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 wait $pid || return 3
743 [ -e $DIR/$tfile ] && return 1
747 run_test 34 "abort recovery before client does replay (test mds_cleanup_orphans)"
749 # bug 2278 - generate one orphan on OST, then destroy it during recovery from llog
753 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
754 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000119"
759 # give a chance to remove from MDS
760 fail_abort $SINGLEMDS
761 $CHECKSTAT -t file $DIR/$tfile && return 1 || true
763 run_test 35 "test recovery from llog for unlink op"
765 # b=2432 resent cancel after replay uses wrong cookie,
766 # so don't resend cancels
768 replay_barrier $SINGLEMDS
770 checkstat $DIR/$tfile
771 facet_failover $SINGLEMDS
773 if dmesg | grep "unknown lock cookie"; then
774 echo "cancel after replay failed"
778 run_test 36 "don't resend cancel"
781 # directory orphans can't be unlinked from PENDING directory
783 rmdir $DIR/$tfile 2>/dev/null
784 multiop_bg_pause $DIR/$tfile dD_c || return 2
788 replay_barrier $SINGLEMDS
789 # clear the dmesg buffer so we only see errors from this recovery
791 fail_abort $SINGLEMDS
793 dmesg | grep "mds_unlink_orphan.*error .* unlinking orphan" && return 1
797 run_test 37 "abort recovery before client does replay (test mds_cleanup_orphans for directories)"
800 createmany -o $DIR/$tfile-%d 800
801 unlinkmany $DIR/$tfile-%d 0 400
802 replay_barrier $SINGLEMDS
804 unlinkmany $DIR/$tfile-%d 400 400
806 $CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
808 run_test 38 "test recovery from unlink llog (test llog_gen_rec) "
810 test_39() { # bug 4176
811 createmany -o $DIR/$tfile-%d 800
812 replay_barrier $SINGLEMDS
813 unlinkmany $DIR/$tfile-%d 0 400
815 unlinkmany $DIR/$tfile-%d 400 400
817 $CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
819 run_test 39 "test recovery from unlink llog (test llog_gen_rec) "
822 lctl get_param -n osc.*.stats | awk -vwrites=0 '/ost_write/ { writes += $2 } END { print writes; }'
827 $LCTL mark multiop $MOUNT/$tfile OS_c
828 multiop $MOUNT/$tfile OS_c &
830 writeme -s $MOUNT/${tfile}-2 &
833 facet_failover $SINGLEMDS
834 #define OBD_FAIL_MDS_CONNECT_NET 0x117
835 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000117"
837 stat1=`count_ost_writes`
839 stat2=`count_ost_writes`
840 echo "$stat1, $stat2"
841 if [ $stat1 -lt $stat2 ]; then
842 echo "writes continuing during recovery"
845 echo "writes not continuing during recovery, bug 2477"
848 echo "waiting for writeme $WRITE_PID"
852 echo "waiting for multiop $PID"
853 wait $PID || return 2
854 do_facet client munlink $MOUNT/$tfile || return 3
855 do_facet client munlink $MOUNT/${tfile}-2 || return 3
858 run_test 40 "cause recovery in ptlrpc, ensure IO continues"
862 # make sure that a read to one osc doesn't try to double-unlock its page just
863 # because another osc is invalid. trigger_group_io used to mistakenly return
864 # an error if any oscs were invalid even after having successfully put rpcs
865 # on valid oscs. This was fatal if the caller was ll_readpage who unlocked
866 # the page, guarnateeing that the unlock from the RPC completion would
867 # assert on trying to unlock the unlocked page.
869 [ $OSTCOUNT -lt 2 ] && \
870 skip "skipping test 41: we don't have a second OST to test with" && \
873 local f=$MOUNT/$tfile
874 # make sure the start of the file is ost1
875 lfs setstripe $f -s $((128 * 1024)) -i 0
876 do_facet client dd if=/dev/zero of=$f bs=4k count=1 || return 3
878 # fail ost2 and read from ost1
879 local osc2dev=`do_facet mds "lctl get_param -n devices | grep ${ost2_svc}-osc-MDT0000" | awk '{print $1}'`
880 [ -z "$osc2dev" ] && echo "OST: $ost2_svc" && lctl get_param -n devices && return 4
881 do_facet mds $LCTL --device $osc2dev deactivate || return 1
882 do_facet client dd if=$f of=/dev/null bs=4k count=1 || return 3
883 do_facet mds $LCTL --device $osc2dev activate || return 2
886 run_test 41 "read from a valid osc while other oscs are invalid"
888 # test MDS recovery after ost failure
890 blocks=`df -P $MOUNT | tail -n 1 | awk '{ print $2 }'`
891 createmany -o $DIR/$tfile-%d 800
893 unlinkmany $DIR/$tfile-%d 0 400
895 lctl set_param debug=-1
898 # osc is evicted, fs is smaller (but only with failout OSTs (bug 7287)
899 #blocks_after=`df -P $MOUNT | tail -n 1 | awk '{ print $2 }'`
900 #[ $blocks_after -lt $blocks ] || return 1
901 echo wait for MDS to timeout and recover
902 sleep $((TIMEOUT * 2))
904 unlinkmany $DIR/$tfile-%d 400 400
905 $CHECKSTAT -t file $DIR/$tfile-* && return 2 || true
907 run_test 42 "recovery after ost failure"
909 # timeout in MDS/OST recovery RPC will LBUG MDS
910 test_43() { # bug 2530
911 replay_barrier $SINGLEMDS
913 # OBD_FAIL_OST_CREATE_NET 0x204
914 do_facet ost1 "lctl set_param fail_loc=0x80000204"
917 do_facet ost1 "lctl set_param fail_loc=0"
921 run_test 43 "mds osc import failure during recovery; don't LBUG"
923 test_44a() { # was test_44
926 mdcdev=`lctl get_param -n devices | awk '/MDT0000-mdc-/ {print $1}'`
927 [ "$mdcdev" ] || exit 2
929 # adaptive timeouts slow this way down
930 if at_is_valid && at_is_enabled; then
931 at_max_saved=$(at_max_get mds)
935 for i in `seq 1 10`; do
936 echo "$i of 10 ($(date +%s))"
937 do_facet mds "grep service $LPROC/mdt/MDS/mds/timeouts"
938 #define OBD_FAIL_TGT_CONN_RACE 0x701
939 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000701"
940 $LCTL --device $mdcdev recover
943 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
944 [ $at_max_saved -ne 0 ] && at_max_set $at_max_saved mds
947 run_test 44a "race in target handle connect"
950 mdcdev=`lctl get_param -n devices | awk '/MDT0000-mdc-/ {print $1}'`
951 [ "$mdcdev" ] || exit 2
952 for i in `seq 1 10`; do
953 echo "$i of 10 ($(date +%s))"
954 do_facet mds "grep service $LPROC/mdt/MDS/mds/timeouts"
955 #define OBD_FAIL_TGT_DELAY_RECONNECT 0x704
956 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000704"
957 $LCTL --device $mdcdev recover
960 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
963 run_test 44b "race in target handle connect"
965 # Handle failed close
967 mdcdev=`lctl get_param -n devices | awk '/MDT0000-mdc-/ {print $1}'`
968 [ "$mdcdev" ] || exit 2
969 $LCTL --device $mdcdev recover
971 multiop_bg_pause $DIR/$tfile O_c || return 1
974 # This will cause the CLOSE to fail before even
975 # allocating a reply buffer
976 $LCTL --device $mdcdev deactivate || return 4
980 wait $pid || return 1
982 $LCTL --device $mdcdev activate || return 5
985 $CHECKSTAT -t file $DIR/$tfile || return 2
988 run_test 45 "Handle failed close"
992 drop_reply "touch $DIR/$tfile"
994 # ironically, the previous test, 45, will cause a real forced close,
995 # so just look for one for this test
996 dmesg | grep -i "force closing client file handle for $tfile" && return 1
999 run_test 46 "Don't leak file handle after open resend (3325)"
1001 test_47() { # bug 2824
1002 # create some files to make sure precreate has been done on all
1003 # OSTs. (just in case this test is run independently)
1004 createmany -o $DIR/$tfile 20 || return 1
1006 # OBD_FAIL_OST_CREATE_NET 0x204
1008 do_facet ost1 "lctl set_param fail_loc=0x80000204"
1009 df $MOUNT || return 2
1011 # let the MDS discover the OST failure, attempt to recover, fail
1012 # and recover again.
1013 sleep $((3 * TIMEOUT))
1015 # Without 2824, this createmany would hang
1016 createmany -o $DIR/$tfile 20 || return 3
1017 unlinkmany $DIR/$tfile 20 || return 4
1019 do_facet ost1 "lctl set_param fail_loc=0"
1022 run_test 47 "MDS->OSC failure during precreate cleanup (2824)"
1025 replay_barrier $SINGLEMDS
1026 createmany -o $DIR/$tfile 20 || return 1
1027 # OBD_FAIL_OST_EROFS 0x216
1029 do_facet ost1 "lctl set_param fail_loc=0x80000216"
1030 df $MOUNT || return 2
1032 createmany -o $DIR/$tfile 20 20 || return 2
1033 unlinkmany $DIR/$tfile 40 || return 3
1035 do_facet ost1 "lctl set_param fail_loc=0"
1038 run_test 48 "MDS->OSC failure during precreate cleanup (2824)"
1041 local oscdev=`do_facet $SINGLEMDS lctl get_param -n devices | grep ${ost1_svc}-osc-MDT0000 | awk '{print $1}'`
1042 [ "$oscdev" ] || return 1
1043 do_facet $SINGLEMDS $LCTL --device $oscdev recover || return 2
1044 do_facet $SINGLEMDS $LCTL --device $oscdev recover || return 3
1045 # give the mds_lov_sync threads a chance to run
1048 run_test 50 "Double OSC recovery, don't LASSERT (3812)"
1050 # b3764 timed out lock replay
1053 cancel_lru_locks mdc
1055 multiop $DIR/$tfile s || return 1
1056 replay_barrier $SINGLEMDS
1057 #define OBD_FAIL_LDLM_REPLY 0x30c
1058 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000030c"
1059 fail $SINGLEMDS || return 2
1060 do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
1062 $CHECKSTAT -t file $DIR/$tfile-* && return 3 || true
1064 run_test 52 "time out lock replay (3764)"
1066 # bug 3462 - simultaneous MDC requests
1068 mkdir -p $DIR/${tdir}-1
1069 mkdir -p $DIR/${tdir}-2
1070 multiop $DIR/${tdir}-1/f O_c &
1072 # give multiop a change to open
1075 #define OBD_FAIL_MDS_CLOSE_NET 0x115
1076 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
1077 kill -USR1 $close_pid
1078 cancel_lru_locks mdc # force the close
1079 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1081 mcreate $DIR/${tdir}-2/f || return 1
1083 # close should still be here
1084 [ -d /proc/$close_pid ] || return 2
1086 replay_barrier_nodf $SINGLEMDS
1088 wait $close_pid || return 3
1090 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 4
1091 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 5
1092 rm -rf $DIR/${tdir}-*
1094 run_test 53a "|X| close request while two MDC requests in flight"
1097 mkdir -p $DIR/${tdir}-1
1098 mkdir -p $DIR/${tdir}-2
1099 multiop $DIR/${tdir}-1/f O_c &
1102 #define OBD_FAIL_MDS_REINT_NET 0x107
1103 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
1104 mcreate $DIR/${tdir}-2/f &
1108 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1109 kill -USR1 $close_pid
1110 cancel_lru_locks mdc # force the close
1111 wait $close_pid || return 1
1112 # open should still be here
1113 [ -d /proc/$open_pid ] || return 2
1115 replay_barrier_nodf $SINGLEMDS
1117 wait $open_pid || return 3
1119 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 4
1120 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 5
1121 rm -rf $DIR/${tdir}-*
1123 run_test 53b "|X| open request while two MDC requests in flight"
1126 mkdir -p $DIR/${tdir}-1
1127 mkdir -p $DIR/${tdir}-2
1128 multiop $DIR/${tdir}-1/f O_c &
1131 #define OBD_FAIL_MDS_REINT_NET 0x107
1132 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
1133 mcreate $DIR/${tdir}-2/f &
1137 #define OBD_FAIL_MDS_CLOSE_NET 0x115
1138 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
1139 kill -USR1 $close_pid
1140 cancel_lru_locks mdc # force the close
1142 replay_barrier_nodf $SINGLEMDS
1143 fail_nodf $SINGLEMDS
1144 wait $open_pid || return 1
1146 # close should be gone
1147 [ -d /proc/$close_pid ] && return 2
1148 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1150 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 3
1151 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 4
1152 rm -rf $DIR/${tdir}-*
1154 run_test 53c "|X| open request and close request while two MDC requests in flight"
1157 mkdir -p $DIR/${tdir}-1
1158 mkdir -p $DIR/${tdir}-2
1159 multiop $DIR/${tdir}-1/f O_c &
1161 # give multiop a chance to open
1164 #define OBD_FAIL_MDS_CLOSE_NET_REP 0x13f
1165 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000013f"
1166 kill -USR1 $close_pid
1167 cancel_lru_locks mdc # force the close
1168 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1169 mcreate $DIR/${tdir}-2/f || return 1
1171 # close should still be here
1172 [ -d /proc/$close_pid ] || return 2
1174 wait $close_pid || return 3
1176 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 4
1177 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 5
1178 rm -rf $DIR/${tdir}-*
1180 run_test 53d "|X| close reply while two MDC requests in flight"
1183 mkdir -p $DIR/${tdir}-1
1184 mkdir -p $DIR/${tdir}-2
1185 multiop $DIR/${tdir}-1/f O_c &
1188 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
1189 do_facet $SINGLEMDS "lctl set_param fail_loc=0x119"
1190 mcreate $DIR/${tdir}-2/f &
1194 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1195 kill -USR1 $close_pid
1196 cancel_lru_locks mdc # force the close
1197 wait $close_pid || return 1
1198 # open should still be here
1199 [ -d /proc/$open_pid ] || return 2
1201 replay_barrier_nodf $SINGLEMDS
1203 wait $open_pid || return 3
1205 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 4
1206 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 5
1207 rm -rf $DIR/${tdir}-*
1209 run_test 53e "|X| open reply while two MDC requests in flight"
1212 mkdir -p $DIR/${tdir}-1
1213 mkdir -p $DIR/${tdir}-2
1214 multiop $DIR/${tdir}-1/f O_c &
1217 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
1218 do_facet $SINGLEMDS "lctl set_param fail_loc=0x119"
1219 mcreate $DIR/${tdir}-2/f &
1223 #define OBD_FAIL_MDS_CLOSE_NET_REP 0x13f
1224 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000013f"
1225 kill -USR1 $close_pid
1226 cancel_lru_locks mdc # force the close
1228 replay_barrier_nodf $SINGLEMDS
1229 fail_nodf $SINGLEMDS
1230 wait $open_pid || return 1
1232 # close should be gone
1233 [ -d /proc/$close_pid ] && return 2
1234 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1236 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 3
1237 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 4
1238 rm -rf $DIR/${tdir}-*
1240 run_test 53f "|X| open reply and close reply while two MDC requests in flight"
1243 mkdir -p $DIR/${tdir}-1
1244 mkdir -p $DIR/${tdir}-2
1245 multiop $DIR/${tdir}-1/f O_c &
1248 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
1249 do_facet $SINGLEMDS "lctl set_param fail_loc=0x119"
1250 mcreate $DIR/${tdir}-2/f &
1254 #define OBD_FAIL_MDS_CLOSE_NET 0x115
1255 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
1256 kill -USR1 $close_pid
1257 cancel_lru_locks mdc # force the close
1259 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1260 replay_barrier_nodf $SINGLEMDS
1261 fail_nodf $SINGLEMDS
1262 wait $open_pid || return 1
1264 # close should be gone
1265 [ -d /proc/$close_pid ] && return 2
1267 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 3
1268 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 4
1269 rm -rf $DIR/${tdir}-*
1271 run_test 53g "|X| drop open reply and close request while close and open are both in flight"
1274 mkdir -p $DIR/${tdir}-1
1275 mkdir -p $DIR/${tdir}-2
1276 multiop $DIR/${tdir}-1/f O_c &
1279 #define OBD_FAIL_MDS_REINT_NET 0x107
1280 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
1281 mcreate $DIR/${tdir}-2/f &
1285 #define OBD_FAIL_MDS_CLOSE_NET_REP 0x13f
1286 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000013f"
1287 kill -USR1 $close_pid
1288 cancel_lru_locks mdc # force the close
1291 replay_barrier_nodf $SINGLEMDS
1292 fail_nodf $SINGLEMDS
1293 wait $open_pid || return 1
1295 # close should be gone
1296 [ -d /proc/$close_pid ] && return 2
1297 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1299 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 3
1300 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 4
1301 rm -rf $DIR/${tdir}-*
1303 run_test 53h "|X| open request and close reply while two MDC requests in flight"
1305 #b_cray 54 "|X| open request and close reply while two MDC requests in flight"
1307 #b3761 ASSERTION(hash != 0) failed
1309 # OBD_FAIL_MDS_OPEN_CREATE | OBD_FAIL_ONCE
1310 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000012b"
1312 # give touch a chance to run
1314 do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
1318 run_test 55 "let MDS_CHECK_RESENT return the original return code instead of 0"
1320 #b3440 ASSERTION(rec->ur_fid2->id) failed
1322 ln -s foo $DIR/$tfile
1323 replay_barrier $SINGLEMDS
1324 #drop_reply "cat $DIR/$tfile"
1328 run_test 56 "don't replay a symlink open request (3440)"
1330 #recovery one mds-ost setattr from llog
1332 #define OBD_FAIL_MDS_OST_SETATTR 0x12c
1333 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000012c"
1335 replay_barrier $SINGLEMDS
1338 $CHECKSTAT -t file $DIR/$tfile || return 1
1339 do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
1342 run_test 57 "test recovery from llog for setattr op"
1344 #recovery many mds-ost setattr from llog
1347 #define OBD_FAIL_MDS_OST_SETATTR 0x12c
1348 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000012c"
1349 createmany -o $DIR/$tdir/$tfile-%d 2500
1350 replay_barrier $SINGLEMDS
1353 $CHECKSTAT -t file $DIR/$tdir/$tfile-* >/dev/null || return 1
1354 do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
1355 unlinkmany $DIR/$tdir/$tfile-%d 2500
1358 run_test 58 "test recovery from llog for setattr op (test llog_gen_rec)"
1362 touch $DIR/$tdir/$tfile
1363 replay_barrier $SINGLEMDS
1364 setfattr -n trusted.foo -v bar $DIR/$tdir/$tfile
1366 VAL=`getfattr --absolute-names --only-value -n trusted.foo $DIR/$tdir/$tfile`
1367 [ x$VAL = x"bar" ] || return 1
1368 rm -f $DIR/$tdir/$tfile
1371 run_test 58a "test replay of setxattr op"
1373 # log_commit_thread vs filter_destroy race used to lead to import use after free
1377 createmany -o $DIR/$tdir/$tfile-%d 200
1379 unlinkmany $DIR/$tdir/$tfile-%d 200
1380 #define OBD_FAIL_PTLRPC_DELAY_RECOV 0x507
1381 do_facet ost1 "lctl set_param fail_loc=0x507"
1384 do_facet ost1 "lctl set_param fail_loc=0x0"
1388 run_test 59 "test log_commit_thread vs filter_destroy race"
1390 # race between add unlink llog vs cat log init in post_recovery (only for b1_6)
1391 # bug 12086: should no oops and No ctxt error for this test
1394 createmany -o $DIR/$tdir/$tfile-%d 200
1395 replay_barrier $SINGLEMDS
1396 unlinkmany $DIR/$tdir/$tfile-%d 0 100
1398 unlinkmany $DIR/$tdir/$tfile-%d 100 100
1399 local no_ctxt=`dmesg | grep "No ctxt"`
1400 [ -z "$no_ctxt" ] || error "ctxt is not initialized in recovery"
1402 run_test 60 "test llog post recovery init vs llog unlink"
1404 #test race llog recovery thread vs llog cleanup
1405 test_61a() { # was test_61
1407 createmany -o $DIR/$tdir/$tfile-%d 800
1409 # OBD_FAIL_OST_LLOG_RECOVERY_TIMEOUT 0x221
1410 unlinkmany $DIR/$tdir/$tfile-%d 800
1411 set_nodes_failloc "$(osts_nodes)" 0x80000221
1416 set_nodes_failloc "$(osts_nodes)" 0x0
1418 $CHECKSTAT -t file $DIR/$tdir/$tfile-* && return 1
1421 run_test 61a "test race llog recovery vs llog cleanup"
1423 #test race mds llog sync vs llog cleanup
1425 # OBD_FAIL_MDS_LLOG_SYNC_TIMEOUT 0x140
1426 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000140"
1427 facet_failover $SINGLEMDS
1430 do_facet client dd if=/dev/zero of=$DIR/$tfile bs=4k count=1 || return 1
1432 run_test 61b "test race mds llog sync vs llog cleanup"
1434 #test race cancel cookie cb vs llog cleanup
1436 # OBD_FAIL_OST_CANCEL_COOKIE_TIMEOUT 0x222
1438 set_nodes_failloc "$(osts_nodes)" 0x80000222
1442 set_nodes_failloc "$(osts_nodes)" 0x0
1444 run_test 61c "test race mds llog sync vs llog cleanup"
1446 test_62() { # Bug 15756 - don't mis-drop resent replay
1448 replay_barrier $SINGLEMDS
1449 createmany -o $DIR/$tdir/$tfile- 25
1450 #define OBD_FAIL_TGT_REPLAY_DROP 0x706
1451 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000707"
1452 facet_failover $SINGLEMDS
1453 df $MOUNT || return 1
1454 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1455 unlinkmany $DIR/$tdir/$tfile- 25 || return 2
1458 run_test 62 "don't mis-drop resent replay"
1460 #Adaptive Timeouts (bug 3055)
1465 if ! at_is_valid; then
1466 skip "AT env is invalid"
1470 if ! at_is_enabled; then
1471 echo "AT is disabled, enable it by force temporarily"
1472 at_max_set 600 mds ost client
1476 if [ -z "$ATOLDBASE" ]; then
1477 local at_history=$(do_facet mds "find /sys/ -name at_history")
1478 [ -z "$at_history" ] && skip "missing /sys/.../at_history " && return 1
1479 ATOLDBASE=$(do_facet mds "cat $at_history")
1480 # speed up the timebase so we can check decreasing AT
1481 do_facet mds "echo 8 >> $at_history"
1482 do_facet ost1 "echo 8 >> $at_history"
1486 test_65a() #bug 3055
1488 at_start || return 0
1489 $LCTL dk > /dev/null
1491 sysctl -w lnet.debug="+other"
1492 # slow down a request
1493 do_facet mds sysctl -w lustre.fail_val=30000
1494 #define OBD_FAIL_PTLRPC_PAUSE_REQ 0x50a
1495 do_facet mds sysctl -w lustre.fail_loc=0x8000050a
1496 createmany -o $DIR/$tfile 10 > /dev/null
1497 unlinkmany $DIR/$tfile 10 > /dev/null
1498 # check for log message
1499 $LCTL dk | grep "Early reply #" || error "No early reply"
1501 # client should show 30s estimates
1502 grep portal $LPROC/mdc/${FSNAME}-MDT0000-mdc-*/timeouts
1504 grep portal $LPROC/mdc/${FSNAME}-MDT0000-mdc-*/timeouts
1506 run_test 65a "AT: verify early replies"
1508 test_65b() #bug 3055
1510 at_start || return 0
1513 sysctl -w lnet.debug="other trace"
1514 $LCTL dk > /dev/null
1515 # slow down bulk i/o
1516 do_facet ost1 sysctl -w lustre.fail_val=30
1517 #define OBD_FAIL_OST_BRW_PAUSE_PACK 0x224
1518 do_facet ost1 sysctl -w lustre.fail_loc=0x224
1521 lfs setstripe $DIR/$tfile --index=0 --count=1
1522 # force some real bulk transfer
1523 multiop $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c
1525 do_facet ost1 sysctl -w lustre.fail_loc=0
1526 # check for log message
1527 $LCTL dk | grep "Early reply #" || error "No early reply"
1529 # client should show 30s estimates
1530 grep portal $LPROC/osc/${FSNAME}-OST0000-osc-*/timeouts
1532 run_test 65b "AT: verify early replies on packed reply / bulk"
1534 test_66a() #bug 3055
1536 at_start || return 0
1537 grep "portal 12" $LPROC/mdc/${FSNAME}-MDT0000-mdc-*/timeouts
1538 # adjust 5s at a time so no early reply is sent (within deadline)
1539 do_facet mds "sysctl -w lustre.fail_val=5000"
1540 #define OBD_FAIL_PTLRPC_PAUSE_REQ 0x50a
1541 do_facet mds "sysctl -w lustre.fail_loc=0x8000050a"
1542 createmany -o $DIR/$tfile 20 > /dev/null
1543 unlinkmany $DIR/$tfile 20 > /dev/null
1544 grep "portal 12" $LPROC/mdc/${FSNAME}-MDT0000-mdc-*/timeouts
1545 do_facet mds "sysctl -w lustre.fail_val=10000"
1546 do_facet mds "sysctl -w lustre.fail_loc=0x8000050a"
1547 createmany -o $DIR/$tfile 20 > /dev/null
1548 unlinkmany $DIR/$tfile 20 > /dev/null
1549 grep "portal 12" $LPROC/mdc/${FSNAME}-MDT0000-mdc-*/timeouts
1550 do_facet mds "sysctl -w lustre.fail_loc=0"
1552 createmany -o $DIR/$tfile 20 > /dev/null
1553 unlinkmany $DIR/$tfile 20 > /dev/null
1554 grep portal $LPROC/mdc/${FSNAME}-MDT0000-mdc-*/timeouts | grep "portal 12"
1555 CUR=$(awk '/portal 12/ {print $5}' $LPROC/mdc/${FSNAME}-MDT0000-mdc-*/timeouts)
1556 WORST=$(awk '/portal 12/ {print $7}' $LPROC/mdc/${FSNAME}-MDT0000-mdc-*/timeouts)
1557 echo "Current MDT timeout $CUR, worst $WORST"
1558 [ $CUR -lt $WORST ] || error "Current $CUR should be less than worst $WORST"
1560 run_test 66a "AT: verify MDT service time adjusts with no early replies"
1562 test_66b() #bug 3055
1564 at_start || return 0
1565 ORIG=$(awk '/network/ {print $4}' $LPROC/mdc/lustre-*/timeouts)
1566 sysctl -w lustre.fail_val=$(($ORIG + 5))
1567 #define OBD_FAIL_PTLRPC_PAUSE_REP 0x50c
1568 sysctl -w lustre.fail_loc=0x50c
1569 ls $DIR/$tfile > /dev/null 2>&1
1570 sysctl -w lustre.fail_loc=0
1571 CUR=$(awk '/network/ {print $4}' $LPROC/mdc/${FSNAME}-*/timeouts)
1572 WORST=$(awk '/network/ {print $6}' $LPROC/mdc/${FSNAME}-*/timeouts)
1573 echo "network timeout orig $ORIG, cur $CUR, worst $WORST"
1574 [ $WORST -gt $ORIG ] || error "Worst $WORST should be worse than orig $ORIG"
1576 run_test 66b "AT: verify net latency adjusts"
1578 test_67a() #bug 3055
1580 at_start || return 0
1581 CONN1=$(awk '/_connect/ {total+=$2} END {print total}' $LPROC/osc/*/stats)
1582 # sleeping threads may drive values above this
1583 do_facet ost1 "sysctl -w lustre.fail_val=400"
1584 #define OBD_FAIL_PTLRPC_PAUSE_REQ 0x50a
1585 do_facet ost1 "sysctl -w lustre.fail_loc=0x50a"
1586 createmany -o $DIR/$tfile 20 > /dev/null
1587 unlinkmany $DIR/$tfile 20 > /dev/null
1588 do_facet ost1 "sysctl -w lustre.fail_loc=0"
1589 CONN2=$(awk '/_connect/ {total+=$2} END {print total}' $LPROC/osc/*/stats)
1590 ATTEMPTS=$(($CONN2 - $CONN1))
1591 echo "$ATTEMPTS osc reconnect attemps on gradual slow"
1592 [ $ATTEMPTS -gt 0 ] && error_ignore 13721 "AT should have prevented reconnect"
1595 run_test 67a "AT: verify slow request processing doesn't induce reconnects"
1597 test_67b() #bug 3055
1599 at_start || return 0
1600 CONN1=$(awk '/_connect/ {total+=$2} END {print total}' $LPROC/osc/*/stats)
1601 #define OBD_FAIL_OST_PAUSE_CREATE 0x223
1602 do_facet ost1 "sysctl -w lustre.fail_val=20000"
1603 do_facet ost1 "sysctl -w lustre.fail_loc=0x80000223"
1604 cp /etc/profile $DIR/$tfile || error "cp failed"
1606 cat $LPROC/ost/OSS/ost_create/timeouts
1608 CONN2=$(awk '/_connect/ {total+=$2} END {print total}' $LPROC/osc/*/stats)
1609 ATTEMPTS=$(($CONN2 - $CONN1))
1610 echo "$ATTEMPTS osc reconnect attemps on instant slow"
1611 # do it again; should not timeout
1612 do_facet ost1 "sysctl -w lustre.fail_loc=0x80000223"
1613 cp /etc/profile $DIR/$tfile || error "cp failed"
1614 do_facet ost1 "sysctl -w lustre.fail_loc=0"
1616 cat $LPROC/ost/OSS/ost_create/timeouts
1617 CONN3=$(awk '/_connect/ {total+=$2} END {print total}' $LPROC/osc/*/stats)
1618 ATTEMPTS=$(($CONN3 - $CONN2))
1619 echo "$ATTEMPTS osc reconnect attemps on 2nd slow"
1620 [ $ATTEMPTS -gt 0 ] && error "AT should have prevented reconnect"
1623 run_test 67b "AT: verify instant slowdown doesn't induce reconnects"
1625 test_68 () #bug 13813
1627 at_start || return 0
1628 local ldlm_enqueue_min=$(find /sys -name ldlm_enqueue_min)
1629 [ -z "$ldlm_enqueue_min" ] && skip "missing /sys/.../ldlm_enqueue_min" && return 0
1630 local ENQ_MIN=$(cat $ldlm_enqueue_min)
1631 echo $TIMEOUT >> $ldlm_enqueue_min
1632 rm -f $DIR/${tfile}_[1-2]
1633 lfs setstripe $DIR/$tfile --index=0 --count=1
1634 #define OBD_FAIL_LDLM_PAUSE_CANCEL 0x312
1635 sysctl -w lustre.fail_val=$(($TIMEOUT - 1))
1636 sysctl -w lustre.fail_loc=0x80000312
1637 cp /etc/profile $DIR/${tfile}_1 || error "1st cp failed $?"
1638 sysctl -w lustre.fail_val=$((TIMEOUT * 3 / 2))
1639 sysctl -w lustre.fail_loc=0x80000312
1640 cp /etc/profile $DIR/${tfile}_2 || error "2nd cp failed $?"
1641 sysctl -w lustre.fail_loc=0
1642 echo $ENQ_MIN >> $ldlm_enqueue_min
1645 run_test 68 "AT: verify slowing locks"
1647 if [ -n "$ATOLDBASE" ]; then
1648 at_history=$(do_facet mds "find /sys/ -name at_history")
1649 do_facet mds "echo $ATOLDBASE >> $at_history" || true
1650 do_facet ost1 "echo $ATOLDBASE >> $at_history" || true
1653 if [ $AT_MAX_SET -ne 0 ]; then
1654 echo "restore AT status to be disabled"
1655 at_max_set 0 mds ost client
1658 # end of AT tests includes above lines
1661 # start multi-client tests
1663 [ -z "$CLIENTS" ] && \
1664 { skip "Need two or more clients." && return; }
1665 [ $CLIENTCOUNT -lt 2 ] && \
1666 { skip "Need two or more clients, have $CLIENTCOUNT" && return; }
1668 echo "mount clients $CLIENTS ..."
1669 zconf_mount_clients $CLIENTS $DIR
1671 local clients=${CLIENTS//,/ }
1672 echo "Write/read files on $DIR ; clients $CLIENTS ... "
1673 for CLIENT in $clients; do
1674 do_node $CLIENT dd bs=1M count=10 if=/dev/zero \
1675 of=$DIR/${tfile}_${CLIENT} 2>/dev/null || \
1676 error "dd failed on $CLIENT"
1679 local prev_client=$(echo $clients | sed 's/^.* \(.\+\)$/\1/')
1680 for C in ${CLIENTS//,/ }; do
1681 do_node $prev_client dd if=$DIR/${tfile}_${C} of=/dev/null 2>/dev/null || \
1682 error "dd if=$DIR/${tfile}_${C} failed on $prev_client"
1688 zconf_umount_clients $CLIENTS $DIR
1690 run_test 70a "check multi client t-f"
1693 [ -z "$CLIENTS" ] && \
1694 { skip "Need two or more clients." && return; }
1695 [ $CLIENTCOUNT -lt 2 ] && \
1696 { skip "Need two or more clients, have $CLIENTCOUNT" && return; }
1698 zconf_mount_clients $CLIENTS $DIR
1700 local duration="-t 60"
1701 local cmd="rundbench 1 $duration "
1703 for CLIENT in ${CLIENTS//,/ }; do
1704 $PDSH $CLIENT "set -x; PATH=:$PATH:$LUSTRE/utils:$LUSTRE/tests/:${DBENCH_LIB} DBENCH_LIB=${DBENCH_LIB} $cmd" &
1706 echo $PID >pid.$CLIENT
1707 echo "Started load PID=`cat pid.$CLIENT`"
1710 replay_barrier $SINGLEMDS
1711 sleep 3 # give clients a time to do operations
1713 log "$TESTNAME fail mds 1"
1716 # wait for client to reconnect to MDS
1719 for CLIENT in ${CLIENTS//,/ }; do
1720 PID=`cat pid.$CLIENT`
1723 echo "load on ${CLIENT} returned $rc"
1726 zconf_umount_clients $CLIENTS $DIR
1728 run_test 70b "mds recovery; $CLIENTCOUNT clients"
1729 # end multi-client tests
1731 equals_msg `basename $0`: test complete, cleaning up
1732 check_and_cleanup_lustre
1733 [ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG || true