3 # Run select tests by setting ONLY, or as arguments to the script.
4 # Skip specific tests by setting EXCEPT.
11 LREPL_LOG=$TMP/lustre_rsync.log
14 LUSTRE=${LUSTRE:-$(dirname $0)/..}
15 . $LUSTRE/tests/test-framework.sh
19 ALWAYS_EXCEPT="$LRSYNC_EXCEPT "
20 # bug number for skipped test: LU-4256
22 # UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT!
26 [ -n "$FILESET" ] && skip "Not functional for FILESET set"
28 check_and_setup_lustre
33 if getent group nobody; then
35 elif getent group nogroup; then
38 error "No generic nobody group"
41 export LRSYNC=${LRSYNC:-"$LUSTRE/utils/lustre_rsync"}
42 [ ! -f "$LRSYNC" ] && export LRSYNC=$(which lustre_rsync)
43 export LRSYNC="$LRSYNC -v -c no -d 2"
45 # Number of seconds to run dbench
46 DBENCH_TIME=${DBENCH_TIME:-60}
49 MDT0=$($LCTL get_param -n mdc.*.mds_server_uuid |
50 awk '{ gsub(/_UUID/,""); print $1 }' | head -n1)
53 changelog_register || error "changelog_register failed"
54 CL_USER=(${CL_USERS[$SINGLEMDS]})
55 echo $MDT0: Registered changelog user $CL_USER
57 echo "No changelog users present on $SINGLEMDS"
66 rm -rf $TGT/$tdir $TGT/d*.lustre_rsync-test 2> /dev/null
67 rm -rf $TGT2/$tdir $TGT2/d*.lustre_rsync-test 2> /dev/null
68 rm -rf ${DIR}/$tdir $DIR/d*.lustre_rsync-test ${DIR}/tgt 2> /dev/null
71 mkdir_on_mdt0 $DIR/$tdir || error "Failed to create target: " $DIR/$tdir
72 mkdir -p ${TGT}/$tdir || error "Failed to create target: " $TGT/$tdir
73 mkdir -p ${TGT2}/$tdir || error "Failed to create target: " $TGT2/$tdir
77 [ ! -d $TGT2/$tdir ] || rm -rf $TGT2/$tdir # /tmp/target2/...
78 [ ! -d $TGT/$tdir ] || rm -rf $TGT/$tdir # /tmp/target/...
79 rm -rf $DIR/$tdir # /mnt/lustre/...
80 rm -rf $DIR/tgt # /mnt/lustre/tgt
83 # Check whether the filesystem supports xattr or not.
85 # "large" - large xattr is supported
86 # "small" - large xattr is unsupported but small xattr is supported
87 # "no" - xattr is unsupported
94 local val="$(generate_string $(max_xattr_size))"
96 if large_xattr_enabled &&
97 setfattr -n user.foo -v $val $tgt 2>/dev/null; then
100 setfattr -n user.foo -v bar $tgt 2>/dev/null && xattr="small"
108 local changelog_file=$(generate_logname "changelog")
110 if [ -e $1 -o -e $2 ]; then
111 diff -rq -x "dev1" $1 $2
113 if [ $RC -ne 0 ]; then
114 $LFS changelog $MDT0 > $changelog_file
115 error "Failure in replication; differences found."
120 procs_are_stopped() {
124 for state in $(ps -p "$pids" -o state=); do
125 if [[ "$state" != T ]]; then
133 # Send SIGSTOP to PIDs and wait up to 60 seconds for them to show a
134 # stopped process state.
140 end=$((SECONDS + 60))
141 while ((SECONDS < end)); do
142 if procs_are_stopped $pids; then
152 # Test 1A - test basic operations
153 test_1A() { # was test_1
156 local xattr=$(check_xattr $TGT/foo)
163 touch $DIR/$tdir/file1
164 cp /etc/hosts $DIR/$tdir/d1/
165 touch $DIR/$tdir/d1/"space in filename"
166 touch $DIR/$tdir/d1/file2
169 mv $DIR/$tdir/d1/file2 $DIR/$tdir/d2/file3
171 # File and directory delete
172 touch $DIR/$tdir/d1/file4
173 mkdir $DIR/$tdir/d1/del
174 touch $DIR/$tdir/d1/del/del1
175 touch $DIR/$tdir/d1/del/del2
176 rm -rf $DIR/$tdir/d1/del
177 rm $DIR/$tdir/d1/file4
179 # Hard and soft links
180 cat /etc/hosts > $DIR/$tdir/d1/link1
181 ln $DIR/$tdir/d1/link1 $DIR/$tdir/d1/link2
182 ln -s $DIR/$tdir/d1/link1 $DIR/$tdir/d1/link3
185 #mknod $DIR/$tdir/dev1 b 8 1
188 local LRSYNC_LOG=$(generate_logname "lrsync_log")
189 echo "Replication #1"
190 $LRSYNC -s $DIR -t $TGT -t $TGT2 -m $MDT0 -u $CL_USER -l $LREPL_LOG \
194 chmod 000 $DIR/$tdir/d2/file3
195 chown nobody:$GROUP $DIR/$tdir/d2/file3
198 if [[ "$xattr" != "no" ]]; then
200 touch $DIR/$tdir/file5
201 [[ "$xattr" = "large" ]] &&
202 value="$(generate_string $(max_xattr_size))" || value="bar"
203 setfattr -n user.foo -v $value $DIR/$tdir/file5 ||
204 error "setfattr failed"
207 echo "Replication #2"
208 $LRSYNC -l $LREPL_LOG -D $LRSYNC_LOG
210 if [[ "$xattr" != "no" ]]; then
211 local xval1=$(get_xattr_value user.foo $TGT/$tdir/file5)
212 local xval2=$(get_xattr_value user.foo $TGT2/$tdir/file5)
214 if [[ "$xval1" != "$value" || "$xval2" != "$value" ]]; then
215 error "Error in replicating xattrs."
219 # Use diff to compare the source and the destination
220 check_diff $DIR/$tdir $TGT/$tdir
221 check_diff $DIR/$tdir $TGT2/$tdir
226 run_test 1A "Simple Replication"
228 # Test 1a - test create/delete operations in ROOT directory
229 test_1a() { # LU-5005
230 rm -rf $TGT/root-* 2> /dev/null
231 rm -rf $DIR/root-* 2> /dev/null
236 mkdir_on_mdt0 $DIR/root-dir
240 touch $DIR/root-file2
243 mv $DIR/root-file2 $DIR/root-file3
245 # File and directory delete
246 touch $DIR/root-file4
247 mkdir_on_mdt0 $DIR/root-dir1
249 rm -rf $DIR/root-dir1
252 local LRSYNC_LOG=$(generate_logname "lrsync_log")
254 $LRSYNC -s $DIR -t $TGT -m $MDT0 -u $CL_USER -l $LREPL_LOG \
258 stat $TGT/root-dir || error "Dir create not replicated"
259 stat $TGT/root-file || error "File create not replicated"
260 stat $TGT/root-file2 && error "Rename not replicated (src)"
261 stat $TGT/root-file3 || error "Rename not replicated (tgt)"
262 stat $TGT/root-dir1 && error "Dir delete not replicated"
263 stat $TGT/root-file4 && error "File delete not replicated"
271 run_test 1a "Replicate create/delete operations in ROOT directory"
273 # Test 2a - Replicate files created by dbench
279 bash rundbench -C -D $DIR/$tdir 2 -t $DBENCH_TIME ||
280 error "dbench failed"
282 local LRSYNC_LOG=$(generate_logname "lrsync_log")
283 # Replicate the changes to $TGT
284 $LRSYNC -s $DIR -t $TGT -t $TGT2 -m $MDT0 -u $CL_USER -l $LREPL_LOG \
287 # Use diff to compare the source and the destination
288 check_diff $DIR/$tdir $TGT/$tdir
289 check_diff $DIR/$tdir $TGT2/$tdir
295 run_test 2a "Replicate files created by dbench."
297 # Test 2b - Replicate files changed by dbench.
304 bash rundbench -C -D $DIR/$tdir 2 -t $DBENCH_TIME &
305 # wait for dbench to start
306 wait_for_function 'child_pid=$(pgrep dbench)' 360
307 # let dbench run for a bit
310 echo PIDs: $child_pid
312 stop_procs $child_pid
314 local LRSYNC_LOG=$(generate_logname "lrsync_log")
315 echo Starting replication
316 $LRSYNC -s $DIR -t $TGT -t $TGT2 -m $MDT0 -u $CL_USER -l $LREPL_LOG \
318 check_diff $DIR/$tdir $TGT/$tdir
321 $KILL -SIGCONT $child_pid
325 stop_procs $child_pid
327 echo Starting replication
328 $LRSYNC -l $LREPL_LOG -D $LRSYNC_LOG
329 check_diff $DIR/$tdir $TGT/$tdir
331 echo "Wait for dbench to finish"
332 $KILL -SIGCONT $child_pid
335 # Replicate the changes to $TGT
336 echo Starting replication
337 $LRSYNC -l $LREPL_LOG -D $LRSYNC_LOG
339 check_diff $DIR/$tdir $TGT/$tdir
340 check_diff $DIR/$tdir $TGT2/$tdir
346 run_test 2b "Replicate files changed by dbench."
348 # Test 2c - Replicate files while dbench is running
354 bash rundbench -C -D $DIR/$tdir 2 -t $DBENCH_TIME &
356 local LRSYNC_LOG=$(generate_logname "lrsync_log")
357 # Replicate the changes to $TGT
358 sleep 10 # give dbench a headstart
360 while [ $quit -le 1 ];
362 echo "Running lustre_rsync"
363 $LRSYNC -s $DIR -t $TGT -t $TGT2 -m ${mds1_svc} -u $CL_USER \
364 -l $LREPL_LOG -D $LRSYNC_LOG
367 if [ $? -ne 0 ]; then
368 quit=$(expr $quit + 1)
372 # Use diff to compare the source and the destination
373 check_diff $DIR/$tdir $TGT/$tdir
374 check_diff $DIR/$tdir $TGT2/$tdir
380 run_test 2c "Replicate files while dbench is running."
382 # Test 3a - Replicate files created by createmany
388 createmany -o $DIR/$tdir/$tfile $numfiles || error "createmany failed"
390 local LRSYNC_LOG=$(generate_logname "lrsync_log")
391 # Replicate the changes to $TGT
392 $LRSYNC -s $DIR -t $TGT -t $TGT2 -m $MDT0 -u $CL_USER -l $LREPL_LOG \
394 check_diff $DIR/$tdir $TGT/$tdir
395 check_diff $DIR/$tdir $TGT2/$tdir
401 run_test 3a "Replicate files created by createmany"
403 # Test 3b - Replicate files created by writemany
410 writemany -q -a $DIR/$tdir/$tfile $time $threads ||
411 error "writemany failed"
413 local LRSYNC_LOG=$(generate_logname "lrsync_log")
414 # Replicate the changes to $TGT
415 $LRSYNC -s $DIR -t $TGT -t $TGT2 -m $MDT0 -u $CL_USER -l $LREPL_LOG \
418 check_diff $DIR/$tdir $TGT/$tdir
419 check_diff $DIR/$tdir $TGT2/$tdir
425 run_test 3b "Replicate files created by writemany"
427 # Test 3c - Replicate files created by createmany/unlinkmany
433 createmany -o $DIR/$tdir/$tfile $numfiles || error "createmany failed"
434 unlinkmany $DIR/$tdir/$tfile $numfiles || error "unlinkmany failed"
436 local LRSYNC_LOG=$(generate_logname "lrsync_log")
437 # Replicate the changes to $TGT
438 $LRSYNC -s $DIR -t $TGT -t $TGT2 -m $MDT0 -u $CL_USER -l $LREPL_LOG \
440 check_diff $DIR/$tdir $TGT/$tdir
441 check_diff $DIR/$tdir $TGT2/$tdir
447 run_test 3c "Replicate files created by createmany/unlinkmany"
449 # Test 4 - Replicate files created by iozone
451 which iozone > /dev/null 2>&1
452 [ $? -ne 0 ] && skip "iozone not found"
457 END_RUN_FILE=${DIR}/$tdir/run LOAD_PID_FILE=${DIR}/$tdir/pid \
458 MOUNT=${DIR}/$tdir run_iozone.sh &
460 child_pid=$(pgrep iozone)
461 stop_procs $child_pid
463 local LRSYNC_LOG=$(generate_logname "lrsync_log")
464 # Replicate the changes to $TGT
465 $LRSYNC -s $DIR -t $TGT -t $TGT2 -m $MDT0 -u $CL_USER -l $LREPL_LOG \
467 check_diff $DIR/$tdir $TGT/$tdir
468 check_diff $DIR/$tdir $TGT2/$tdir
470 $KILL -SIGCONT $child_pid
472 $KILL -SIGKILL $(pgrep run_iozone.sh)
473 $KILL -SIGKILL $(pgrep iozone)
475 # After killing 'run_iozone.sh', process 'iozone' becomes the
476 # child of PID 1. Hence 'wait' does not wait for it. Killing
477 # iozone first, means more iozone processes are spawned off which
478 # is not desirable. So, after sending a sigkill, the test goes
479 # into a wait loop for iozone to cleanup and exit.
481 while [ "$(pgrep "iozone")" != "" ];
483 ps -ef | grep iozone | grep -v grep
487 $LRSYNC -l $LREPL_LOG -D $LRSYNC_LOG
488 check_diff $DIR/$tdir $TGT/$tdir
489 check_diff $DIR/$tdir $TGT2/$tdir
495 run_test 4 "Replicate files created by iozone"
497 # Test 5a - Stop / start lustre_rsync
503 createmany -o $DIR/$tdir/$tfile $NUMTEST
505 # Replicate the changes to $TGT
506 local LRSYNC_LOG=$(generate_logname "lrsync_log")
507 $LRSYNC -s $DIR -t $TGT -t $TGT2 -m $MDT0 -u $CL_USER -l $LREPL_LOG \
511 $KILL -SIGHUP $child_pid
513 $LRSYNC -l $LREPL_LOG -D $LRSYNC_LOG
515 check_diff $DIR/$tdir $TGT/$tdir
516 check_diff $DIR/$tdir $TGT2/$tdir
522 run_test 5a "Stop / start lustre_rsync"
524 # Test 5b - Kill / restart lustre_rsync
530 createmany -o $DIR/$tdir/$tfile $NUMTEST
532 # Replicate the changes to $TGT
533 local LRSYNC_LOG=$(generate_logname "lrsync_log")
534 $LRSYNC -s $DIR -t $TGT -t $TGT2 -m $MDT0 -u $CL_USER -l $LREPL_LOG \
538 $KILL -SIGKILL $child_pid
540 $LRSYNC -l $LREPL_LOG -D $LRSYNC_LOG
542 check_diff $DIR/$tdir $TGT/$tdir
543 check_diff $DIR/$tdir $TGT2/$tdir
549 run_test 5b "Kill / restart lustre_rsync"
551 # Test 6 - lustre_rsync large no of hard links
559 touch $DIR/$tdir/link0
560 for ((i = 1; i < num_links - 1; i++)); do
561 ln $DIR/$tdir/link0 $DIR/$tdir/link$i
563 # create an extra hard link of src name ending with dest name
564 ln $DIR/$tdir/link0 $DIR/$tdir/ink0
566 local LRSYNC_LOG=$(generate_logname "lrsync_log")
567 # Replicate the changes to $TGT
568 $LRSYNC -s $DIR -t $TGT -t $TGT2 -m $MDT0 -u $CL_USER -l $LREPL_LOG \
570 check_diff $DIR/$tdir $TGT/$tdir
571 check_diff $DIR/$tdir $TGT2/$tdir
573 local count1=$(stat --format=%h $TGT/$tdir/link0)
574 local count2=$(stat --format=%h $TGT2/$tdir/link0)
575 if ((count1 != num_links || count2 != num_links)); then
576 ls -l $TGT/$tdir/link0 $TGT2/$tdir/link0
577 error "Incorrect no of hard links found $count1, $count2"
584 run_test 6 "lustre_rsync large no of hard links"
586 # Test 7 - lustre_rsync stripesize
591 mkdir -p ${DIR}/tgt/$tdir
594 $LFS setstripe -c $OSTCOUNT $DIR/$tdir ||
595 error "$LFS setstripe failed"
596 createmany -o $DIR/$tdir/$tfile $numfiles
598 # To simulate replication to another lustre filesystem, replicate
599 # the changes to $DIR/tgt. We can't turn off the changelogs
600 # while we are registered, so lustre_rsync better not try to
601 # replicate the replication steps. It seems ok :)
603 local LRSYNC_LOG=$(generate_logname "lrsync_log")
604 $LRSYNC -s $DIR -t $DIR/tgt -m $MDT0 -u $CL_USER -l $LREPL_LOG \
606 check_diff ${DIR}/$tdir $DIR/tgt/$tdir
609 while [ $i -lt $numfiles ];
611 local count=$($LFS getstripe $DIR/tgt/$tdir/${tfile}$i |
612 awk '/stripe_count/ {print $2}')
613 if [ $count -ne $OSTCOUNT ]; then
614 error "Stripe size not replicated"
622 run_test 7 "lustre_rsync stripesize"
624 # Test 8 - Replicate multiple file/directory moves
629 for i in 1 2 3 4 5 6 7 8 9; do
631 for j in 1 2 3 4 5 6 7 8 9; do
632 mkdir $DIR/$tdir/d$i/d$i$j
633 createmany -o $DIR/$tdir/d$i/d$i$j/a 10 \
635 mv $DIR/$tdir/d$i/d$i$j $DIR/$tdir/d$i/d0$i$j
636 createmany -o $DIR/$tdir/d$i/d0$i$j/b 10 \
638 mv $DIR/$tdir/d$i/d0$i$j/a0 $DIR/$tdir/d$i/d0$i$j/c0
640 mv $DIR/$tdir/d$i $DIR/$tdir/d0$i
643 local LRSYNC_LOG=$(generate_logname "lrsync_log")
644 $LRSYNC -s $DIR -t $TGT -m $MDT0 -u $CL_USER -l $LREPL_LOG \
647 check_diff ${DIR}/$tdir $TGT/$tdir
653 run_test 8 "Replicate multiple file/directory moves"
660 touch $DIR/$tdir/foo/a1
662 local LRSYNC_LOG=$(generate_logname "lrsync_log")
663 $LRSYNC -s $DIR -t $TGT -m $MDT0 -u $CL_USER -l $LREPL_LOG \
666 check_diff ${DIR}/$tdir $TGT/$tdir
668 rm -rf $DIR/$tdir/foo
670 $LRSYNC -s $DIR -t $TGT -m $MDT0 -u $CL_USER -l $LREPL_LOG \
673 check_diff ${DIR}/$tdir $TGT/$tdir
679 run_test 9 "Replicate recursive directory removal"
682 complete_test $SECONDS
683 check_and_cleanup_lustre