 lctl get_param -n osc.*.stats | awk -vwrites=0 '/ost_write/ { writes += $2 } END { print writes; }'
826 $LCTL mark multiop $MOUNT/$tfile OS_c
827 multiop $MOUNT/$tfile OS_c &
829 writeme -s $MOUNT/${tfile}-2 &
832 facet_failover $SINGLEMDS
833 #define OBD_FAIL_MDS_CONNECT_NET 0x117
834 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000117"
836 stat1=`count_ost_writes`
838 stat2=`count_ost_writes`
839 echo "$stat1, $stat2"
840 if [ $stat1 -lt $stat2 ]; then
841 echo "writes continuing during recovery"
844 echo "writes not continuing during recovery, bug 2477"
847 echo "waiting for writeme $WRITE_PID"
851 echo "waiting for multiop $PID"
852 wait $PID || return 2
853 do_facet client munlink $MOUNT/$tfile || return 3
854 do_facet client munlink $MOUNT/${tfile}-2 || return 3
857 run_test 40 "cause recovery in ptlrpc, ensure IO continues"
861 # make sure that a read to one osc doesn't try to double-unlock its page just
862 # because another osc is invalid. trigger_group_io used to mistakenly return
863 # an error if any oscs were invalid even after having successfully put rpcs
864 # on valid oscs. This was fatal if the caller was ll_readpage who unlocked
865 # the page, guarnateeing that the unlock from the RPC completion would
866 # assert on trying to unlock the unlocked page.
868 [ $OSTCOUNT -lt 2 ] && \
869 skip "skipping test 41: we don't have a second OST to test with" && \
872 local f=$MOUNT/$tfile
873 # make sure the start of the file is ost1
874 lfs setstripe $f -s $((128 * 1024)) -i 0
875 do_facet client dd if=/dev/zero of=$f bs=4k count=1 || return 3
877 # fail ost2 and read from ost1
878 local osc2dev=`do_facet mds "lctl get_param -n devices | grep ${ost2_svc}-osc-MDT0000" | awk '{print $1}'`
879 [ -z "$osc2dev" ] && echo "OST: $ost2_svc" && lctl get_param -n devices && return 4
880 do_facet mds $LCTL --device $osc2dev deactivate || return 1
881 do_facet client dd if=$f of=/dev/null bs=4k count=1 || return 3
882 do_facet mds $LCTL --device $osc2dev activate || return 2
885 run_test 41 "read from a valid osc while other oscs are invalid"
887 # test MDS recovery after ost failure
889 blocks=`df -P $MOUNT | tail -n 1 | awk '{ print $2 }'`
890 createmany -o $DIR/$tfile-%d 800
892 unlinkmany $DIR/$tfile-%d 0 400
894 lctl set_param debug=-1
897 # osc is evicted, fs is smaller (but only with failout OSTs (bug 7287)
898 #blocks_after=`df -P $MOUNT | tail -n 1 | awk '{ print $2 }'`
899 #[ $blocks_after -lt $blocks ] || return 1
900 echo wait for MDS to timeout and recover
901 sleep $((TIMEOUT * 2))
903 unlinkmany $DIR/$tfile-%d 400 400
904 $CHECKSTAT -t file $DIR/$tfile-* && return 2 || true
906 run_test 42 "recovery after ost failure"
908 # timeout in MDS/OST recovery RPC will LBUG MDS
909 test_43() { # bug 2530
910 replay_barrier $SINGLEMDS
912 # OBD_FAIL_OST_CREATE_NET 0x204
913 do_facet ost1 "lctl set_param fail_loc=0x80000204"
916 do_facet ost1 "lctl set_param fail_loc=0"
920 run_test 43 "mds osc import failure during recovery; don't LBUG"
922 test_44a() { # was test_44
925 mdcdev=`lctl get_param -n devices | awk '/MDT0000-mdc-/ {print $1}'`
926 [ "$mdcdev" ] || exit 2
928 # adaptive timeouts slow this way down
929 if at_is_valid && at_is_enabled; then
930 at_max_saved=$(at_max_get mds)
934 for i in `seq 1 10`; do
935 echo "$i of 10 ($(date +%s))"
936 do_facet mds "grep service $LPROC/mdt/MDS/mds/timeouts"
937 #define OBD_FAIL_TGT_CONN_RACE 0x701
938 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000701"
939 $LCTL --device $mdcdev recover
942 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
943 [ $at_max_saved -ne 0 ] && at_max_set $at_max_saved mds
946 run_test 44a "race in target handle connect"
949 mdcdev=`lctl get_param -n devices | awk '/MDT0000-mdc-/ {print $1}'`
950 [ "$mdcdev" ] || exit 2
951 for i in `seq 1 10`; do
952 echo "$i of 10 ($(date +%s))"
953 do_facet mds "grep service $LPROC/mdt/MDS/mds/timeouts"
954 #define OBD_FAIL_TGT_DELAY_RECONNECT 0x704
955 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000704"
956 $LCTL --device $mdcdev recover
959 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
962 run_test 44b "race in target handle connect"
964 # Handle failed close
966 mdcdev=`lctl get_param -n devices | awk '/MDT0000-mdc-/ {print $1}'`
967 [ "$mdcdev" ] || exit 2
968 $LCTL --device $mdcdev recover
970 multiop_bg_pause $DIR/$tfile O_c || return 1
973 # This will cause the CLOSE to fail before even
974 # allocating a reply buffer
975 $LCTL --device $mdcdev deactivate || return 4
979 wait $pid || return 1
981 $LCTL --device $mdcdev activate || return 5
984 $CHECKSTAT -t file $DIR/$tfile || return 2
987 run_test 45 "Handle failed close"
991 drop_reply "touch $DIR/$tfile"
993 # ironically, the previous test, 45, will cause a real forced close,
994 # so just look for one for this test
995 dmesg | grep -i "force closing client file handle for $tfile" && return 1
998 run_test 46 "Don't leak file handle after open resend (3325)"
1000 test_47() { # bug 2824
1001 # create some files to make sure precreate has been done on all
1002 # OSTs. (just in case this test is run independently)
1003 createmany -o $DIR/$tfile 20 || return 1
1005 # OBD_FAIL_OST_CREATE_NET 0x204
1007 do_facet ost1 "lctl set_param fail_loc=0x80000204"
1008 df $MOUNT || return 2
1010 # let the MDS discover the OST failure, attempt to recover, fail
1011 # and recover again.
1012 sleep $((3 * TIMEOUT))
1014 # Without 2824, this createmany would hang
1015 createmany -o $DIR/$tfile 20 || return 3
1016 unlinkmany $DIR/$tfile 20 || return 4
1018 do_facet ost1 "lctl set_param fail_loc=0"
1021 run_test 47 "MDS->OSC failure during precreate cleanup (2824)"
1024 replay_barrier $SINGLEMDS
1025 createmany -o $DIR/$tfile 20 || return 1
1026 # OBD_FAIL_OST_EROFS 0x216
1028 do_facet ost1 "lctl set_param fail_loc=0x80000216"
1029 df $MOUNT || return 2
1031 createmany -o $DIR/$tfile 20 20 || return 2
1032 unlinkmany $DIR/$tfile 40 || return 3
1034 do_facet ost1 "lctl set_param fail_loc=0"
1037 run_test 48 "MDS->OSC failure during precreate cleanup (2824)"
1040 local oscdev=`do_facet $SINGLEMDS lctl get_param -n devices | grep ${ost1_svc}-osc-MDT0000 | awk '{print $1}'`
1041 [ "$oscdev" ] || return 1
1042 do_facet $SINGLEMDS $LCTL --device $oscdev recover || return 2
1043 do_facet $SINGLEMDS $LCTL --device $oscdev recover || return 3
1044 # give the mds_lov_sync threads a chance to run
1047 run_test 50 "Double OSC recovery, don't LASSERT (3812)"
1049 # b3764 timed out lock replay
1052 cancel_lru_locks mdc
1054 multiop $DIR/$tfile s || return 1
1055 replay_barrier $SINGLEMDS
1056 #define OBD_FAIL_LDLM_REPLY 0x30c
1057 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000030c"
1058 fail $SINGLEMDS || return 2
1059 do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
1061 $CHECKSTAT -t file $DIR/$tfile-* && return 3 || true
1063 run_test 52 "time out lock replay (3764)"
1065 # bug 3462 - simultaneous MDC requests
1067 mkdir -p $DIR/${tdir}-1
1068 mkdir -p $DIR/${tdir}-2
1069 multiop $DIR/${tdir}-1/f O_c &
1071 # give multiop a change to open
1074 #define OBD_FAIL_MDS_CLOSE_NET 0x115
1075 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
1076 kill -USR1 $close_pid
1077 cancel_lru_locks mdc # force the close
1078 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1080 mcreate $DIR/${tdir}-2/f || return 1
1082 # close should still be here
1083 [ -d /proc/$close_pid ] || return 2
1085 replay_barrier_nodf $SINGLEMDS
1087 wait $close_pid || return 3
1089 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 4
1090 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 5
1091 rm -rf $DIR/${tdir}-*
1093 run_test 53a "|X| close request while two MDC requests in flight"
1096 mkdir -p $DIR/${tdir}-1
1097 mkdir -p $DIR/${tdir}-2
1098 multiop $DIR/${tdir}-1/f O_c &
1101 #define OBD_FAIL_MDS_REINT_NET 0x107
1102 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
1103 mcreate $DIR/${tdir}-2/f &
1107 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1108 kill -USR1 $close_pid
1109 cancel_lru_locks mdc # force the close
1110 wait $close_pid || return 1
1111 # open should still be here
1112 [ -d /proc/$open_pid ] || return 2
1114 replay_barrier_nodf $SINGLEMDS
1116 wait $open_pid || return 3
1118 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 4
1119 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 5
1120 rm -rf $DIR/${tdir}-*
1122 run_test 53b "|X| open request while two MDC requests in flight"
1125 mkdir -p $DIR/${tdir}-1
1126 mkdir -p $DIR/${tdir}-2
1127 multiop $DIR/${tdir}-1/f O_c &
1130 #define OBD_FAIL_MDS_REINT_NET 0x107
1131 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
1132 mcreate $DIR/${tdir}-2/f &
1136 #define OBD_FAIL_MDS_CLOSE_NET 0x115
1137 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
1138 kill -USR1 $close_pid
1139 cancel_lru_locks mdc # force the close
1141 replay_barrier_nodf $SINGLEMDS
1142 fail_nodf $SINGLEMDS
1143 wait $open_pid || return 1
1145 # close should be gone
1146 [ -d /proc/$close_pid ] && return 2
1147 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1149 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 3
1150 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 4
1151 rm -rf $DIR/${tdir}-*
1153 run_test 53c "|X| open request and close request while two MDC requests in flight"
1156 mkdir -p $DIR/${tdir}-1
1157 mkdir -p $DIR/${tdir}-2
1158 multiop $DIR/${tdir}-1/f O_c &
1160 # give multiop a chance to open
1163 #define OBD_FAIL_MDS_CLOSE_NET_REP 0x13f
1164 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000013f"
1165 kill -USR1 $close_pid
1166 cancel_lru_locks mdc # force the close
1167 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1168 mcreate $DIR/${tdir}-2/f || return 1
1170 # close should still be here
1171 [ -d /proc/$close_pid ] || return 2
1173 wait $close_pid || return 3
1175 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 4
1176 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 5
1177 rm -rf $DIR/${tdir}-*
1179 run_test 53d "|X| close reply while two MDC requests in flight"
1182 mkdir -p $DIR/${tdir}-1
1183 mkdir -p $DIR/${tdir}-2
1184 multiop $DIR/${tdir}-1/f O_c &
1187 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
1188 do_facet $SINGLEMDS "lctl set_param fail_loc=0x119"
1189 mcreate $DIR/${tdir}-2/f &
1193 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1194 kill -USR1 $close_pid
1195 cancel_lru_locks mdc # force the close
1196 wait $close_pid || return 1
1197 # open should still be here
1198 [ -d /proc/$open_pid ] || return 2
1200 replay_barrier_nodf $SINGLEMDS
1202 wait $open_pid || return 3
1204 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 4
1205 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 5
1206 rm -rf $DIR/${tdir}-*
1208 run_test 53e "|X| open reply while two MDC requests in flight"
1211 mkdir -p $DIR/${tdir}-1
1212 mkdir -p $DIR/${tdir}-2
1213 multiop $DIR/${tdir}-1/f O_c &
1216 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
1217 do_facet $SINGLEMDS "lctl set_param fail_loc=0x119"
1218 mcreate $DIR/${tdir}-2/f &
1222 #define OBD_FAIL_MDS_CLOSE_NET_REP 0x13f
1223 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000013f"
1224 kill -USR1 $close_pid
1225 cancel_lru_locks mdc # force the close
1227 replay_barrier_nodf $SINGLEMDS
1228 fail_nodf $SINGLEMDS
1229 wait $open_pid || return 1
1231 # close should be gone
1232 [ -d /proc/$close_pid ] && return 2
1233 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1235 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 3
1236 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 4
1237 rm -rf $DIR/${tdir}-*
1239 run_test 53f "|X| open reply and close reply while two MDC requests in flight"
1242 mkdir -p $DIR/${tdir}-1
1243 mkdir -p $DIR/${tdir}-2
1244 multiop $DIR/${tdir}-1/f O_c &
1247 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
1248 do_facet $SINGLEMDS "lctl set_param fail_loc=0x119"
1249 mcreate $DIR/${tdir}-2/f &
1253 #define OBD_FAIL_MDS_CLOSE_NET 0x115
1254 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
1255 kill -USR1 $close_pid
1256 cancel_lru_locks mdc # force the close
1258 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1259 replay_barrier_nodf $SINGLEMDS
1260 fail_nodf $SINGLEMDS
1261 wait $open_pid || return 1
1263 # close should be gone
1264 [ -d /proc/$close_pid ] && return 2
1266 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 3
1267 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 4
1268 rm -rf $DIR/${tdir}-*
1270 run_test 53g "|X| drop open reply and close request while close and open are both in flight"
1273 mkdir -p $DIR/${tdir}-1
1274 mkdir -p $DIR/${tdir}-2
1275 multiop $DIR/${tdir}-1/f O_c &
1278 #define OBD_FAIL_MDS_REINT_NET 0x107
1279 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
1280 mcreate $DIR/${tdir}-2/f &
1284 #define OBD_FAIL_MDS_CLOSE_NET_REP 0x13f
1285 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000013f"
1286 kill -USR1 $close_pid
1287 cancel_lru_locks mdc # force the close
1290 replay_barrier_nodf $SINGLEMDS
1291 fail_nodf $SINGLEMDS
1292 wait $open_pid || return 1
1294 # close should be gone
1295 [ -d /proc/$close_pid ] && return 2
1296 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1298 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 3
1299 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 4
1300 rm -rf $DIR/${tdir}-*
1302 run_test 53h "|X| open request and close reply while two MDC requests in flight"
1304 #b_cray 54 "|X| open request and close reply while two MDC requests in flight"
1306 #b3761 ASSERTION(hash != 0) failed
1308 # OBD_FAIL_MDS_OPEN_CREATE | OBD_FAIL_ONCE
1309 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000012b"
1311 # give touch a chance to run
1313 do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
1317 run_test 55 "let MDS_CHECK_RESENT return the original return code instead of 0"
1319 #b3440 ASSERTION(rec->ur_fid2->id) failed
1321 ln -s foo $DIR/$tfile
1322 replay_barrier $SINGLEMDS
1323 #drop_reply "cat $DIR/$tfile"
1327 run_test 56 "don't replay a symlink open request (3440)"
1329 #recovery one mds-ost setattr from llog
1331 #define OBD_FAIL_MDS_OST_SETATTR 0x12c
1332 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000012c"
1334 replay_barrier $SINGLEMDS
1337 $CHECKSTAT -t file $DIR/$tfile || return 1
1338 do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
1341 run_test 57 "test recovery from llog for setattr op"
1343 #recovery many mds-ost setattr from llog
1346 #define OBD_FAIL_MDS_OST_SETATTR 0x12c
1347 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000012c"
1348 createmany -o $DIR/$tdir/$tfile-%d 2500
1349 replay_barrier $SINGLEMDS
1352 $CHECKSTAT -t file $DIR/$tdir/$tfile-* >/dev/null || return 1
1353 do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
1354 unlinkmany $DIR/$tdir/$tfile-%d 2500
1357 run_test 58 "test recovery from llog for setattr op (test llog_gen_rec)"
1359 # log_commit_thread vs filter_destroy race used to lead to import use after free
1363 createmany -o $DIR/$tdir/$tfile-%d 200
1365 unlinkmany $DIR/$tdir/$tfile-%d 200
1366 #define OBD_FAIL_PTLRPC_DELAY_RECOV 0x507
1367 do_facet ost1 "lctl set_param fail_loc=0x507"
1370 do_facet ost1 "lctl set_param fail_loc=0x0"
1374 run_test 59 "test log_commit_thread vs filter_destroy race"
1376 # race between add unlink llog vs cat log init in post_recovery (only for b1_6)
1377 # bug 12086: should no oops and No ctxt error for this test
1380 createmany -o $DIR/$tdir/$tfile-%d 200
1381 replay_barrier $SINGLEMDS
1382 unlinkmany $DIR/$tdir/$tfile-%d 0 100
1384 unlinkmany $DIR/$tdir/$tfile-%d 100 100
1385 local no_ctxt=`dmesg | grep "No ctxt"`
1386 [ -z "$no_ctxt" ] || error "ctxt is not initialized in recovery"
1388 run_test 60 "test llog post recovery init vs llog unlink"
1390 #test race llog recovery thread vs llog cleanup
1391 test_61a() { # was test_61
1393 createmany -o $DIR/$tdir/$tfile-%d 800
1395 # OBD_FAIL_OST_LLOG_RECOVERY_TIMEOUT 0x221
1396 unlinkmany $DIR/$tdir/$tfile-%d 800
1397 set_nodes_failloc "$(osts_nodes)" 0x80000221
1402 set_nodes_failloc "$(osts_nodes)" 0x0
1404 $CHECKSTAT -t file $DIR/$tdir/$tfile-* && return 1
1407 run_test 61a "test race llog recovery vs llog cleanup"
1409 #test race mds llog sync vs llog cleanup
1411 # OBD_FAIL_MDS_LLOG_SYNC_TIMEOUT 0x140
1412 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000140"
1413 facet_failover $SINGLEMDS
1416 do_facet client dd if=/dev/zero of=$DIR/$tfile bs=4k count=1 || return 1
1418 run_test 61b "test race mds llog sync vs llog cleanup"
1420 #test race cancel cookie cb vs llog cleanup
1422 # OBD_FAIL_OST_CANCEL_COOKIE_TIMEOUT 0x222
1424 set_nodes_failloc "$(osts_nodes)" 0x80000222
1428 set_nodes_failloc "$(osts_nodes)" 0x0
1430 run_test 61c "test race mds llog sync vs llog cleanup"
1432 test_62() { # Bug 15756 - don't mis-drop resent replay
1434 replay_barrier $SINGLEMDS
1435 createmany -o $DIR/$tdir/$tfile- 25
1436 #define OBD_FAIL_TGT_REPLAY_DROP 0x706
1437 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000707"
1438 facet_failover $SINGLEMDS
1439 df $MOUNT || return 1
1440 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1441 unlinkmany $DIR/$tdir/$tfile- 25 || return 2
1444 run_test 62 "don't mis-drop resent replay"
1446 #Adaptive Timeouts (bug 3055)
1451 if ! at_is_valid; then
1452 skip "AT env is invalid"
1456 if ! at_is_enabled; then
1457 echo "AT is disabled, enable it by force temporarily"
1458 at_max_set 600 mds ost client
1462 if [ -z "$ATOLDBASE" ]; then
1463 local at_history=$(do_facet mds "find /sys/ -name at_history")
1464 [ -z "$at_history" ] && skip "missing /sys/.../at_history " && return 1
1465 ATOLDBASE=$(do_facet mds "cat $at_history")
1466 # speed up the timebase so we can check decreasing AT
1467 do_facet mds "echo 8 >> $at_history"
1468 do_facet ost1 "echo 8 >> $at_history"
1472 test_65a() #bug 3055
1474 at_start || return 0
1475 $LCTL dk > /dev/null
1477 sysctl -w lnet.debug="+other"
1478 # slow down a request
1479 do_facet mds sysctl -w lustre.fail_val=30000
1480 #define OBD_FAIL_PTLRPC_PAUSE_REQ 0x50a
1481 do_facet mds sysctl -w lustre.fail_loc=0x8000050a
1482 createmany -o $DIR/$tfile 10 > /dev/null
1483 unlinkmany $DIR/$tfile 10 > /dev/null
1484 # check for log message
1485 $LCTL dk | grep "Early reply #" || error "No early reply"
1487 # client should show 30s estimates
1488 grep portal $LPROC/mdc/${FSNAME}-MDT0000-mdc-*/timeouts
1490 grep portal $LPROC/mdc/${FSNAME}-MDT0000-mdc-*/timeouts
1492 run_test 65a "AT: verify early replies"
1494 test_65b() #bug 3055
1496 at_start || return 0
1499 sysctl -w lnet.debug="other trace"
1500 $LCTL dk > /dev/null
1501 # slow down bulk i/o
1502 do_facet ost1 sysctl -w lustre.fail_val=30
1503 #define OBD_FAIL_OST_BRW_PAUSE_PACK 0x224
1504 do_facet ost1 sysctl -w lustre.fail_loc=0x224
1507 lfs setstripe $DIR/$tfile --index=0 --count=1
1508 # force some real bulk transfer
1509 multiop $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c
1511 do_facet ost1 sysctl -w lustre.fail_loc=0
1512 # check for log message
1513 $LCTL dk | grep "Early reply #" || error "No early reply"
1515 # client should show 30s estimates
1516 grep portal $LPROC/osc/${FSNAME}-OST0000-osc-*/timeouts
1518 run_test 65b "AT: verify early replies on packed reply / bulk"
1520 test_66a() #bug 3055
1522 at_start || return 0
1523 grep "portal 12" $LPROC/mdc/${FSNAME}-MDT0000-mdc-*/timeouts
1524 # adjust 5s at a time so no early reply is sent (within deadline)
1525 do_facet mds "sysctl -w lustre.fail_val=5000"
1526 #define OBD_FAIL_PTLRPC_PAUSE_REQ 0x50a
1527 do_facet mds "sysctl -w lustre.fail_loc=0x8000050a"
1528 createmany -o $DIR/$tfile 20 > /dev/null
1529 unlinkmany $DIR/$tfile 20 > /dev/null
1530 grep "portal 12" $LPROC/mdc/${FSNAME}-MDT0000-mdc-*/timeouts
1531 do_facet mds "sysctl -w lustre.fail_val=10000"
1532 do_facet mds "sysctl -w lustre.fail_loc=0x8000050a"
1533 createmany -o $DIR/$tfile 20 > /dev/null
1534 unlinkmany $DIR/$tfile 20 > /dev/null
1535 grep "portal 12" $LPROC/mdc/${FSNAME}-MDT0000-mdc-*/timeouts
1536 do_facet mds "sysctl -w lustre.fail_loc=0"
1538 createmany -o $DIR/$tfile 20 > /dev/null
1539 unlinkmany $DIR/$tfile 20 > /dev/null
1540 grep portal $LPROC/mdc/${FSNAME}-MDT0000-mdc-*/timeouts | grep "portal 12"
1541 CUR=$(awk '/portal 12/ {print $5}' $LPROC/mdc/${FSNAME}-MDT0000-mdc-*/timeouts)
1542 WORST=$(awk '/portal 12/ {print $7}' $LPROC/mdc/${FSNAME}-MDT0000-mdc-*/timeouts)
1543 echo "Current MDT timeout $CUR, worst $WORST"
1544 [ $CUR -lt $WORST ] || error "Current $CUR should be less than worst $WORST"
1546 run_test 66a "AT: verify MDT service time adjusts with no early replies"
1548 test_66b() #bug 3055
1550 at_start || return 0
1551 ORIG=$(awk '/network/ {print $4}' $LPROC/mdc/lustre-*/timeouts)
1552 sysctl -w lustre.fail_val=$(($ORIG + 5))
1553 #define OBD_FAIL_PTLRPC_PAUSE_REP 0x50c
1554 sysctl -w lustre.fail_loc=0x50c
1555 ls $DIR/$tfile > /dev/null 2>&1
1556 sysctl -w lustre.fail_loc=0
1557 CUR=$(awk '/network/ {print $4}' $LPROC/mdc/${FSNAME}-*/timeouts)
1558 WORST=$(awk '/network/ {print $6}' $LPROC/mdc/${FSNAME}-*/timeouts)
1559 echo "network timeout orig $ORIG, cur $CUR, worst $WORST"
1560 [ $WORST -gt $ORIG ] || error "Worst $WORST should be worse than orig $ORIG"
1562 run_test 66b "AT: verify net latency adjusts"
1564 test_67a() #bug 3055
1566 at_start || return 0
1567 CONN1=$(awk '/_connect/ {total+=$2} END {print total}' $LPROC/osc/*/stats)
1568 # sleeping threads may drive values above this
1569 do_facet ost1 "sysctl -w lustre.fail_val=400"
1570 #define OBD_FAIL_PTLRPC_PAUSE_REQ 0x50a
1571 do_facet ost1 "sysctl -w lustre.fail_loc=0x50a"
1572 createmany -o $DIR/$tfile 20 > /dev/null
1573 unlinkmany $DIR/$tfile 20 > /dev/null
1574 do_facet ost1 "sysctl -w lustre.fail_loc=0"
1575 CONN2=$(awk '/_connect/ {total+=$2} END {print total}' $LPROC/osc/*/stats)
1576 ATTEMPTS=$(($CONN2 - $CONN1))
1577 echo "$ATTEMPTS osc reconnect attemps on gradual slow"
1578 [ $ATTEMPTS -gt 0 ] && error_ignore 13721 "AT should have prevented reconnect"
1581 run_test 67a "AT: verify slow request processing doesn't induce reconnects"
1583 test_67b() #bug 3055
1585 at_start || return 0
1586 CONN1=$(awk '/_connect/ {total+=$2} END {print total}' $LPROC/osc/*/stats)
1587 #define OBD_FAIL_OST_PAUSE_CREATE 0x223
1588 do_facet ost1 "sysctl -w lustre.fail_val=20000"
1589 do_facet ost1 "sysctl -w lustre.fail_loc=0x80000223"
1590 cp /etc/profile $DIR/$tfile || error "cp failed"
1592 cat $LPROC/ost/OSS/ost_create/timeouts
1594 CONN2=$(awk '/_connect/ {total+=$2} END {print total}' $LPROC/osc/*/stats)
1595 ATTEMPTS=$(($CONN2 - $CONN1))
1596 echo "$ATTEMPTS osc reconnect attemps on instant slow"
1597 # do it again; should not timeout
1598 do_facet ost1 "sysctl -w lustre.fail_loc=0x80000223"
1599 cp /etc/profile $DIR/$tfile || error "cp failed"
1600 do_facet ost1 "sysctl -w lustre.fail_loc=0"
1602 cat $LPROC/ost/OSS/ost_create/timeouts
1603 CONN3=$(awk '/_connect/ {total+=$2} END {print total}' $LPROC/osc/*/stats)
1604 ATTEMPTS=$(($CONN3 - $CONN2))
1605 echo "$ATTEMPTS osc reconnect attemps on 2nd slow"
1606 [ $ATTEMPTS -gt 0 ] && error "AT should have prevented reconnect"
1609 run_test 67b "AT: verify instant slowdown doesn't induce reconnects"
1611 test_68 () #bug 13813
1613 at_start || return 0
1614 local ldlm_enqueue_min=$(find /sys -name ldlm_enqueue_min)
1615 [ -z "$ldlm_enqueue_min" ] && skip "missing /sys/.../ldlm_enqueue_min" && return 0
1616 local ENQ_MIN=$(cat $ldlm_enqueue_min)
1617 echo $TIMEOUT >> $ldlm_enqueue_min
1618 rm -f $DIR/${tfile}_[1-2]
1619 lfs setstripe $DIR/$tfile --index=0 --count=1
1620 #define OBD_FAIL_LDLM_PAUSE_CANCEL 0x312
1621 sysctl -w lustre.fail_val=$(($TIMEOUT - 1))
1622 sysctl -w lustre.fail_loc=0x80000312
1623 cp /etc/profile $DIR/${tfile}_1 || error "1st cp failed $?"
1624 sysctl -w lustre.fail_val=$((TIMEOUT * 3 / 2))
1625 sysctl -w lustre.fail_loc=0x80000312
1626 cp /etc/profile $DIR/${tfile}_2 || error "2nd cp failed $?"
1627 sysctl -w lustre.fail_loc=0
1628 echo $ENQ_MIN >> $ldlm_enqueue_min
1631 run_test 68 "AT: verify slowing locks"
1633 if [ -n "$ATOLDBASE" ]; then
1634 at_history=$(do_facet mds "find /sys/ -name at_history")
1635 do_facet mds "echo $ATOLDBASE >> $at_history" || true
1636 do_facet ost1 "echo $ATOLDBASE >> $at_history" || true
1639 if [ $AT_MAX_SET -ne 0 ]; then
1640 echo "restore AT status to be disabled"
1641 at_max_set 0 mds ost client
1644 # end of AT tests includes above lines
1647 # start multi-client tests
1649 [ -z "$CLIENTS" ] && \
1650 { skip "Need two or more clients." && return; }
1651 [ $CLIENTCOUNT -lt 2 ] && \
1652 { skip "Need two or more clients, have $CLIENTCOUNT" && return; }
1654 echo "mount clients $CLIENTS ..."
1655 zconf_mount_clients $CLIENTS $DIR
1657 local clients=${CLIENTS//,/ }
1658 echo "Write/read files on $DIR ; clients $CLIENTS ... "
1659 for CLIENT in $clients; do
1660 do_node $CLIENT dd bs=1M count=10 if=/dev/zero \
1661 of=$DIR/${tfile}_${CLIENT} 2>/dev/null || \
1662 error "dd failed on $CLIENT"
1665 local prev_client=$(echo $clients | sed 's/^.* \(.\+\)$/\1/')
1666 for C in ${CLIENTS//,/ }; do
1667 do_node $prev_client dd if=$DIR/${tfile}_${C} of=/dev/null 2>/dev/null || \
1668 error "dd if=$DIR/${tfile}_${C} failed on $prev_client"
1674 zconf_umount_clients $CLIENTS $DIR
1676 run_test 70a "check multi client t-f"
1679 [ -z "$CLIENTS" ] && \
1680 { skip "Need two or more clients." && return; }
1681 [ $CLIENTCOUNT -lt 2 ] && \
1682 { skip "Need two or more clients, have $CLIENTCOUNT" && return; }
1684 zconf_mount_clients $CLIENTS $DIR
1686 local duration="-t 60"
1687 local cmd="rundbench 1 $duration "
1689 for CLIENT in ${CLIENTS//,/ }; do
1690 $PDSH $CLIENT "set -x; PATH=:$PATH:$LUSTRE/utils:$LUSTRE/tests/:${DBENCH_LIB} DBENCH_LIB=${DBENCH_LIB} $cmd" &
1692 echo $PID >pid.$CLIENT
1693 echo "Started load PID=`cat pid.$CLIENT`"
1696 replay_barrier $SINGLEMDS
1697 sleep 3 # give clients a time to do operations
1699 log "$TESTNAME fail mds 1"
1702 # wait for client to reconnect to MDS
1705 for CLIENT in ${CLIENTS//,/ }; do
1706 PID=`cat pid.$CLIENT`
1709 echo "load on ${CLIENT} returned $rc"
1712 zconf_umount_clients $CLIENTS $DIR
1714 run_test 70b "mds recovery; $CLIENTCOUNT clients"
1715 # end multi-client tests
1717 equals_msg `basename $0`: test complete, cleaning up
1718 check_and_cleanup_lustre
1719 [ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG || true