3 # Run select tests by setting ONLY, or as arguments to the script.
4 # Skip specific tests by setting EXCEPT.
10 ALWAYS_EXCEPT="$SANITY_SCRUB_EXCEPT"
12 [ "$SLOW" = "no" ] && EXCEPT_SLOW=""
13 # UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT!
15 LUSTRE=${LUSTRE:-$(cd $(dirname $0)/..; echo $PWD)}
16 . $LUSTRE/tests/test-framework.sh
18 . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
21 require_dsh_mds || exit 0
25 if ! check_versions; then
26 skip "It is NOT necessary to test scrub under interoperation mode"
30 [ $(facet_fstype $SINGLEMDS) != "ldiskfs" ] &&
31 skip "test OI scrub only for ldiskfs" && exit 0
33 [ $(facet_fstype ost1) != "ldiskfs" ] &&
34 skip "test OI scrub only for ldiskfs" && exit 0
36 [[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.2.90) ]] &&
37 skip "Need MDS version at least 2.2.90" && exit 0
39 SAVED_MDSSIZE=${MDSSIZE}
40 SAVED_OSTSIZE=${OSTSIZE}
41 SAVED_OSTCOUNT=${OSTCOUNT}
42 # use small MDS + OST size to speed formatting time
43 # do not use too small MDSSIZE/OSTSIZE, which affect the default journal size
44 # 200M MDT device can guarantee uninitialized groups during the OI scrub
47 # no need too much OSTs, to reduce the format/start/stop overhead
49 [ $OSTCOUNT -gt 4 ] && OSTCOUNT=4
53 # build up a clean test environment.
57 [[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.3.90) ]] &&
58 ALWAYS_EXCEPT="$ALWAYS_EXCEPT 1a"
60 [[ $(lustre_version_code $SINGLEMDS) -le $(version_code 2.6.50) ]] &&
61 ALWAYS_EXCEPT="$ALWAYS_EXCEPT 4"
63 [[ $(lustre_version_code $SINGLEMDS) -le $(version_code 2.4.1) ]] &&
64 ALWAYS_EXCEPT="$ALWAYS_EXCEPT 15"
66 [[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.4.90) ]] &&
67 [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.4.50) ]] &&
68 ALWAYS_EXCEPT="$ALWAYS_EXCEPT 15"
70 [[ $(lustre_version_code ost1) -lt $(version_code 2.4.50) ]] &&
71 ALWAYS_EXCEPT="$ALWAYS_EXCEPT 11 12 13 14"
73 [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.5.59) ]] &&
78 MDT_DEV="${FSNAME}-MDT0000"
79 OST_DEV="${FSNAME}-OST0000"
80 MDT_DEVNAME=$(mdsdevname ${SINGLEMDS//mds/})
86 # use "lfsck_start -A" when we no longer need testing interop
87 for n in $(seq $MDSCOUNT); do
88 do_facet mds$n $LCTL lfsck_start -M $(facet_svc mds$n) \
90 error "($error_id) Failed to start OI scrub on mds$n"
98 # use "lfsck_stop -A" when we no longer need testing interop
99 for n in $(seq $MDSCOUNT); do
100 do_facet mds$n $LCTL lfsck_stop -M $(facet_svc mds$n) ||
101 error "($error_id) Failed to stop OI scrub on mds$n"
108 do_facet mds$n $LCTL get_param -n \
109 osd-ldiskfs.$(facet_svc mds$n).oi_scrub
112 START_SCRUB="do_facet $SINGLEMDS $LCTL lfsck_start -M ${MDT_DEV} $SCRUB_ONLY"
113 START_SCRUB_ON_OST="do_facet ost1 $LCTL lfsck_start -M ${OST_DEV} $SCRUB_ONLY"
114 STOP_SCRUB="do_facet $SINGLEMDS $LCTL lfsck_stop -M ${MDT_DEV}"
115 SHOW_SCRUB="do_facet $SINGLEMDS \
116 $LCTL get_param -n osd-ldiskfs.${MDT_DEV}.oi_scrub"
117 SHOW_SCRUB_ON_OST="do_facet ost1 \
118 $LCTL get_param -n osd-ldiskfs.${OST_DEV}.oi_scrub"
119 MOUNT_OPTS_SCRUB="-o user_xattr"
120 MOUNT_OPTS_NOSCRUB="-o user_xattr,noscrub"
128 echo "preparing... $(date)"
129 for n in $(seq $MDSCOUNT); do
130 echo "creating $nfiles files on mds$n"
131 test_mkdir -i $((n - 1)) $DIR/$tdir/mds$n ||
132 error "Failed to create directory mds$n"
133 cp $LUSTRE/tests/*.sh $DIR/$tdir/mds$n ||
134 error "Failed to copy files to mds$n"
135 mkdir -p $DIR/$tdir/mds$n/d_$tfile ||
136 error "mkdir failed on mds$n"
137 createmany -m $DIR/$tdir/mds$n/d_$tfile/f 2 > \
138 /dev/null || error "create failed on mds$n"
139 if [[ $nfiles -gt 0 ]]; then
140 createmany -m $DIR/$tdir/mds$n/$tfile $nfiles > \
141 /dev/null || error "createmany failed on mds$n"
144 echo "prepared $(date)."
145 cleanup_mount $MOUNT > /dev/null || error "Fail to stop client!"
146 for n in $(seq $MDSCOUNT); do
148 stop mds$n > /dev/null || error "Fail to stop MDS$n!"
157 for n in $(seq $MDSCOUNT); do
158 start mds$n $(mdsdevname $n) $opts >/dev/null ||
159 error "($error_id) Failed to start mds$n"
167 for n in $(seq $MDSCOUNT); do
168 echo "stopping mds$n"
169 stop mds$n >/dev/null ||
170 error "($error_id) Failed to stop mds$n"
174 scrub_check_status() {
179 for n in $(seq $MDSCOUNT); do
180 wait_update_facet mds$n "$LCTL get_param -n \
181 osd-ldiskfs.$(facet_svc mds$n).oi_scrub |
182 awk '/^status/ { print \\\$2 }'" "$expected" 6 ||
183 error "($error_id) Expected '$expected' on mds$n"
187 scrub_check_flags() {
193 for n in $(seq $MDSCOUNT); do
194 actual=$(do_facet mds$n $LCTL get_param -n \
195 osd-ldiskfs.$(facet_svc mds$n).oi_scrub |
196 awk '/^flags/ { print $2 }')
197 if [ "$actual" != "$expected" ]; then
198 error "($error_id) Expected '$expected' on mds$n, but" \
204 scrub_check_params() {
210 for n in $(seq $MDSCOUNT); do
211 actual=$(do_facet mds$n $LCTL get_param -n \
212 osd-ldiskfs.$(facet_svc mds$n).oi_scrub |
213 awk '/^param/ { print $2 }')
214 if [ "$actual" != "$expected" ]; then
215 error "($error_id) Expected '$expected' on mds$n, but" \
221 scrub_check_repaired() {
227 for n in $(seq $MDSCOUNT); do
228 actual=$(do_facet mds$n $LCTL get_param -n \
229 osd-ldiskfs.$(facet_svc mds$n).oi_scrub |
230 awk '/^updated/ { print $2 }')
232 if [ $expected -eq 0 -a $actual -ne 0 ]; then
233 error "($error_id) Expected no repaired on mds$n, but" \
237 if [ $expected -ne 0 -a $actual -lt $expected ]; then
238 error "($error_id) Expected '$expected' on mds$n, but" \
248 for n in $(seq $MDSCOUNT); do
249 diff -q $LUSTRE/tests/test-framework.sh \
250 $DIR/$tdir/mds$n/test-framework.sh ||
251 error "($error_id) File data check failed"
255 scrub_check_data2() {
260 for n in $(seq $MDSCOUNT); do
261 diff -q $LUSTRE/tests/$filename \
262 $DIR/$tdir/mds$n/$filename ||
263 error "($error_id) File data check failed"
272 for n in $(seq $MDSCOUNT); do
273 mds_remove_ois mds$n $index ||
274 error "($error_id) Failed to remove OI .$index on mds$n"
278 scrub_backup_restore() {
283 for n in $(seq $MDSCOUNT); do
284 mds_backup_restore mds$n $igif ||
285 error "($error_id) Backup/restore on mds$n failed"
289 scrub_enable_auto() {
290 do_nodes $(comma_list $(mdts_nodes)) $LCTL set_param -n \
291 osd-ldiskfs.*.auto_scrub=1
295 [[ $(lustre_version_code $SINGLEMDS) -le $(version_code 2.6.50) ]] &&
300 do_nodes $(comma_list $(mdts_nodes)) $LCTL set_param -n \
301 osd-ldiskfs.*.full_scrub_ratio=$ratio
304 full_scrub_threshold_rate() {
305 [[ $(lustre_version_code $SINGLEMDS) -le $(version_code 2.6.50) ]] &&
310 do_nodes $(comma_list $(mdts_nodes)) $LCTL set_param -n \
311 osd-ldiskfs.*.full_scrub_threshold_rate=$rate
316 echo "starting MDTs without disabling OI scrub"
317 scrub_start_mds 1 "$MOUNT_OPTS_SCRUB"
318 scrub_check_status 2 init
319 scrub_check_flags 3 ""
320 mount_client $MOUNT || error "(4) Fail to start client!"
323 run_test 0 "Do not auto trigger OI scrub for non-backup/restore case"
327 echo "start $SINGLEMDS without disabling OI scrub"
328 scrub_start_mds 1 "$MOUNT_OPTS_SCRUB"
330 local FLAGS=$($SHOW_SCRUB | awk '/^flags/ { print $2 }')
331 [ -z "$FLAGS" ] || error "(3) Expect empty flags, but got '$FLAGS'"
333 mount_client $MOUNT || error "(4) Fail to start client!"
334 #define OBD_FAIL_OSD_FID_MAPPING 0x193
335 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x193
336 # update .lustre OI mapping
338 do_facet $SINGLEMDS $LCTL set_param fail_loc=0
339 umount_client $MOUNT || error "(5) Fail to stop client!"
341 echo "stop $SINGLEMDS"
342 stop $SINGLEMDS > /dev/null || error "(6) Fail to stop MDS!"
344 echo "start $SINGLEMDS with disabling OI scrub"
345 start $SINGLEMDS $MDT_DEVNAME $MOUNT_OPTS_NOSCRUB > /dev/null ||
346 error "(7) Fail to start MDS!"
348 local FLAGS=$($SHOW_SCRUB | awk '/^flags/ { print $2 }')
349 [ "$FLAGS" == "inconsistent" ] ||
350 error "(9) Expect 'inconsistent', but got '$FLAGS'"
352 run_test 1a "Auto trigger initial OI scrub when server mounts"
357 echo "start MDTs without disabling OI scrub"
358 scrub_start_mds 2 "$MOUNT_OPTS_SCRUB"
359 scrub_check_status 3 completed
360 mount_client $MOUNT || error "(4) Fail to start client!"
363 run_test 1b "Trigger OI scrub when MDT mounts for OI files remove/recreate case"
368 # OI files to be removed:
370 # idx 2: oi.16.{2,4,8,16,32}
371 # idx 3: oi.16.{3,9,27}
372 for index in 0 2 3; do
374 scrub_remove_ois 1 $index
375 echo "start MDTs with OI scrub disabled"
376 scrub_start_mds 2 "$MOUNT_OPTS_NOSCRUB"
377 scrub_check_flags 3 recreated
379 scrub_check_status 5 completed
380 scrub_check_flags 6 ""
383 run_test 1c "Auto detect kinds of OI file(s) removed/recreated cases"
387 scrub_backup_restore 1
388 echo "starting MDTs without disabling OI scrub"
389 scrub_start_mds 2 "$MOUNT_OPTS_SCRUB"
390 scrub_check_status 3 completed
391 mount_client $MOUNT || error "(4) Fail to start client!"
394 run_test 2 "Trigger OI scrub when MDT mounts for backup/restore case"
396 # test_3 is obsolete, it will be covered by test_5.
398 formatall > /dev/null
402 scrub_backup_restore 1
403 echo "starting MDTs with OI scrub disabled"
404 scrub_start_mds 2 "$MOUNT_OPTS_NOSCRUB"
405 scrub_check_status 3 init
406 scrub_check_flags 4 recreated,inconsistent
408 #run_test 3 "Do not trigger OI scrub when MDT mounts if 'noscrub' specified"
412 scrub_backup_restore 1
413 echo "starting MDTs with OI scrub disabled"
414 scrub_start_mds 2 "$MOUNT_OPTS_NOSCRUB"
415 scrub_check_flags 4 recreated,inconsistent
416 mount_client $MOUNT || error "(5) Fail to start client!"
422 scrub_check_status 7 completed
423 scrub_check_flags 8 ""
426 for n in $(seq $MDSCOUNT); do
427 updated0[$n]=$(scrub_status $n |
428 awk '/^prior_updated/ { print $2 }')
431 scrub_check_data2 sanity-scrub.sh 9
435 for n in $(seq $MDSCOUNT); do
436 updated1[$n]=$(scrub_status $n |
437 awk '/^prior_updated/ { print $2 }')
438 [ ${updated0[$n]} -eq ${updated1[$n]} ] ||
439 error "(10) NOT auto trigger full scrub as expected"
442 run_test 4a "Auto trigger OI scrub if bad OI mapping was found (1)"
446 scrub_backup_restore 1
447 echo "starting MDTs with OI scrub disabled"
448 scrub_start_mds 2 "$MOUNT_OPTS_NOSCRUB"
449 scrub_check_flags 4 recreated,inconsistent
450 mount_client $MOUNT || error "(5) Fail to start client!"
453 full_scrub_threshold_rate 10000
457 scrub_check_status 7 completed
458 scrub_check_flags 8 ""
461 for n in $(seq $MDSCOUNT); do
462 updated0[$n]=$(scrub_status $n |
463 awk '/^prior_updated/ { print $2 }')
466 scrub_check_data2 sanity-scrub.sh 9
469 scrub_check_status 10 completed
470 scrub_check_flags 11 ""
473 for n in $(seq $MDSCOUNT); do
474 updated1[$n]=$(scrub_status $n |
475 awk '/^prior_updated/ { print $2 }')
476 [ ${updated0[$n]} -lt ${updated1[$n]} ] ||
477 error "(12) Auto trigger full scrub unexpectedly"
480 for n in $(seq $MDSCOUNT); do
481 ls -l $DIR/$tdir/mds$n/*.sh > /dev/null ||
482 error "(13) fail to ls"
486 scrub_check_status 14 completed
487 scrub_check_flags 15 ""
489 for n in $(seq $MDSCOUNT); do
490 updated0[$n]=$(scrub_status $n |
491 awk '/^prior_updated/ { print $2 }')
492 [ ${updated0[$n]} -gt ${updated1[$n]} ] ||
493 error "(16) Auto trigger full scrub unexpectedly"
496 for n in $(seq $MDSCOUNT); do
497 ls -l $DIR/$tdir/mds$n/d_${tfile}/ || error "(17) fail to ls"
501 for n in $(seq $MDSCOUNT); do
502 updated1[$n]=$(scrub_status $n |
503 awk '/^prior_updated/ { print $2 }')
504 [ ${updated0[$n]} -eq ${updated1[$n]} ] ||
505 error "(18) NOT auto trigger full scrub as expected"
508 run_test 4b "Auto trigger OI scrub if bad OI mapping was found (2)"
512 scrub_backup_restore 1
513 echo "starting MDTs with OI scrub disabled"
514 scrub_start_mds 2 "$MOUNT_OPTS_NOSCRUB"
515 scrub_check_flags 4 recreated,inconsistent
516 mount_client $MOUNT || error "(5) Fail to start client!"
519 full_scrub_threshold_rate 20
523 scrub_check_status 7 completed
524 scrub_check_flags 8 ""
527 for n in $(seq $MDSCOUNT); do
528 updated0[$n]=$(scrub_status $n |
529 awk '/^prior_updated/ { print $2 }')
532 scrub_check_data2 sanity-scrub.sh 9
535 scrub_check_status 10 completed
536 scrub_check_flags 11 ""
539 for n in $(seq $MDSCOUNT); do
540 updated1[$n]=$(scrub_status $n |
541 awk '/^prior_updated/ { print $2 }')
542 [ ${updated0[$n]} -lt ${updated1[$n]} ] ||
543 error "(12) Auto trigger full scrub unexpectedly"
546 for n in $(seq $MDSCOUNT); do
547 ls -l $DIR/$tdir/mds$n/*.sh > /dev/null ||
548 error "(13) fail to ls"
552 scrub_check_status 14 completed
553 scrub_check_flags 15 ""
555 for n in $(seq $MDSCOUNT); do
556 updated0[$n]=$(scrub_status $n |
557 awk '/^prior_updated/ { print $2 }')
558 [ ${updated0[$n]} -gt ${updated1[$n]} ] ||
559 error "(16) Auto trigger full scrub unexpectedly"
562 for n in $(seq $MDSCOUNT); do
563 ls -l $DIR/$tdir/mds$n/${tfile}1 || error "(17) fail to ls"
567 for n in $(seq $MDSCOUNT); do
568 updated1[$n]=$(scrub_status $n |
569 awk '/^prior_updated/ { print $2 }')
570 [ ${updated0[$n]} -eq ${updated1[$n]} ] ||
571 error "(18) NOT auto trigger full scrub as expected"
574 run_test 4c "Auto trigger OI scrub if bad OI mapping was found (3)"
577 formatall > /dev/null
581 scrub_backup_restore 1
582 echo "starting MDTs with OI scrub disabled (1)"
583 scrub_start_mds 2 "$MOUNT_OPTS_NOSCRUB"
584 scrub_check_status 3 init
585 scrub_check_flags 4 recreated,inconsistent
586 mount_client $MOUNT || error "(5) Fail to start client!"
590 #define OBD_FAIL_OSD_SCRUB_DELAY 0x190
591 do_nodes $(comma_list $(mdts_nodes)) \
592 $LCTL set_param fail_val=3 fail_loc=0x190
595 umount_client $MOUNT || error "(7) Fail to stop client!"
596 scrub_check_status 8 scanning
598 #define OBD_FAIL_OSD_SCRUB_CRASH 0x191
599 do_nodes $(comma_list $(mdts_nodes)) $LCTL set_param fail_loc=0x191
604 do_nodes $(comma_list $(mdts_nodes)) \
605 $LCTL set_param fail_loc=0 fail_val=0
607 echo "starting MDTs with OI scrub disabled (2)"
608 scrub_start_mds 10 "$MOUNT_OPTS_NOSCRUB"
609 scrub_check_status 11 crashed
612 #define OBD_FAIL_OSD_SCRUB_DELAY 0x190
613 do_nodes $(comma_list $(mdts_nodes)) \
614 $LCTL set_param fail_val=3 fail_loc=0x190
616 echo "starting MDTs without disabling OI scrub"
617 scrub_start_mds 13 "$MOUNT_OPTS_SCRUB"
618 scrub_check_status 14 scanning
620 #define OBD_FAIL_OSD_SCRUB_FATAL 0x192
621 do_nodes $(comma_list $(mdts_nodes)) $LCTL set_param fail_loc=0x192
623 scrub_check_status 15 failed
624 mount_client $MOUNT || error "(16) Fail to start client!"
627 #define OBD_FAIL_OSD_SCRUB_DELAY 0x190
628 do_nodes $(comma_list $(mdts_nodes)) \
629 $LCTL set_param fail_val=3 fail_loc=0x190
632 for n in $(seq $MDSCOUNT); do
633 stat $DIR/$tdir/mds$n/${tfile}800 ||
634 error "(17) Failed to stat mds$n/${tfile}800"
637 scrub_check_status 18 scanning
639 do_nodes $(comma_list $(mdts_nodes)) \
640 $LCTL set_param fail_loc=0 fail_val=0
642 scrub_check_status 19 completed
643 scrub_check_flags 20 ""
645 run_test 5 "OI scrub state machine"
649 scrub_backup_restore 1
650 echo "starting MDTs with OI scrub disabled"
651 scrub_start_mds 2 "$MOUNT_OPTS_NOSCRUB"
652 scrub_check_flags 4 recreated,inconsistent
653 mount_client $MOUNT || error "(5) Fail to start client!"
657 #define OBD_FAIL_OSD_SCRUB_DELAY 0x190
658 do_nodes $(comma_list $(mdts_nodes)) \
659 $LCTL set_param fail_val=2 fail_loc=0x190
663 # Sleep 5 sec to guarantee at least one object processed by OI scrub
665 # Fail the OI scrub to guarantee there is at least one checkpoint
666 #define OBD_FAIL_OSD_SCRUB_FATAL 0x192
667 do_nodes $(comma_list $(mdts_nodes)) $LCTL set_param fail_loc=0x192
669 scrub_check_status 7 failed
671 #define OBD_FAIL_OSD_SCRUB_DELAY 0x190
672 do_nodes $(comma_list $(mdts_nodes)) \
673 $LCTL set_param fail_val=3 fail_loc=0x190
676 for n in $(seq $MDSCOUNT); do
677 # stat will re-trigger OI scrub
678 stat $DIR/$tdir/mds$n/${tfile}800 ||
679 error "(8) Failed to stat mds$n/${tfile}800"
682 umount_client $MOUNT || error "(9) Fail to stop client!"
683 scrub_check_status 10 scanning
685 #define OBD_FAIL_OSD_SCRUB_CRASH 0x191
686 do_nodes $(comma_list $(mdts_nodes)) $LCTL set_param fail_loc=0x191
690 for n in $(seq $MDSCOUNT); do
691 position0[$n]=$(scrub_status $n |
692 awk '/^last_checkpoint_position/ {print $2}')
693 position0[$n]=$((${position0[$n]} + 1))
698 #define OBD_FAIL_OSD_SCRUB_DELAY 0x190
699 do_nodes $(comma_list $(mdts_nodes)) \
700 $LCTL set_param fail_val=3 fail_loc=0x190
702 echo "starting MDTs without disabling OI scrub"
703 scrub_start_mds 12 "$MOUNT_OPTS_SCRUB"
705 scrub_check_status 13 scanning
708 for n in $(seq $MDSCOUNT); do
709 position1[$n]=$(scrub_status $n |
710 awk '/^latest_start_position/ {print $2}')
711 if [ ${position0[$n]} -ne ${position1[$n]} ]; then
712 error "(14) Expected position ${position0[$n]}, but" \
713 "got ${position1[$n]}"
717 do_nodes $(comma_list $(mdts_nodes)) \
718 $LCTL set_param fail_loc=0 fail_val=0
720 scrub_check_status 15 completed
721 scrub_check_flags 16 ""
723 run_test 6 "OI scrub resumes from last checkpoint"
727 scrub_backup_restore 1
728 echo "starting MDTs with OI scrub disabled"
729 scrub_start_mds 2 "$MOUNT_OPTS_NOSCRUB"
730 scrub_check_flags 4 recreated,inconsistent
731 mount_client $MOUNT || error "(5) Fail to start client!"
735 #define OBD_FAIL_OSD_SCRUB_DELAY 0x190
736 do_nodes $(comma_list $(mdts_nodes)) \
737 $LCTL set_param fail_val=3 fail_loc=0x190
742 for n in $(seq $MDSCOUNT); do
743 stat $DIR/$tdir/mds$n/${tfile}300 ||
744 error "(7) Failed to stat mds$n/${tfile}300!"
747 scrub_check_status 8 scanning
748 scrub_check_flags 9 recreated,inconsistent,auto
750 do_nodes $(comma_list $(mdts_nodes)) \
751 $LCTL set_param fail_loc=0 fail_val=0
753 scrub_check_status 10 completed
756 run_test 7 "System is available during OI scrub scanning"
760 scrub_backup_restore 1
761 echo "starting MDTs with OI scrub disabled"
762 scrub_start_mds 2 "$MOUNT_OPTS_NOSCRUB"
763 scrub_check_flags 4 recreated,inconsistent
765 #define OBD_FAIL_OSD_SCRUB_DELAY 0x190
766 do_nodes $(comma_list $(mdts_nodes)) \
767 $LCTL set_param fail_val=1 fail_loc=0x190
770 scrub_check_status 6 scanning
772 scrub_check_status 8 stopped
774 scrub_check_status 10 scanning
776 do_nodes $(comma_list $(mdts_nodes)) \
777 $LCTL set_param fail_loc=0 fail_val=0
779 scrub_check_status 11 completed
780 scrub_check_flags 12 ""
782 run_test 8 "Control OI scrub manually"
785 if [ -z "$(grep "processor.*: 1" /proc/cpuinfo)" ]; then
786 skip "Testing on UP system, the speed may be inaccurate."
791 scrub_backup_restore 1
793 echo "starting MDTs with OI scrub disabled"
794 scrub_start_mds 2 "$MOUNT_OPTS_NOSCRUB"
795 scrub_check_flags 4 recreated,inconsistent
797 local BASE_SPEED1=100
799 # OI scrub should run with full speed under inconsistent case
800 scrub_start 5 -s $BASE_SPEED1
803 scrub_check_status 6 completed
804 scrub_check_flags 7 ""
805 # OI scrub should run with limited speed under non-inconsistent case
806 scrub_start 8 -s $BASE_SPEED1 -r
809 scrub_check_status 9 scanning
811 # Do NOT ignore that there are 1024 pre-fetched items. And there
812 # may be time error, normally it should be less than 2 seconds.
813 # We allow another 20% schedule error.
814 local PRE_FETCHED=1024
816 # MAX_MARGIN = 1.2 = 12 / 10
817 local MAX_SPEED=$(((PRE_FETCHED + BASE_SPEED1 * \
818 (RUN_TIME1 + TIME_DIFF)) / RUN_TIME1 * 12 / 10))
820 for n in $(seq $MDSCOUNT); do
821 local SPEED=$(scrub_status $n | \
822 awk '/^average_speed/ { print $2 }')
823 [ $SPEED -lt $MAX_SPEED ] ||
824 error "(10) Got speed $SPEED, expected less than" \
829 local BASE_SPEED2=300
831 for n in $(seq $MDSCOUNT); do
832 do_facet mds$n $LCTL set_param -n \
833 mdd.$(facet_svc mds$n).lfsck_speed_limit $BASE_SPEED2
837 # MIN_MARGIN = 0.8 = 8 / 10
838 local MIN_SPEED=$(((PRE_FETCHED + \
839 BASE_SPEED1 * (RUN_TIME1 - TIME_DIFF) + \
840 BASE_SPEED2 * (RUN_TIME2 - TIME_DIFF)) / \
841 (RUN_TIME1 + RUN_TIME2) * 8 / 10))
842 # MAX_MARGIN = 1.2 = 12 / 10
843 MAX_SPEED=$(((PRE_FETCHED + \
844 BASE_SPEED1 * (RUN_TIME1 + TIME_DIFF) + \
845 BASE_SPEED2 * (RUN_TIME2 + TIME_DIFF)) / \
846 (RUN_TIME1 + RUN_TIME2) * 12 / 10))
847 for n in $(seq $MDSCOUNT); do
848 SPEED=$(scrub_status $n | awk '/^average_speed/ { print $2 }')
849 [ $SPEED -gt $MIN_SPEED ] ||
850 error "(11) Got speed $SPEED, expected more than" \
852 [ $SPEED -lt $MAX_SPEED ] ||
853 error "(12) Got speed $SPEED, expected less than" \
856 do_facet mds$n $LCTL set_param -n \
857 mdd.$(facet_svc mds$n).lfsck_speed_limit 0
860 scrub_check_status 13 completed
862 run_test 9 "OI scrub speed control"
866 scrub_backup_restore 1
867 echo "starting mds$n with OI scrub disabled (1)"
868 scrub_start_mds 2 "$MOUNT_OPTS_NOSCRUB"
869 scrub_check_flags 4 recreated,inconsistent
870 mount_client $MOUNT || error "(5) Fail to start client!"
874 #define OBD_FAIL_OSD_SCRUB_DELAY 0x190
875 do_nodes $(comma_list $(mdts_nodes)) \
876 $LCTL set_param fail_val=1 fail_loc=0x190
879 scrub_check_status 7 scanning
880 umount_client $MOUNT || error "(8) Fail to stop client!"
882 echo "starting MDTs with OI scrub disabled (2)"
883 scrub_start_mds 10 "$MOUNT_OPTS_NOSCRUB"
884 scrub_check_status 11 paused
886 echo "starting MDTs without disabling OI scrub"
887 scrub_start_mds 13 "$MOUNT_OPTS_SCRUB"
888 scrub_check_status 14 scanning
890 do_nodes $(comma_list $(mdts_nodes)) \
891 $LCTL set_param fail_loc=0 fail_val=0
893 scrub_check_status 15 completed
894 scrub_check_flags 16 ""
896 run_test 10a "non-stopped OI scrub should auto restarts after MDS remount (1)"
898 # test_10b is obsolete, it will be coverded by related sanity-lfsck tests.
901 scrub_backup_restore 1
902 echo "starting MDTs with OI scrub disabled"
903 scrub_start_mds 2 "$MOUNT_OPTS_NOSCRUB"
904 scrub_check_flags 4 recreated,inconsistent
906 #define OBD_FAIL_OSD_SCRUB_DELAY 0x190
907 do_nodes $(comma_list $(mdts_nodes)) \
908 $LCTL set_param fail_val=3 fail_loc=0x190
911 scrub_check_status 6 scanning
913 echo "starting MDTs with OI scrub disabled"
914 scrub_start_mds 8 "$MOUNT_OPTS_NOSCRUB"
915 scrub_check_status 9 paused
917 echo "starting MDTs without disabling OI scrub"
918 scrub_start_mds 11 "$MOUNT_OPTS_SCRUB"
919 scrub_check_status 12 scanning
921 do_nodes $(comma_list $(mdts_nodes)) \
922 $LCTL set_param fail_loc=0 fail_val=0
924 scrub_check_status 13 completed
925 scrub_check_flags 14 ""
927 #run_test 10b "non-stopped OI scrub should auto restarts after MDS remount (2)"
935 for n in $(seq $MDSCOUNT); do
936 test_mkdir -i $((n - 1)) $DIR/$tdir/mds$n ||
937 error "(1) Fail to mkdir $DIR/$tdir/mds$n"
939 createmany -o $DIR/$tdir/mds$n/f $CREATED ||
940 error "(2) Fail to create under $tdir/mds$n"
943 # reset OI scrub start point by force
945 scrub_check_status 4 completed
950 # OI scrub should skip the new created objects for the first accessing
951 # notice we're creating a new llog for every OST on every startup
952 # new features can make this even less stable, so we only check that
953 # the number of skipped files is more than the number or known created
954 local MINIMUM=$((CREATED + 1)) # files + directory
955 for n in $(seq $MDSCOUNT); do
956 local SKIPPED=$(scrub_status $n | awk '/^noscrub/ { print $2 }')
957 [ $SKIPPED -lt $MINIMUM ] &&
958 error "(5) Expect at least $MINIMUM objects" \
959 "skipped on mds$n, but got $SKIPPED"
961 checked0[$n]=$(scrub_status $n | awk '/^checked/ { print $2 }')
964 # reset OI scrub start point by force
966 scrub_check_status 7 completed
968 # OI scrub should skip the new created object only once
969 for n in $(seq $MDSCOUNT); do
970 SKIPPED=$(scrub_status $n | awk '/^noscrub/ { print $2 }')
971 checked1[$n]=$(scrub_status $n | awk '/^checked/ { print $2 }')
973 [ ${checked0[$n]} -ne ${checked1[$n]} -o $SKIPPED -eq 0 ] ||
974 error "(8) Expect 0 objects skipped on mds$n, but" \
978 run_test 11 "OI scrub skips the new created objects only once"
982 $SETSTRIPE -c 1 -i 0 $DIR/$tdir
984 #define OBD_FAIL_OSD_COMPAT_INVALID_ENTRY 0x195
985 do_facet ost1 $LCTL set_param fail_loc=0x195
986 local count=$(precreated_ost_obj_count 0 0)
988 createmany -o $DIR/$tdir/f $((count + 32))
989 umount_client $MOUNT || error "(1) Fail to stop client!"
991 stop ost1 || error "(2) Fail to stop ost1"
993 #define OBD_FAIL_OST_NODESTROY 0x233
994 do_facet ost1 $LCTL set_param fail_loc=0x233
996 start ost1 $(ostdevname 1) $MOUNT_OPTS_NOSCRUB ||
997 error "(3) Fail to start ost1"
999 mount_client $MOUNT || error "(4) Fail to start client!"
1001 ls -ail $DIR/$tdir > /dev/null 2>&1 && error "(5) ls should fail"
1003 $START_SCRUB_ON_OST -r || error "(6) Fail to start OI scrub on OST!"
1005 do_facet ost1 $LCTL set_param fail_loc=0
1006 wait_update_facet ost1 "$LCTL get_param -n \
1007 osd-ldiskfs.$(facet_svc ost1).oi_scrub |
1008 awk '/^status/ { print \\\$2 }'" "completed" 6 ||
1009 error "(7) Expected '$expected' on ost1"
1011 ls -ail $DIR/$tdir > /dev/null || {
1013 error "(8) ls should succeed"
1016 run_test 12 "OI scrub can rebuild invalid /O entries"
1019 check_mount_and_prep
1020 $SETSTRIPE -c 1 -i 0 $DIR/$tdir
1022 #define OBD_FAIL_OSD_COMPAT_NO_ENTRY 0x196
1023 do_facet ost1 $LCTL set_param fail_loc=0x196
1024 local count=$(precreated_ost_obj_count 0 0)
1026 createmany -o $DIR/$tdir/f $((count + 32))
1027 do_facet ost1 $LCTL set_param fail_loc=0
1029 umount_client $MOUNT || error "(1) Fail to stop client!"
1031 stop ost1 || error "(2) Fail to stop ost1"
1033 start ost1 $(ostdevname 1) $MOUNT_OPTS_NOSCRUB ||
1034 error "(3) Fail to start ost1"
1036 mount_client $MOUNT || error "(4) Fail to start client!"
1038 ls -ail $DIR/$tdir > /dev/null 2>&1 && error "(5) ls should fail"
1040 $START_SCRUB_ON_OST -r || error "(6) Fail to start OI scrub on OST!"
1042 wait_update_facet ost1 "$LCTL get_param -n \
1043 osd-ldiskfs.$(facet_svc ost1).oi_scrub |
1044 awk '/^status/ { print \\\$2 }'" "completed" 6 ||
1045 error "(7) Expected '$expected' on ost1"
1047 ls -ail $DIR/$tdir > /dev/null || error "(8) ls should succeed"
1049 run_test 13 "OI scrub can rebuild missed /O entries"
1052 check_mount_and_prep
1053 $SETSTRIPE -c 1 -i 0 $DIR/$tdir
1055 #define OBD_FAIL_OSD_COMPAT_NO_ENTRY 0x196
1056 do_facet ost1 $LCTL set_param fail_loc=0x196
1057 local count=$(precreated_ost_obj_count 0 0)
1059 createmany -o $DIR/$tdir/f $((count + 32))
1060 do_facet ost1 $LCTL set_param fail_loc=0
1062 umount_client $MOUNT || error "(1) Fail to stop client!"
1064 stop ost1 || error "(2) Fail to stop ost1"
1067 run_e2fsck $(facet_host ost1) $(ostdevname 1) "-y" ||
1068 error "(3) Fail to run e2fsck error"
1070 start ost1 $(ostdevname 1) $OST_MOUNT_OPTS ||
1071 error "(4) Fail to start ost1"
1073 mount_client $MOUNT || error "(5) Fail to start client!"
1075 local LF_REPAIRED=$($SHOW_SCRUB_ON_OST |
1076 awk '/^lf_repa[ri]*ed/ { print $2 }')
1077 [ $LF_REPAIRED -gt 0 ] ||
1078 error "(6) Some entry under /lost+found should be repaired"
1080 ls -ail $DIR/$tdir > /dev/null || error "(7) ls should succeed"
1082 run_test 14 "OI scrub can repair objects under lost+found"
1085 local server_version=$(lustre_version_code $SINGLEMDS)
1087 scrub_backup_restore 1
1088 echo "starting MDTs with OI scrub disabled"
1089 scrub_start_mds 2 "$MOUNT_OPTS_NOSCRUB"
1090 scrub_check_status 3 init
1091 scrub_check_flags 4 recreated,inconsistent
1093 # run under dryrun mode
1094 if [ $server_version -lt $(version_code 2.5.58) ]; then
1095 scrub_start 5 --dryrun on
1097 scrub_start 5 --dryrun
1099 scrub_check_status 6 completed
1100 scrub_check_flags 7 recreated,inconsistent
1101 scrub_check_params 8 dryrun
1102 scrub_check_repaired 9 20
1104 # run under dryrun mode again
1105 if [ $server_version -lt $(version_code 2.5.58) ]; then
1106 scrub_start 10 --dryrun on
1108 scrub_start 10 --dryrun
1110 scrub_check_status 11 completed
1111 scrub_check_flags 12 recreated,inconsistent
1112 scrub_check_params 13 dryrun
1113 scrub_check_repaired 14 20
1115 # run under normal mode
1117 # Lustre-2.x (x <= 5) used "-n off" to disable dryrun which does not
1118 # work under Lustre-2.y (y >= 6), the test script should be fixed as
1119 # "-noff" or "--dryrun=off" or nothing by default.
1120 if [ $server_version -lt $(version_code 2.5.58) ]; then
1121 scrub_start 15 --dryrun off
1125 scrub_check_status 16 completed
1126 scrub_check_flags 17 ""
1127 scrub_check_params 18 ""
1128 scrub_check_repaired 19 20
1130 # run under normal mode again
1131 if [ $server_version -lt $(version_code 2.5.58) ]; then
1132 scrub_start 20 --dryrun off
1136 scrub_check_status 21 completed
1137 scrub_check_flags 22 ""
1138 scrub_check_params 23 ""
1139 scrub_check_repaired 24 0
1141 run_test 15 "Dryrun mode OI scrub"
1143 # restore MDS/OST size
1144 MDSSIZE=${SAVED_MDSSIZE}
1145 OSTSIZE=${SAVED_OSTSIZE}
1146 OSTCOUNT=${SAVED_OSTCOUNT}
1148 # cleanup the system at last