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
82 # Check whether the filesystem supports xattr or not.
84 # "large" - large xattr is supported
85 # "small" - large xattr is unsupported but small xattr is supported
86 # "no" - xattr is unsupported
93 local val="$(generate_string $(max_xattr_size))"
94 if large_xattr_enabled &&
95 setfattr -n user.foo -v $val $tgt 2>/dev/null; then
98 setfattr -n user.foo -v bar $tgt 2>/dev/null && xattr="small"
106 local changelog_file=$(generate_logname "changelog")
108 if [ -e $1 -o -e $2 ]; then
109 diff -rq -x "dev1" $1 $2
111 if [ $RC -ne 0 ]; then
112 $LFS changelog $MDT0 > $changelog_file
113 error "Failure in replication; differences found."
118 procs_are_stopped() {
122 for state in $(ps -p "$pids" -o state=); do
123 if [[ "$state" != T ]]; then
131 # Send SIGSTOP to PIDs and wait up to 60 seconds for them to show a
132 # stopped process state.
138 end=$((SECONDS + 60))
139 while ((SECONDS < end)); do
140 if procs_are_stopped $pids; then
150 # Test 1A - test basic operations
151 test_1A() { # was test_1
154 local xattr=$(check_xattr $TGT/foo)
161 touch $DIR/$tdir/file1
162 cp /etc/hosts $DIR/$tdir/d1/
163 touch $DIR/$tdir/d1/"space in filename"
164 touch $DIR/$tdir/d1/file2
167 mv $DIR/$tdir/d1/file2 $DIR/$tdir/d2/file3
169 # File and directory delete
170 touch $DIR/$tdir/d1/file4
171 mkdir $DIR/$tdir/d1/del
172 touch $DIR/$tdir/d1/del/del1
173 touch $DIR/$tdir/d1/del/del2
174 rm -rf $DIR/$tdir/d1/del
175 rm $DIR/$tdir/d1/file4
178 cat /etc/hosts > $DIR/$tdir/d1/link1
179 ln $DIR/$tdir/d1/link1 $DIR/$tdir/d1/link2
180 ln -s $DIR/$tdir/d1/link1 $DIR/$tdir/d1/link3
183 #mknod $DIR/$tdir/dev1 b 8 1
186 local LRSYNC_LOG=$(generate_logname "lrsync_log")
187 echo "Replication #1"
188 $LRSYNC -s $DIR -t $TGT -t $TGT2 -m $MDT0 -u $CL_USER -l $LREPL_LOG \
192 chmod 000 $DIR/$tdir/d2/file3
193 chown nobody:$GROUP $DIR/$tdir/d2/file3
196 if [[ "$xattr" != "no" ]]; then
198 touch $DIR/$tdir/file5
199 [[ "$xattr" = "large" ]] &&
200 value="$(generate_string $(max_xattr_size))" || value="bar"
201 setfattr -n user.foo -v $value $DIR/$tdir/file5
204 echo "Replication #2"
205 $LRSYNC -l $LREPL_LOG -D $LRSYNC_LOG
207 if [[ "$xattr" != "no" ]]; then
208 local xval1=$(get_xattr_value user.foo $TGT/$tdir/file5)
209 local xval2=$(get_xattr_value user.foo $TGT2/$tdir/file5)
211 if [[ "$xval1" != "$value" || "$xval2" != "$value" ]]; then
212 error "Error in replicating xattrs."
216 # Use diff to compare the source and the destination
217 check_diff $DIR/$tdir $TGT/$tdir
218 check_diff $DIR/$tdir $TGT2/$tdir
223 run_test 1A "Simple Replication"
225 # Test 1a - test create/delete operations in ROOT directory
226 test_1a() { # LU-5005
227 rm -rf $TGT/root-* 2> /dev/null
228 rm -rf $DIR/root-* 2> /dev/null
233 mkdir_on_mdt0 $DIR/root-dir
237 touch $DIR/root-file2
240 mv $DIR/root-file2 $DIR/root-file3
242 # File and directory delete
243 touch $DIR/root-file4
244 mkdir_on_mdt0 $DIR/root-dir1
246 rm -rf $DIR/root-dir1
249 local LRSYNC_LOG=$(generate_logname "lrsync_log")
251 $LRSYNC -s $DIR -t $TGT -m $MDT0 -u $CL_USER -l $LREPL_LOG \
255 stat $TGT/root-dir || error "Dir create not replicated"
256 stat $TGT/root-file || error "File create not replicated"
257 stat $TGT/root-file2 && error "Rename not replicated (src)"
258 stat $TGT/root-file3 || error "Rename not replicated (tgt)"
259 stat $TGT/root-dir1 && error "Dir delete not replicated"
260 stat $TGT/root-file4 && error "File delete not replicated"
268 run_test 1a "Replicate create/delete operations in ROOT directory"
270 # Test 2a - Replicate files created by dbench
276 bash rundbench -C -D $DIR/$tdir 2 -t $DBENCH_TIME ||
277 error "dbench failed"
279 local LRSYNC_LOG=$(generate_logname "lrsync_log")
280 # Replicate the changes to $TGT
281 $LRSYNC -s $DIR -t $TGT -t $TGT2 -m $MDT0 -u $CL_USER -l $LREPL_LOG \
284 # Use diff to compare the source and the destination
285 check_diff $DIR/$tdir $TGT/$tdir
286 check_diff $DIR/$tdir $TGT2/$tdir
292 run_test 2a "Replicate files created by dbench."
294 # Test 2b - Replicate files changed by dbench.
301 bash rundbench -C -D $DIR/$tdir 2 -t $DBENCH_TIME &
302 # wait for dbench to start
303 wait_for_function 'child_pid=$(pgrep dbench)' 360
304 # let dbench run for a bit
307 echo PIDs: $child_pid
309 stop_procs $child_pid
311 local LRSYNC_LOG=$(generate_logname "lrsync_log")
312 echo Starting replication
313 $LRSYNC -s $DIR -t $TGT -t $TGT2 -m $MDT0 -u $CL_USER -l $LREPL_LOG \
315 check_diff $DIR/$tdir $TGT/$tdir
318 $KILL -SIGCONT $child_pid
322 stop_procs $child_pid
324 echo Starting replication
325 $LRSYNC -l $LREPL_LOG -D $LRSYNC_LOG
326 check_diff $DIR/$tdir $TGT/$tdir
328 echo "Wait for dbench to finish"
329 $KILL -SIGCONT $child_pid
332 # Replicate the changes to $TGT
333 echo Starting replication
334 $LRSYNC -l $LREPL_LOG -D $LRSYNC_LOG
336 check_diff $DIR/$tdir $TGT/$tdir
337 check_diff $DIR/$tdir $TGT2/$tdir
343 run_test 2b "Replicate files changed by dbench."
345 # Test 2c - Replicate files while dbench is running
351 bash rundbench -C -D $DIR/$tdir 2 -t $DBENCH_TIME &
353 local LRSYNC_LOG=$(generate_logname "lrsync_log")
354 # Replicate the changes to $TGT
355 sleep 10 # give dbench a headstart
357 while [ $quit -le 1 ];
359 echo "Running lustre_rsync"
360 $LRSYNC -s $DIR -t $TGT -t $TGT2 -m ${mds1_svc} -u $CL_USER \
361 -l $LREPL_LOG -D $LRSYNC_LOG
364 if [ $? -ne 0 ]; then
365 quit=$(expr $quit + 1)
369 # Use diff to compare the source and the destination
370 check_diff $DIR/$tdir $TGT/$tdir
371 check_diff $DIR/$tdir $TGT2/$tdir
377 run_test 2c "Replicate files while dbench is running."
379 # Test 3a - Replicate files created by createmany
385 createmany -o $DIR/$tdir/$tfile $numfiles || error "createmany failed"
387 local LRSYNC_LOG=$(generate_logname "lrsync_log")
388 # Replicate the changes to $TGT
389 $LRSYNC -s $DIR -t $TGT -t $TGT2 -m $MDT0 -u $CL_USER -l $LREPL_LOG \
391 check_diff $DIR/$tdir $TGT/$tdir
392 check_diff $DIR/$tdir $TGT2/$tdir
398 run_test 3a "Replicate files created by createmany"
400 # Test 3b - Replicate files created by writemany
407 writemany -q -a $DIR/$tdir/$tfile $time $threads ||
408 error "writemany failed"
410 local LRSYNC_LOG=$(generate_logname "lrsync_log")
411 # Replicate the changes to $TGT
412 $LRSYNC -s $DIR -t $TGT -t $TGT2 -m $MDT0 -u $CL_USER -l $LREPL_LOG \
415 check_diff $DIR/$tdir $TGT/$tdir
416 check_diff $DIR/$tdir $TGT2/$tdir
422 run_test 3b "Replicate files created by writemany"
424 # Test 3c - Replicate files created by createmany/unlinkmany
430 createmany -o $DIR/$tdir/$tfile $numfiles || error "createmany failed"
431 unlinkmany $DIR/$tdir/$tfile $numfiles || error "unlinkmany failed"
433 local LRSYNC_LOG=$(generate_logname "lrsync_log")
434 # Replicate the changes to $TGT
435 $LRSYNC -s $DIR -t $TGT -t $TGT2 -m $MDT0 -u $CL_USER -l $LREPL_LOG \
437 check_diff $DIR/$tdir $TGT/$tdir
438 check_diff $DIR/$tdir $TGT2/$tdir
444 run_test 3c "Replicate files created by createmany/unlinkmany"
446 # Test 4 - Replicate files created by iozone
448 which iozone > /dev/null 2>&1
449 [ $? -ne 0 ] && skip "iozone not found"
454 END_RUN_FILE=${DIR}/$tdir/run LOAD_PID_FILE=${DIR}/$tdir/pid \
455 MOUNT=${DIR}/$tdir run_iozone.sh &
457 child_pid=$(pgrep iozone)
458 stop_procs $child_pid
460 local LRSYNC_LOG=$(generate_logname "lrsync_log")
461 # Replicate the changes to $TGT
462 $LRSYNC -s $DIR -t $TGT -t $TGT2 -m $MDT0 -u $CL_USER -l $LREPL_LOG \
464 check_diff $DIR/$tdir $TGT/$tdir
465 check_diff $DIR/$tdir $TGT2/$tdir
467 $KILL -SIGCONT $child_pid
469 $KILL -SIGKILL $(pgrep run_iozone.sh)
470 $KILL -SIGKILL $(pgrep iozone)
472 # After killing 'run_iozone.sh', process 'iozone' becomes the
473 # child of PID 1. Hence 'wait' does not wait for it. Killing
474 # iozone first, means more iozone processes are spawned off which
475 # is not desirable. So, after sending a sigkill, the test goes
476 # into a wait loop for iozone to cleanup and exit.
478 while [ "$(pgrep "iozone")" != "" ];
480 ps -ef | grep iozone | grep -v grep
484 $LRSYNC -l $LREPL_LOG -D $LRSYNC_LOG
485 check_diff $DIR/$tdir $TGT/$tdir
486 check_diff $DIR/$tdir $TGT2/$tdir
492 run_test 4 "Replicate files created by iozone"
494 # Test 5a - Stop / start lustre_rsync
500 createmany -o $DIR/$tdir/$tfile $NUMTEST
502 # Replicate the changes to $TGT
503 local LRSYNC_LOG=$(generate_logname "lrsync_log")
504 $LRSYNC -s $DIR -t $TGT -t $TGT2 -m $MDT0 -u $CL_USER -l $LREPL_LOG \
508 $KILL -SIGHUP $child_pid
510 $LRSYNC -l $LREPL_LOG -D $LRSYNC_LOG
512 check_diff $DIR/$tdir $TGT/$tdir
513 check_diff $DIR/$tdir $TGT2/$tdir
519 run_test 5a "Stop / start lustre_rsync"
521 # Test 5b - Kill / restart lustre_rsync
527 createmany -o $DIR/$tdir/$tfile $NUMTEST
529 # Replicate the changes to $TGT
530 local LRSYNC_LOG=$(generate_logname "lrsync_log")
531 $LRSYNC -s $DIR -t $TGT -t $TGT2 -m $MDT0 -u $CL_USER -l $LREPL_LOG \
535 $KILL -SIGKILL $child_pid
537 $LRSYNC -l $LREPL_LOG -D $LRSYNC_LOG
539 check_diff $DIR/$tdir $TGT/$tdir
540 check_diff $DIR/$tdir $TGT2/$tdir
546 run_test 5b "Kill / restart lustre_rsync"
548 # Test 6 - lustre_rsync large no of hard links
556 touch $DIR/$tdir/link0
557 for ((i = 1; i < num_links - 1; i++)); do
558 ln $DIR/$tdir/link0 $DIR/$tdir/link$i
560 # create an extra hard link of src name ending with dest name
561 ln $DIR/$tdir/link0 $DIR/$tdir/ink0
563 local LRSYNC_LOG=$(generate_logname "lrsync_log")
564 # Replicate the changes to $TGT
565 $LRSYNC -s $DIR -t $TGT -t $TGT2 -m $MDT0 -u $CL_USER -l $LREPL_LOG \
567 check_diff $DIR/$tdir $TGT/$tdir
568 check_diff $DIR/$tdir $TGT2/$tdir
570 local count1=$(stat --format=%h $TGT/$tdir/link0)
571 local count2=$(stat --format=%h $TGT2/$tdir/link0)
572 if ((count1 != num_links || count2 != num_links)); then
573 ls -l $TGT/$tdir/link0 $TGT2/$tdir/link0
574 error "Incorrect no of hard links found $count1, $count2"
581 run_test 6 "lustre_rsync large no of hard links"
583 # Test 7 - lustre_rsync stripesize
588 mkdir -p ${DIR}/tgt/$tdir
591 $LFS setstripe -c $OSTCOUNT $DIR/$tdir ||
592 error "$LFS setstripe failed"
593 createmany -o $DIR/$tdir/$tfile $numfiles
595 # To simulate replication to another lustre filesystem, replicate
596 # the changes to $DIR/tgt. We can't turn off the changelogs
597 # while we are registered, so lustre_rsync better not try to
598 # replicate the replication steps. It seems ok :)
600 local LRSYNC_LOG=$(generate_logname "lrsync_log")
601 $LRSYNC -s $DIR -t $DIR/tgt -m $MDT0 -u $CL_USER -l $LREPL_LOG \
603 check_diff ${DIR}/$tdir $DIR/tgt/$tdir
606 while [ $i -lt $numfiles ];
608 local count=$($LFS getstripe $DIR/tgt/$tdir/${tfile}$i |
609 awk '/stripe_count/ {print $2}')
610 if [ $count -ne $OSTCOUNT ]; then
611 error "Stripe size not replicated"
619 run_test 7 "lustre_rsync stripesize"
621 # Test 8 - Replicate multiple file/directory moves
626 for i in 1 2 3 4 5 6 7 8 9; do
628 for j in 1 2 3 4 5 6 7 8 9; do
629 mkdir $DIR/$tdir/d$i/d$i$j
630 createmany -o $DIR/$tdir/d$i/d$i$j/a 10 \
632 mv $DIR/$tdir/d$i/d$i$j $DIR/$tdir/d$i/d0$i$j
633 createmany -o $DIR/$tdir/d$i/d0$i$j/b 10 \
635 mv $DIR/$tdir/d$i/d0$i$j/a0 $DIR/$tdir/d$i/d0$i$j/c0
637 mv $DIR/$tdir/d$i $DIR/$tdir/d0$i
640 local LRSYNC_LOG=$(generate_logname "lrsync_log")
641 $LRSYNC -s $DIR -t $TGT -m $MDT0 -u $CL_USER -l $LREPL_LOG \
644 check_diff ${DIR}/$tdir $TGT/$tdir
650 run_test 8 "Replicate multiple file/directory moves"
657 touch $DIR/$tdir/foo/a1
659 local LRSYNC_LOG=$(generate_logname "lrsync_log")
660 $LRSYNC -s $DIR -t $TGT -m $MDT0 -u $CL_USER -l $LREPL_LOG \
663 check_diff ${DIR}/$tdir $TGT/$tdir
665 rm -rf $DIR/$tdir/foo
667 $LRSYNC -s $DIR -t $TGT -m $MDT0 -u $CL_USER -l $LREPL_LOG \
670 check_diff ${DIR}/$tdir $TGT/$tdir
676 run_test 9 "Replicate recursive directory removal"
680 check_and_cleanup_lustre