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"
493 wait $pid || return 1
496 run_test 20c "check that client eviction does not affect file content"
499 replay_barrier $SINGLEMDS
500 multiop_bg_pause $DIR/$tfile O_tSc || return 5
503 touch $DIR/g11 || return 1
507 wait $pid || return 2
508 [ -e $DIR/$tfile ] && return 3
509 touch $DIR/h11 || return 4
512 run_test 21 "|X| open(O_CREAT), unlink touch new, replay, close (test mds_cleanup_orphans)"
515 multiop_bg_pause $DIR/$tfile O_tSc || return 3
518 replay_barrier $SINGLEMDS
523 wait $pid || return 1
524 [ -e $DIR/$tfile ] && return 2
527 run_test 22 "open(O_CREAT), |X| unlink, replay, close (test mds_cleanup_orphans)"
530 multiop_bg_pause $DIR/$tfile O_tSc || return 5
533 replay_barrier $SINGLEMDS
535 touch $DIR/g11 || return 1
539 wait $pid || return 2
540 [ -e $DIR/$tfile ] && return 3
541 touch $DIR/h11 || return 4
544 run_test 23 "open(O_CREAT), |X| unlink touch new, replay, close (test mds_cleanup_orphans)"
547 multiop_bg_pause $DIR/$tfile O_tSc || return 3
550 replay_barrier $SINGLEMDS
554 wait $pid || return 1
555 [ -e $DIR/$tfile ] && return 2
558 run_test 24 "open(O_CREAT), replay, unlink, close (test mds_cleanup_orphans)"
561 multiop_bg_pause $DIR/$tfile O_tSc || return 3
565 replay_barrier $SINGLEMDS
568 wait $pid || return 1
569 [ -e $DIR/$tfile ] && return 2
572 run_test 25 "open(O_CREAT), unlink, replay, close (test mds_cleanup_orphans)"
575 replay_barrier $SINGLEMDS
576 multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
578 multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
583 wait $pid2 || return 1
587 wait $pid1 || return 2
588 [ -e $DIR/$tfile-1 ] && return 3
589 [ -e $DIR/$tfile-2 ] && return 4
592 run_test 26 "|X| open(O_CREAT), unlink two, close one, replay, close one (test mds_cleanup_orphans)"
595 replay_barrier $SINGLEMDS
596 multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
598 multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
605 wait $pid1 || return 1
607 wait $pid2 || return 2
608 [ -e $DIR/$tfile-1 ] && return 3
609 [ -e $DIR/$tfile-2 ] && return 4
612 run_test 27 "|X| open(O_CREAT), unlink two, replay, close two (test mds_cleanup_orphans)"
615 multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
617 multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
619 replay_barrier $SINGLEMDS
623 wait $pid2 || return 1
627 wait $pid1 || return 2
628 [ -e $DIR/$tfile-1 ] && return 3
629 [ -e $DIR/$tfile-2 ] && return 4
632 run_test 28 "open(O_CREAT), |X| unlink two, close one, replay, close one (test mds_cleanup_orphans)"
635 multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
637 multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
639 replay_barrier $SINGLEMDS
645 wait $pid1 || return 1
647 wait $pid2 || return 2
648 [ -e $DIR/$tfile-1 ] && return 3
649 [ -e $DIR/$tfile-2 ] && return 4
652 run_test 29 "open(O_CREAT), |X| unlink two, replay, close two (test mds_cleanup_orphans)"
655 multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
657 multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
662 replay_barrier $SINGLEMDS
665 wait $pid1 || return 1
667 wait $pid2 || return 2
668 [ -e $DIR/$tfile-1 ] && return 3
669 [ -e $DIR/$tfile-2 ] && return 4
672 run_test 30 "open(O_CREAT) two, unlink two, replay, close two (test mds_cleanup_orphans)"
675 multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
677 multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
681 replay_barrier $SINGLEMDS
685 wait $pid1 || return 1
687 wait $pid2 || return 2
688 [ -e $DIR/$tfile-1 ] && return 3
689 [ -e $DIR/$tfile-2 ] && return 4
692 run_test 31 "open(O_CREAT) two, unlink one, |X| unlink one, close two (test mds_cleanup_orphans)"
694 # tests for bug 2104; completion without crashing is success. The close is
695 # stale, but we always return 0 for close, so the app never sees it.
697 multiop_bg_pause $DIR/$tfile O_c || return 2
699 multiop_bg_pause $DIR/$tfile O_c || return 3
702 df $MOUNT || sleep 1 && df $MOUNT || return 1
705 wait $pid1 || return 4
706 wait $pid2 || return 5
709 run_test 32 "close() notices client eviction; close() after client eviction"
711 # Abort recovery before client complete
712 test_33a() { # was test_33
713 replay_barrier $SINGLEMDS
714 createmany -o $DIR/$tfile-%d 100
715 fail_abort $SINGLEMDS
716 # this file should be gone, because the replay was aborted
717 $CHECKSTAT -t file $DIR/$tfile-* && return 3
718 unlinkmany $DIR/$tfile-%d 0 100
721 run_test 33a "abort recovery before client does replay"
724 test_33b() { # was test_33a
725 replay_barrier $SINGLEMDS
726 createmany -o $DIR/$tfile-%d 10
727 fail_abort $SINGLEMDS
728 unlinkmany $DIR/$tfile-%d 0 10
729 # recreate shouldn't fail
730 createmany -o $DIR/$tfile-%d 10 || return 3
731 unlinkmany $DIR/$tfile-%d 0 10
734 run_test 33b "fid shouldn't be reused after abort recovery"
737 multiop_bg_pause $DIR/$tfile O_c || return 2
741 replay_barrier $SINGLEMDS
742 fail_abort $SINGLEMDS
744 wait $pid || return 3
745 [ -e $DIR/$tfile ] && return 1
749 run_test 34 "abort recovery before client does replay (test mds_cleanup_orphans)"
751 # bug 2278 - generate one orphan on OST, then destroy it during recovery from llog
755 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
756 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000119"
761 # give a chance to remove from MDS
762 fail_abort $SINGLEMDS
763 $CHECKSTAT -t file $DIR/$tfile && return 1 || true
765 run_test 35 "test recovery from llog for unlink op"
767 # b=2432 resent cancel after replay uses wrong cookie,
768 # so don't resend cancels
770 replay_barrier $SINGLEMDS
772 checkstat $DIR/$tfile
773 facet_failover $SINGLEMDS
775 if dmesg | grep "unknown lock cookie"; then
776 echo "cancel after replay failed"
780 run_test 36 "don't resend cancel"
783 # directory orphans can't be unlinked from PENDING directory
785 rmdir $DIR/$tfile 2>/dev/null
786 multiop_bg_pause $DIR/$tfile dD_c || return 2
790 replay_barrier $SINGLEMDS
791 # clear the dmesg buffer so we only see errors from this recovery
793 fail_abort $SINGLEMDS
795 dmesg | grep "mds_unlink_orphan.*error .* unlinking orphan" && return 1
796 wait $pid || return 3
800 run_test 37 "abort recovery before client does replay (test mds_cleanup_orphans for directories)"
803 createmany -o $DIR/$tfile-%d 800
804 unlinkmany $DIR/$tfile-%d 0 400
805 replay_barrier $SINGLEMDS
807 unlinkmany $DIR/$tfile-%d 400 400
809 $CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
811 run_test 38 "test recovery from unlink llog (test llog_gen_rec) "
813 test_39() { # bug 4176
814 createmany -o $DIR/$tfile-%d 800
815 replay_barrier $SINGLEMDS
816 unlinkmany $DIR/$tfile-%d 0 400
818 unlinkmany $DIR/$tfile-%d 400 400
820 $CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
822 run_test 39 "test recovery from unlink llog (test llog_gen_rec) "
825 lctl get_param -n osc.*.stats | awk -vwrites=0 '/ost_write/ { writes += $2 } END { print writes; }'
830 $LCTL mark multiop $MOUNT/$tfile OS_c
831 multiop $MOUNT/$tfile OS_c &
833 writeme -s $MOUNT/${tfile}-2 &
836 facet_failover $SINGLEMDS
837 #define OBD_FAIL_MDS_CONNECT_NET 0x117
838 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000117"
840 stat1=`count_ost_writes`
842 stat2=`count_ost_writes`
843 echo "$stat1, $stat2"
844 if [ $stat1 -lt $stat2 ]; then
845 echo "writes continuing during recovery"
848 echo "writes not continuing during recovery, bug 2477"
851 echo "waiting for writeme $WRITE_PID"
855 echo "waiting for multiop $PID"
856 wait $PID || return 2
857 do_facet client munlink $MOUNT/$tfile || return 3
858 do_facet client munlink $MOUNT/${tfile}-2 || return 3
861 run_test 40 "cause recovery in ptlrpc, ensure IO continues"
865 # make sure that a read to one osc doesn't try to double-unlock its page just
866 # because another osc is invalid. trigger_group_io used to mistakenly return
867 # an error if any oscs were invalid even after having successfully put rpcs
868 # on valid oscs. This was fatal if the caller was ll_readpage who unlocked
869 # the page, guarnateeing that the unlock from the RPC completion would
870 # assert on trying to unlock the unlocked page.
872 [ $OSTCOUNT -lt 2 ] && \
873 skip "skipping test 41: we don't have a second OST to test with" && \
876 local f=$MOUNT/$tfile
877 # make sure the start of the file is ost1
878 lfs setstripe $f -s $((128 * 1024)) -i 0
879 do_facet client dd if=/dev/zero of=$f bs=4k count=1 || return 3
881 # fail ost2 and read from ost1
882 local osc2dev=`do_facet mds "lctl get_param -n devices | grep ${ost2_svc}-osc-MDT0000" | awk '{print $1}'`
883 [ -z "$osc2dev" ] && echo "OST: $ost2_svc" && lctl get_param -n devices && return 4
884 do_facet mds $LCTL --device $osc2dev deactivate || return 1
885 do_facet client dd if=$f of=/dev/null bs=4k count=1 || return 3
886 do_facet mds $LCTL --device $osc2dev activate || return 2
889 run_test 41 "read from a valid osc while other oscs are invalid"
891 # test MDS recovery after ost failure
893 blocks=`df -P $MOUNT | tail -n 1 | awk '{ print $2 }'`
894 createmany -o $DIR/$tfile-%d 800
896 unlinkmany $DIR/$tfile-%d 0 400
898 lctl set_param debug=-1
901 # osc is evicted, fs is smaller (but only with failout OSTs (bug 7287)
902 #blocks_after=`df -P $MOUNT | tail -n 1 | awk '{ print $2 }'`
903 #[ $blocks_after -lt $blocks ] || return 1
904 echo wait for MDS to timeout and recover
905 sleep $((TIMEOUT * 2))
907 unlinkmany $DIR/$tfile-%d 400 400
908 $CHECKSTAT -t file $DIR/$tfile-* && return 2 || true
910 run_test 42 "recovery after ost failure"
912 # timeout in MDS/OST recovery RPC will LBUG MDS
913 test_43() { # bug 2530
914 replay_barrier $SINGLEMDS
916 # OBD_FAIL_OST_CREATE_NET 0x204
917 do_facet ost1 "lctl set_param fail_loc=0x80000204"
920 do_facet ost1 "lctl set_param fail_loc=0"
924 run_test 43 "mds osc import failure during recovery; don't LBUG"
926 test_44a() { # was test_44
929 mdcdev=`lctl get_param -n devices | awk '/MDT0000-mdc-/ {print $1}'`
930 [ "$mdcdev" ] || exit 2
932 # adaptive timeouts slow this way down
933 if at_is_valid && at_is_enabled; then
934 at_max_saved=$(at_max_get mds)
938 for i in `seq 1 10`; do
939 echo "$i of 10 ($(date +%s))"
940 do_facet mds "grep service $LPROC/mdt/MDS/mds/timeouts"
941 #define OBD_FAIL_TGT_CONN_RACE 0x701
942 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000701"
943 $LCTL --device $mdcdev recover
946 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
947 [ $at_max_saved -ne 0 ] && at_max_set $at_max_saved mds
950 run_test 44a "race in target handle connect"
953 mdcdev=`lctl get_param -n devices | awk '/MDT0000-mdc-/ {print $1}'`
954 [ "$mdcdev" ] || exit 2
955 for i in `seq 1 10`; do
956 echo "$i of 10 ($(date +%s))"
957 do_facet mds "grep service $LPROC/mdt/MDS/mds/timeouts"
958 #define OBD_FAIL_TGT_DELAY_RECONNECT 0x704
959 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000704"
960 $LCTL --device $mdcdev recover
963 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
966 run_test 44b "race in target handle connect"
968 # Handle failed close
970 mdcdev=`lctl get_param -n devices | awk '/MDT0000-mdc-/ {print $1}'`
971 [ "$mdcdev" ] || exit 2
972 $LCTL --device $mdcdev recover
974 multiop_bg_pause $DIR/$tfile O_c || return 1
977 # This will cause the CLOSE to fail before even
978 # allocating a reply buffer
979 $LCTL --device $mdcdev deactivate || return 4
983 wait $pid || return 1
985 $LCTL --device $mdcdev activate || return 5
988 $CHECKSTAT -t file $DIR/$tfile || return 2
991 run_test 45 "Handle failed close"
995 drop_reply "touch $DIR/$tfile"
997 # ironically, the previous test, 45, will cause a real forced close,
998 # so just look for one for this test
999 dmesg | grep -i "force closing client file handle for $tfile" && return 1
1002 run_test 46 "Don't leak file handle after open resend (3325)"
1004 test_47() { # bug 2824
1005 # create some files to make sure precreate has been done on all
1006 # OSTs. (just in case this test is run independently)
1007 createmany -o $DIR/$tfile 20 || return 1
1009 # OBD_FAIL_OST_CREATE_NET 0x204
1011 do_facet ost1 "lctl set_param fail_loc=0x80000204"
1012 df $MOUNT || return 2
1014 # let the MDS discover the OST failure, attempt to recover, fail
1015 # and recover again.
1016 sleep $((3 * TIMEOUT))
1018 # Without 2824, this createmany would hang
1019 createmany -o $DIR/$tfile 20 || return 3
1020 unlinkmany $DIR/$tfile 20 || return 4
1022 do_facet ost1 "lctl set_param fail_loc=0"
1025 run_test 47 "MDS->OSC failure during precreate cleanup (2824)"
1028 replay_barrier $SINGLEMDS
1029 createmany -o $DIR/$tfile 20 || return 1
1030 # OBD_FAIL_OST_EROFS 0x216
1032 do_facet ost1 "lctl set_param fail_loc=0x80000216"
1033 df $MOUNT || return 2
1035 createmany -o $DIR/$tfile 20 20 || return 2
1036 unlinkmany $DIR/$tfile 40 || return 3
1038 do_facet ost1 "lctl set_param fail_loc=0"
1041 run_test 48 "MDS->OSC failure during precreate cleanup (2824)"
1044 local oscdev=`do_facet $SINGLEMDS lctl get_param -n devices | grep ${ost1_svc}-osc-MDT0000 | awk '{print $1}'`
1045 [ "$oscdev" ] || return 1
1046 do_facet $SINGLEMDS $LCTL --device $oscdev recover || return 2
1047 do_facet $SINGLEMDS $LCTL --device $oscdev recover || return 3
1048 # give the mds_lov_sync threads a chance to run
1051 run_test 50 "Double OSC recovery, don't LASSERT (3812)"
1053 # b3764 timed out lock replay
1056 cancel_lru_locks mdc
1058 multiop $DIR/$tfile s || return 1
1059 replay_barrier $SINGLEMDS
1060 #define OBD_FAIL_LDLM_REPLY 0x30c
1061 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000030c"
1062 fail $SINGLEMDS || return 2
1063 do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
1065 $CHECKSTAT -t file $DIR/$tfile-* && return 3 || true
1067 run_test 52 "time out lock replay (3764)"
1069 # bug 3462 - simultaneous MDC requests
1071 mkdir -p $DIR/${tdir}-1
1072 mkdir -p $DIR/${tdir}-2
1073 multiop $DIR/${tdir}-1/f O_c &
1075 # give multiop a change to open
1078 #define OBD_FAIL_MDS_CLOSE_NET 0x115
1079 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
1080 kill -USR1 $close_pid
1081 cancel_lru_locks mdc # force the close
1082 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1084 mcreate $DIR/${tdir}-2/f || return 1
1086 # close should still be here
1087 [ -d /proc/$close_pid ] || return 2
1089 replay_barrier_nodf $SINGLEMDS
1091 wait $close_pid || return 3
1093 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 4
1094 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 5
1095 rm -rf $DIR/${tdir}-*
1097 run_test 53a "|X| close request while two MDC requests in flight"
1100 mkdir -p $DIR/${tdir}-1
1101 mkdir -p $DIR/${tdir}-2
1102 multiop $DIR/${tdir}-1/f O_c &
1105 #define OBD_FAIL_MDS_REINT_NET 0x107
1106 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
1107 mcreate $DIR/${tdir}-2/f &
1111 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1112 kill -USR1 $close_pid
1113 cancel_lru_locks mdc # force the close
1114 wait $close_pid || return 1
1115 # open should still be here
1116 [ -d /proc/$open_pid ] || return 2
1118 replay_barrier_nodf $SINGLEMDS
1120 wait $open_pid || return 3
1122 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 4
1123 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 5
1124 rm -rf $DIR/${tdir}-*
1126 run_test 53b "|X| open request while two MDC requests in flight"
1129 mkdir -p $DIR/${tdir}-1
1130 mkdir -p $DIR/${tdir}-2
1131 multiop $DIR/${tdir}-1/f O_c &
1134 #define OBD_FAIL_MDS_REINT_NET 0x107
1135 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
1136 mcreate $DIR/${tdir}-2/f &
1140 #define OBD_FAIL_MDS_CLOSE_NET 0x115
1141 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
1142 kill -USR1 $close_pid
1143 cancel_lru_locks mdc # force the close
1145 replay_barrier_nodf $SINGLEMDS
1146 fail_nodf $SINGLEMDS
1147 wait $open_pid || return 1
1149 # close should be gone
1150 [ -d /proc/$close_pid ] && return 2
1151 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1153 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 3
1154 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 4
1155 rm -rf $DIR/${tdir}-*
1157 run_test 53c "|X| open request and close request while two MDC requests in flight"
1160 mkdir -p $DIR/${tdir}-1
1161 mkdir -p $DIR/${tdir}-2
1162 multiop $DIR/${tdir}-1/f O_c &
1164 # give multiop a chance to open
1167 #define OBD_FAIL_MDS_CLOSE_NET_REP 0x13f
1168 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000013f"
1169 kill -USR1 $close_pid
1170 cancel_lru_locks mdc # force the close
1171 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1172 mcreate $DIR/${tdir}-2/f || return 1
1174 # close should still be here
1175 [ -d /proc/$close_pid ] || return 2
1177 wait $close_pid || return 3
1179 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 4
1180 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 5
1181 rm -rf $DIR/${tdir}-*
1183 run_test 53d "|X| close reply while two MDC requests in flight"
1186 mkdir -p $DIR/${tdir}-1
1187 mkdir -p $DIR/${tdir}-2
1188 multiop $DIR/${tdir}-1/f O_c &
1191 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
1192 do_facet $SINGLEMDS "lctl set_param fail_loc=0x119"
1193 mcreate $DIR/${tdir}-2/f &
1197 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1198 kill -USR1 $close_pid
1199 cancel_lru_locks mdc # force the close
1200 wait $close_pid || return 1
1201 # open should still be here
1202 [ -d /proc/$open_pid ] || return 2
1204 replay_barrier_nodf $SINGLEMDS
1206 wait $open_pid || return 3
1208 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 4
1209 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 5
1210 rm -rf $DIR/${tdir}-*
1212 run_test 53e "|X| open reply while two MDC requests in flight"
1215 mkdir -p $DIR/${tdir}-1
1216 mkdir -p $DIR/${tdir}-2
1217 multiop $DIR/${tdir}-1/f O_c &
1220 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
1221 do_facet $SINGLEMDS "lctl set_param fail_loc=0x119"
1222 mcreate $DIR/${tdir}-2/f &
1226 #define OBD_FAIL_MDS_CLOSE_NET_REP 0x13f
1227 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000013f"
1228 kill -USR1 $close_pid
1229 cancel_lru_locks mdc # force the close
1231 replay_barrier_nodf $SINGLEMDS
1232 fail_nodf $SINGLEMDS
1233 wait $open_pid || return 1
1235 # close should be gone
1236 [ -d /proc/$close_pid ] && return 2
1237 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1239 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 3
1240 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 4
1241 rm -rf $DIR/${tdir}-*
1243 run_test 53f "|X| open reply and close reply while two MDC requests in flight"
1246 mkdir -p $DIR/${tdir}-1
1247 mkdir -p $DIR/${tdir}-2
1248 multiop $DIR/${tdir}-1/f O_c &
1251 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
1252 do_facet $SINGLEMDS "lctl set_param fail_loc=0x119"
1253 mcreate $DIR/${tdir}-2/f &
1257 #define OBD_FAIL_MDS_CLOSE_NET 0x115
1258 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
1259 kill -USR1 $close_pid
1260 cancel_lru_locks mdc # force the close
1262 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1263 replay_barrier_nodf $SINGLEMDS
1264 fail_nodf $SINGLEMDS
1265 wait $open_pid || return 1
1267 # close should be gone
1268 [ -d /proc/$close_pid ] && return 2
1270 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 3
1271 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 4
1272 rm -rf $DIR/${tdir}-*
1274 run_test 53g "|X| drop open reply and close request while close and open are both in flight"
1277 mkdir -p $DIR/${tdir}-1
1278 mkdir -p $DIR/${tdir}-2
1279 multiop $DIR/${tdir}-1/f O_c &
1282 #define OBD_FAIL_MDS_REINT_NET 0x107
1283 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
1284 mcreate $DIR/${tdir}-2/f &
1288 #define OBD_FAIL_MDS_CLOSE_NET_REP 0x13f
1289 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000013f"
1290 kill -USR1 $close_pid
1291 cancel_lru_locks mdc # force the close
1294 replay_barrier_nodf $SINGLEMDS
1295 fail_nodf $SINGLEMDS
1296 wait $open_pid || return 1
1298 # close should be gone
1299 [ -d /proc/$close_pid ] && return 2
1300 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1302 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 3
1303 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 4
1304 rm -rf $DIR/${tdir}-*
1306 run_test 53h "|X| open request and close reply while two MDC requests in flight"
1308 #b_cray 54 "|X| open request and close reply while two MDC requests in flight"
1310 #b3761 ASSERTION(hash != 0) failed
1312 # OBD_FAIL_MDS_OPEN_CREATE | OBD_FAIL_ONCE
1313 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000012b"
1315 # give touch a chance to run
1317 do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
1321 run_test 55 "let MDS_CHECK_RESENT return the original return code instead of 0"
1323 #b3440 ASSERTION(rec->ur_fid2->id) failed
1325 ln -s foo $DIR/$tfile
1326 replay_barrier $SINGLEMDS
1327 #drop_reply "cat $DIR/$tfile"
1331 run_test 56 "don't replay a symlink open request (3440)"
1333 #recovery one mds-ost setattr from llog
1335 #define OBD_FAIL_MDS_OST_SETATTR 0x12c
1336 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000012c"
1338 replay_barrier $SINGLEMDS
1341 $CHECKSTAT -t file $DIR/$tfile || return 1
1342 do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
1345 run_test 57 "test recovery from llog for setattr op"
1347 #recovery many mds-ost setattr from llog
1350 #define OBD_FAIL_MDS_OST_SETATTR 0x12c
1351 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000012c"
1352 createmany -o $DIR/$tdir/$tfile-%d 2500
1353 replay_barrier $SINGLEMDS
1356 $CHECKSTAT -t file $DIR/$tdir/$tfile-* >/dev/null || return 1
1357 do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
1358 unlinkmany $DIR/$tdir/$tfile-%d 2500
1361 run_test 58 "test recovery from llog for setattr op (test llog_gen_rec)"
1365 touch $DIR/$tdir/$tfile
1366 replay_barrier $SINGLEMDS
1367 setfattr -n trusted.foo -v bar $DIR/$tdir/$tfile
1369 VAL=`getfattr --absolute-names --only-value -n trusted.foo $DIR/$tdir/$tfile`
1370 [ x$VAL = x"bar" ] || return 1
1371 rm -f $DIR/$tdir/$tfile
1374 run_test 58a "test replay of setxattr op"
1376 # log_commit_thread vs filter_destroy race used to lead to import use after free
1380 createmany -o $DIR/$tdir/$tfile-%d 200
1382 unlinkmany $DIR/$tdir/$tfile-%d 200
1383 #define OBD_FAIL_PTLRPC_DELAY_RECOV 0x507
1384 do_facet ost1 "lctl set_param fail_loc=0x507"
1387 do_facet ost1 "lctl set_param fail_loc=0x0"
1391 run_test 59 "test log_commit_thread vs filter_destroy race"
1393 # race between add unlink llog vs cat log init in post_recovery (only for b1_6)
1394 # bug 12086: should no oops and No ctxt error for this test
1397 createmany -o $DIR/$tdir/$tfile-%d 200
1398 replay_barrier $SINGLEMDS
1399 unlinkmany $DIR/$tdir/$tfile-%d 0 100
1401 unlinkmany $DIR/$tdir/$tfile-%d 100 100
1402 local no_ctxt=`dmesg | grep "No ctxt"`
1403 [ -z "$no_ctxt" ] || error "ctxt is not initialized in recovery"
1405 run_test 60 "test llog post recovery init vs llog unlink"
1407 #test race llog recovery thread vs llog cleanup
1408 test_61a() { # was test_61
1410 createmany -o $DIR/$tdir/$tfile-%d 800
1412 # OBD_FAIL_OST_LLOG_RECOVERY_TIMEOUT 0x221
1413 unlinkmany $DIR/$tdir/$tfile-%d 800
1414 set_nodes_failloc "$(osts_nodes)" 0x80000221
1419 set_nodes_failloc "$(osts_nodes)" 0x0
1421 $CHECKSTAT -t file $DIR/$tdir/$tfile-* && return 1
1424 run_test 61a "test race llog recovery vs llog cleanup"
1426 #test race mds llog sync vs llog cleanup
1428 # OBD_FAIL_MDS_LLOG_SYNC_TIMEOUT 0x140
1429 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000140"
1430 facet_failover $SINGLEMDS
1433 do_facet client dd if=/dev/zero of=$DIR/$tfile bs=4k count=1 || return 1
1435 run_test 61b "test race mds llog sync vs llog cleanup"
1437 #test race cancel cookie cb vs llog cleanup
1439 # OBD_FAIL_OST_CANCEL_COOKIE_TIMEOUT 0x222
1441 set_nodes_failloc "$(osts_nodes)" 0x80000222
1445 set_nodes_failloc "$(osts_nodes)" 0x0
1447 run_test 61c "test race mds llog sync vs llog cleanup"
1449 test_62() { # Bug 15756 - don't mis-drop resent replay
1451 replay_barrier $SINGLEMDS
1452 createmany -o $DIR/$tdir/$tfile- 25
1453 #define OBD_FAIL_TGT_REPLAY_DROP 0x706
1454 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000707"
1455 facet_failover $SINGLEMDS
1456 df $MOUNT || return 1
1457 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1458 unlinkmany $DIR/$tdir/$tfile- 25 || return 2
1461 run_test 62 "don't mis-drop resent replay"
1463 #Adaptive Timeouts (bug 3055)
1465 # Suppose that all osts have the same at_max
1466 for facet in mds client ost; do
1467 eval AT_MAX_SAVE_${facet}=$(at_max_get $facet)
1472 local at_max_new=600
1473 if ! at_is_valid; then
1474 skip "AT env is invalid"
1480 for facet in mds client ost; do
1481 at_max=$(at_max_get $facet)
1482 if [ $at_max -ne $at_max_new ]; then
1483 echo "AT value on $facet is $at_max, set it by force temporarily to $at_max_new"
1484 at_max_set $at_max_new $facet
1489 if [ -z "$ATOLDBASE" ]; then
1490 local at_history=$(do_facet mds "find /sys/ -name at_history")
1491 [ -z "$at_history" ] && skip "missing /sys/.../at_history " && return 1
1492 ATOLDBASE=$(do_facet mds "cat $at_history")
1493 # speed up the timebase so we can check decreasing AT
1494 do_facet mds "echo 8 >> $at_history"
1495 do_facet ost1 "echo 8 >> $at_history"
1499 test_65a() #bug 3055
1501 at_start || return 0
1502 $LCTL dk > /dev/null
1504 sysctl -w lnet.debug="+other"
1505 # slow down a request
1506 do_facet mds sysctl -w lustre.fail_val=30000
1507 #define OBD_FAIL_PTLRPC_PAUSE_REQ 0x50a
1508 do_facet mds sysctl -w lustre.fail_loc=0x8000050a
1509 createmany -o $DIR/$tfile 10 > /dev/null
1510 unlinkmany $DIR/$tfile 10 > /dev/null
1511 # check for log message
1512 $LCTL dk | grep "Early reply #" || error "No early reply"
1514 # client should show 30s estimates
1515 grep portal $LPROC/mdc/${FSNAME}-MDT0000-mdc-*/timeouts
1517 grep portal $LPROC/mdc/${FSNAME}-MDT0000-mdc-*/timeouts
1519 run_test 65a "AT: verify early replies"
1521 test_65b() #bug 3055
1523 at_start || return 0
1526 sysctl -w lnet.debug="other trace"
1527 $LCTL dk > /dev/null
1528 # slow down bulk i/o
1529 do_facet ost1 sysctl -w lustre.fail_val=30
1530 #define OBD_FAIL_OST_BRW_PAUSE_PACK 0x224
1531 do_facet ost1 sysctl -w lustre.fail_loc=0x224
1534 lfs setstripe $DIR/$tfile --index=0 --count=1
1535 # force some real bulk transfer
1536 multiop $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c
1538 do_facet ost1 sysctl -w lustre.fail_loc=0
1539 # check for log message
1540 $LCTL dk | grep "Early reply #" || error "No early reply"
1542 # client should show 30s estimates
1543 grep portal $LPROC/osc/${FSNAME}-OST0000-osc-*/timeouts
1545 run_test 65b "AT: verify early replies on packed reply / bulk"
1547 test_66a() #bug 3055
1549 at_start || return 0
1550 grep "portal 12" $LPROC/mdc/${FSNAME}-MDT0000-mdc-*/timeouts
1551 # adjust 5s at a time so no early reply is sent (within deadline)
1552 do_facet mds "sysctl -w lustre.fail_val=5000"
1553 #define OBD_FAIL_PTLRPC_PAUSE_REQ 0x50a
1554 do_facet mds "sysctl -w lustre.fail_loc=0x8000050a"
1555 createmany -o $DIR/$tfile 20 > /dev/null
1556 unlinkmany $DIR/$tfile 20 > /dev/null
1557 grep "portal 12" $LPROC/mdc/${FSNAME}-MDT0000-mdc-*/timeouts
1558 do_facet mds "sysctl -w lustre.fail_val=10000"
1559 do_facet mds "sysctl -w lustre.fail_loc=0x8000050a"
1560 createmany -o $DIR/$tfile 20 > /dev/null
1561 unlinkmany $DIR/$tfile 20 > /dev/null
1562 grep "portal 12" $LPROC/mdc/${FSNAME}-MDT0000-mdc-*/timeouts
1563 do_facet mds "sysctl -w lustre.fail_loc=0"
1565 createmany -o $DIR/$tfile 20 > /dev/null
1566 unlinkmany $DIR/$tfile 20 > /dev/null
1567 grep portal $LPROC/mdc/${FSNAME}-MDT0000-mdc-*/timeouts | grep "portal 12"
1568 CUR=$(awk '/portal 12/ {print $5}' $LPROC/mdc/${FSNAME}-MDT0000-mdc-*/timeouts)
1569 WORST=$(awk '/portal 12/ {print $7}' $LPROC/mdc/${FSNAME}-MDT0000-mdc-*/timeouts)
1570 echo "Current MDT timeout $CUR, worst $WORST"
1571 [ $CUR -lt $WORST ] || error "Current $CUR should be less than worst $WORST"
1573 run_test 66a "AT: verify MDT service time adjusts with no early replies"
1575 test_66b() #bug 3055
1577 at_start || return 0
1578 ORIG=$(awk '/network/ {print $4}' $LPROC/mdc/lustre-*/timeouts)
1579 sysctl -w lustre.fail_val=$(($ORIG + 5))
1580 #define OBD_FAIL_PTLRPC_PAUSE_REP 0x50c
1581 sysctl -w lustre.fail_loc=0x50c
1582 ls $DIR/$tfile > /dev/null 2>&1
1583 sysctl -w lustre.fail_loc=0
1584 CUR=$(awk '/network/ {print $4}' $LPROC/mdc/${FSNAME}-*/timeouts)
1585 WORST=$(awk '/network/ {print $6}' $LPROC/mdc/${FSNAME}-*/timeouts)
1586 echo "network timeout orig $ORIG, cur $CUR, worst $WORST"
1587 [ $WORST -gt $ORIG ] || error "Worst $WORST should be worse than orig $ORIG"
1589 run_test 66b "AT: verify net latency adjusts"
1591 test_67a() #bug 3055
1593 at_start || return 0
1594 CONN1=$(awk '/_connect/ {total+=$2} END {print total}' $LPROC/osc/*/stats)
1595 # sleeping threads may drive values above this
1596 do_facet ost1 "sysctl -w lustre.fail_val=400"
1597 #define OBD_FAIL_PTLRPC_PAUSE_REQ 0x50a
1598 do_facet ost1 "sysctl -w lustre.fail_loc=0x50a"
1599 createmany -o $DIR/$tfile 20 > /dev/null
1600 unlinkmany $DIR/$tfile 20 > /dev/null
1601 do_facet ost1 "sysctl -w lustre.fail_loc=0"
1602 CONN2=$(awk '/_connect/ {total+=$2} END {print total}' $LPROC/osc/*/stats)
1603 ATTEMPTS=$(($CONN2 - $CONN1))
1604 echo "$ATTEMPTS osc reconnect attemps on gradual slow"
1605 [ $ATTEMPTS -gt 0 ] && error_ignore 13721 "AT should have prevented reconnect"
1608 run_test 67a "AT: verify slow request processing doesn't induce reconnects"
1610 test_67b() #bug 3055
1612 at_start || return 0
1613 CONN1=$(awk '/_connect/ {total+=$2} END {print total}' $LPROC/osc/*/stats)
1614 #define OBD_FAIL_OST_PAUSE_CREATE 0x223
1615 do_facet ost1 "sysctl -w lustre.fail_val=20000"
1616 do_facet ost1 "sysctl -w lustre.fail_loc=0x80000223"
1617 cp /etc/profile $DIR/$tfile || error "cp failed"
1619 cat $LPROC/ost/OSS/ost_create/timeouts
1621 CONN2=$(awk '/_connect/ {total+=$2} END {print total}' $LPROC/osc/*/stats)
1622 ATTEMPTS=$(($CONN2 - $CONN1))
1623 echo "$ATTEMPTS osc reconnect attemps on instant slow"
1624 # do it again; should not timeout
1625 do_facet ost1 "sysctl -w lustre.fail_loc=0x80000223"
1626 cp /etc/profile $DIR/$tfile || error "cp failed"
1627 do_facet ost1 "sysctl -w lustre.fail_loc=0"
1629 cat $LPROC/ost/OSS/ost_create/timeouts
1630 CONN3=$(awk '/_connect/ {total+=$2} END {print total}' $LPROC/osc/*/stats)
1631 ATTEMPTS=$(($CONN3 - $CONN2))
1632 echo "$ATTEMPTS osc reconnect attemps on 2nd slow"
1633 [ $ATTEMPTS -gt 0 ] && error "AT should have prevented reconnect"
1636 run_test 67b "AT: verify instant slowdown doesn't induce reconnects"
1638 test_68 () #bug 13813
1640 at_start || return 0
1641 local ldlm_enqueue_min=$(find /sys -name ldlm_enqueue_min)
1642 [ -z "$ldlm_enqueue_min" ] && skip "missing /sys/.../ldlm_enqueue_min" && return 0
1643 local ENQ_MIN=$(cat $ldlm_enqueue_min)
1644 echo $TIMEOUT >> $ldlm_enqueue_min
1645 rm -f $DIR/${tfile}_[1-2]
1646 lfs setstripe $DIR/$tfile --index=0 --count=1
1647 #define OBD_FAIL_LDLM_PAUSE_CANCEL 0x312
1648 sysctl -w lustre.fail_val=$(($TIMEOUT - 1))
1649 sysctl -w lustre.fail_loc=0x80000312
1650 cp /etc/profile $DIR/${tfile}_1 || error "1st cp failed $?"
1651 sysctl -w lustre.fail_val=$((TIMEOUT * 3 / 2))
1652 sysctl -w lustre.fail_loc=0x80000312
1653 cp /etc/profile $DIR/${tfile}_2 || error "2nd cp failed $?"
1654 sysctl -w lustre.fail_loc=0
1655 echo $ENQ_MIN >> $ldlm_enqueue_min
1658 run_test 68 "AT: verify slowing locks"
1660 if [ -n "$ATOLDBASE" ]; then
1661 at_history=$(do_facet mds "find /sys/ -name at_history")
1662 do_facet mds "echo $ATOLDBASE >> $at_history" || true
1663 do_facet ost1 "echo $ATOLDBASE >> $at_history" || true
1666 if [ $AT_MAX_SET -ne 0 ]; then
1667 for facet in mds client ost; do
1668 var=AT_MAX_SAVE_${facet}
1669 echo restore AT on $facet to saved value ${!var}
1670 at_max_set ${!var} $facet
1671 AT_NEW=$(at_max_get $facet)
1672 echo Restored AT value on $facet $AT_NEW
1673 [ $AT_NEW -ne ${!var} ] && \
1674 error "$facet : AT value was not restored SAVED ${!var} NEW $AT_NEW"
1678 # end of AT tests includes above lines
1681 # start multi-client tests
1683 [ -z "$CLIENTS" ] && \
1684 { skip "Need two or more clients." && return; }
1685 [ $CLIENTCOUNT -lt 2 ] && \
1686 { skip "Need two or more clients, have $CLIENTCOUNT" && return; }
1688 echo "mount clients $CLIENTS ..."
1689 zconf_mount_clients $CLIENTS $DIR
1691 local clients=${CLIENTS//,/ }
1692 echo "Write/read files on $DIR ; clients $CLIENTS ... "
1693 for CLIENT in $clients; do
1694 do_node $CLIENT dd bs=1M count=10 if=/dev/zero \
1695 of=$DIR/${tfile}_${CLIENT} 2>/dev/null || \
1696 error "dd failed on $CLIENT"
1699 local prev_client=$(echo $clients | sed 's/^.* \(.\+\)$/\1/')
1700 for C in ${CLIENTS//,/ }; do
1701 do_node $prev_client dd if=$DIR/${tfile}_${C} of=/dev/null 2>/dev/null || \
1702 error "dd if=$DIR/${tfile}_${C} failed on $prev_client"
1708 zconf_umount_clients $CLIENTS $DIR
1710 run_test 70a "check multi client t-f"
1713 [ -z "$CLIENTS" ] && \
1714 { skip "Need two or more clients." && return; }
1715 [ $CLIENTCOUNT -lt 2 ] && \
1716 { skip "Need two or more clients, have $CLIENTCOUNT" && return; }
1718 zconf_mount_clients $CLIENTS $DIR
1720 local duration="-t 60"
1721 local cmd="rundbench 1 $duration "
1723 for CLIENT in ${CLIENTS//,/ }; do
1724 $PDSH $CLIENT "set -x; PATH=:$PATH:$LUSTRE/utils:$LUSTRE/tests/:${DBENCH_LIB} DBENCH_LIB=${DBENCH_LIB} $cmd" &
1726 echo $PID >pid.$CLIENT
1727 echo "Started load PID=`cat pid.$CLIENT`"
1730 replay_barrier $SINGLEMDS
1731 sleep 3 # give clients a time to do operations
1733 log "$TESTNAME fail mds 1"
1736 # wait for client to reconnect to MDS
1739 for CLIENT in ${CLIENTS//,/ }; do
1740 PID=`cat pid.$CLIENT`
1743 echo "load on ${CLIENT} returned $rc"
1746 zconf_umount_clients $CLIENTS $DIR
1748 run_test 70b "mds recovery; $CLIENTCOUNT clients"
1749 # end multi-client tests
1751 equals_msg `basename $0`: test complete, cleaning up
1752 check_and_cleanup_lustre
1753 [ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG || true