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 always_except LU-4256 2b
21 [[ $(uname -r) = *"debug" ]] &&
22 always_except LU-16489 2c
26 [ -n "$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 to complete $DBENCH_TIME seconds"
282 local log=$(generate_logname "lrsync_log")
283 # Replicate the changes to $TGT
284 $LRSYNC -s $DIR -t $TGT -m $MDT0 -u $CL_USER -l $LREPL_LOG -D $log
286 # Use diff to compare the source and the destination
287 check_diff $DIR/$tdir $TGT/$tdir
293 run_test 2a "Replicate files created by dbench."
295 # Test 2b - Replicate files changed by dbench.
302 bash rundbench -C -D $DIR/$tdir 2 -t $DBENCH_TIME &
303 # wait for dbench to start
304 wait_for_function 'child_pid=$(pgrep dbench)' 360
305 # let dbench run for a bit
308 echo PIDs: $child_pid
310 stop_procs $child_pid
312 local log=$(generate_logname "lrsync_log")
313 echo Starting replication
314 $LRSYNC -s $DIR -t $TGT -m $MDT0 -u $CL_USER -l $LREPL_LOG -D $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 $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 $log
336 check_diff $DIR/$tdir $TGT/$tdir
342 run_test 2b "Replicate files changed by dbench."
344 # Test 2c - Replicate files while dbench is running
350 bash rundbench -C -D $DIR/$tdir 2 -t $DBENCH_TIME &
352 local LRSYNC_LOG=$(generate_logname "lrsync_log")
353 # Replicate the changes to $TGT
354 sleep 10 # give dbench a headstart
356 while [ $quit -le 1 ];
358 echo "Running lustre_rsync"
359 $LRSYNC -s $DIR -t $TGT -t $TGT2 -m ${mds1_svc} -u $CL_USER \
360 -l $LREPL_LOG -D $LRSYNC_LOG
363 if [ $? -ne 0 ]; then
364 quit=$(expr $quit + 1)
368 # Use diff to compare the source and the destination
369 check_diff $DIR/$tdir $TGT/$tdir
370 check_diff $DIR/$tdir $TGT2/$tdir
376 run_test 2c "Replicate files while dbench is running."
378 # Test 3a - Replicate files created by createmany
384 createmany -o $DIR/$tdir/$tfile $numfiles || error "createmany failed"
386 local LRSYNC_LOG=$(generate_logname "lrsync_log")
387 # Replicate the changes to $TGT
388 $LRSYNC -s $DIR -t $TGT -t $TGT2 -m $MDT0 -u $CL_USER -l $LREPL_LOG \
390 check_diff $DIR/$tdir $TGT/$tdir
391 check_diff $DIR/$tdir $TGT2/$tdir
397 run_test 3a "Replicate files created by createmany"
399 # Test 3b - Replicate files created by writemany
406 writemany -q -a $DIR/$tdir/$tfile $time $threads ||
407 error "writemany failed"
409 local LRSYNC_LOG=$(generate_logname "lrsync_log")
410 # Replicate the changes to $TGT
411 $LRSYNC -s $DIR -t $TGT -t $TGT2 -m $MDT0 -u $CL_USER -l $LREPL_LOG \
414 check_diff $DIR/$tdir $TGT/$tdir
415 check_diff $DIR/$tdir $TGT2/$tdir
421 run_test 3b "Replicate files created by writemany"
423 # Test 3c - Replicate files created by createmany/unlinkmany
429 createmany -o $DIR/$tdir/$tfile $numfiles || error "createmany failed"
430 unlinkmany $DIR/$tdir/$tfile $numfiles || error "unlinkmany failed"
432 local LRSYNC_LOG=$(generate_logname "lrsync_log")
433 # Replicate the changes to $TGT
434 $LRSYNC -s $DIR -t $TGT -t $TGT2 -m $MDT0 -u $CL_USER -l $LREPL_LOG \
436 check_diff $DIR/$tdir $TGT/$tdir
437 check_diff $DIR/$tdir $TGT2/$tdir
443 run_test 3c "Replicate files created by createmany/unlinkmany"
445 # Test 4 - Replicate files created by iozone
447 which iozone > /dev/null 2>&1
448 [ $? -ne 0 ] && skip "iozone not found"
453 END_RUN_FILE=${DIR}/$tdir/run LOAD_PID_FILE=${DIR}/$tdir/pid \
454 MOUNT=${DIR}/$tdir RECOVERY_SCALE_ENABLE_REMOTE_DIRS=false \
455 RECOVERY_SCALE_ENABLE_STRIPED_DIRS=false run_iozone.sh &
457 child_pid=$(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"
679 complete_test $SECONDS
680 check_and_cleanup_lustre