X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Ftests%2Flustre-rsync-test.sh;h=9c8658dba1038ba4dcb22c2b55bed166bb1b000c;hb=0652c828c93f266b9999883b1b2967bfaac11c89;hp=de9f4283f879449d6bf9a62fc0e57ab6655fa0ec;hpb=7b6aaef15a6547550c99cdaa538ea8b1d4d87c88;p=fs%2Flustre-release.git diff --git a/lustre/tests/lustre-rsync-test.sh b/lustre/tests/lustre-rsync-test.sh index de9f428..9c8658d 100644 --- a/lustre/tests/lustre-rsync-test.sh +++ b/lustre/tests/lustre-rsync-test.sh @@ -12,13 +12,14 @@ SRCDIR=`dirname $0` export PATH=$PWD/$SRCDIR:$SRCDIR:$PWD/$SRCDIR/../utils:$PATH:/sbin ONLY=${ONLY:-"$*"} -[ -n "$ONLY" ] && SLOW=yes -ALWAYS_EXCEPT="$LRSYNC_EXCEPT" -# bug number for skipped test: +# bug number for skipped test: LU-4256 +ALWAYS_EXCEPT="$LRSYNC_EXCEPT 2b" # UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT! -[ "$ALWAYS_EXCEPT$EXCEPT" ] && \ - echo "Skipping tests: `echo $ALWAYS_EXCEPT $EXCEPT`" +[ "$SLOW" = "no" ] && EXCEPT_SLOW="" + +[ "$ALWAYS_EXCEPT$EXCEPT" ] && + echo "Skipping tests: `echo $ALWAYS_EXCEPT $EXCEPT`" KILL=/bin/kill @@ -37,51 +38,54 @@ check_and_setup_lustre DIR=${DIR:-$MOUNT} assert_DIR +if getent group nobody; then + GROUP=nobody +elif getent group nogroup; then + GROUP=nogroup +else + error "No generic nobody group" +fi build_test_filter export LRSYNC=${LRSYNC:-"$LUSTRE/utils/lustre_rsync"} [ ! -f "$LRSYNC" ] && export LRSYNC=$(which lustre_rsync) -export LRSYNC="$LRSYNC -v -c no" # -a +export LRSYNC="$LRSYNC -v -c no -d 2" -# control the time of tests -DBENCH_TIME=${DBENCH_TIME:-60} # No of seconds to run dbench +# Number of seconds to run dbench +DBENCH_TIME=${DBENCH_TIME:-60} TGT=$TMP/target TGT2=$TMP/target2 -MDT0=$($LCTL get_param -n mdc.*.mds_server_uuid | \ - awk '{gsub(/_UUID/,""); print $1}' | head -1) +MDT0=$($LCTL get_param -n mdc.*.mds_server_uuid | + awk '{ gsub(/_UUID/,""); print $1 }' | head -n1) init_changelog() { - CL_USER=$(do_facet $SINGLEMDS lctl --device $MDT0 changelog_register -n) - echo $MDT0: Registered changelog user $CL_USER - CL_USERS=$(( $(do_facet $SINGLEMDS lctl get_param -n \ - mdd.$MDT0.changelog_users | wc -l) - 2 )) - [ $CL_USERS -ne 1 ] && \ - echo "Other changelog users present ($CL_USERS)" + changelog_register || error "changelog_register failed" + CL_USER=(${CL_USERS[$SINGLEMDS]}) + echo $MDT0: Registered changelog user $CL_USER + [ -z $CL_USER ] && + echo "No changelog users present on $SINGLEMDS" } -init_src() { - rm -rf $TGT/$tdir $TGT/d*.lustre_rsync-test 2> /dev/null - rm -rf $TGT2/$tdir $TGT2/d*.lustre_rsync-test 2> /dev/null - rm -rf ${DIR}/$tdir $DIR/d*.lustre_rsync-test ${DIR}/tgt 2> /dev/null - rm -f $LREPL_LOG - mkdir -p ${DIR}/$tdir - mkdir -p ${TGT}/$tdir - mkdir -p ${TGT2}/$tdir - if [ $? -ne 0 ]; then - error "Failed to create target: " $TGT - fi +fini_changelog() { + changelog_clear + changelog_deregister } -cleanup_src_tgt() { - rm -rf $TGT/$tdir - rm -rf $DIR/$tdir - rm -rf $DIR/tgt +init_src() { + rm -rf $TGT/$tdir $TGT/d*.lustre_rsync-test 2> /dev/null + rm -rf $TGT2/$tdir $TGT2/d*.lustre_rsync-test 2> /dev/null + rm -rf ${DIR}/$tdir $DIR/d*.lustre_rsync-test ${DIR}/tgt 2> /dev/null + rm -f $LREPL_LOG + mkdir -p ${DIR}/$tdir || error "Failed to create target: " $DIR/$tdir + mkdir -p ${TGT}/$tdir || error "Failed to create target: " $TGT/$tdir + mkdir -p ${TGT2}/$tdir || error "Failed to create target: " $TGT2/$tdir } -fini_changelog() { - $LFS changelog_clear $MDT0 $CL_USER 0 - do_facet $SINGLEMDS lctl --device $MDT0 changelog_deregister $CL_USER +cleanup_src_tgt() { + rm -rf $TGT/$tdir + rm -rf $DIR/$tdir + rm -rf $DIR/tgt } # Check whether the filesystem supports xattr or not. @@ -108,7 +112,7 @@ check_xattr() { } check_diff() { - local changelog_file=$LOGDIR/${TESTSUITE}.test_${3}.changelog + local changelog_file=$(generate_logname "changelog") if [ -e $1 -o -e $2 ]; then diff -rq -x "dev1" $1 $2 @@ -120,8 +124,40 @@ check_diff() { fi } -# Test 1 - test basic operations -test_1() { +procs_are_stopped() { + local pids="$*" + local state + + for state in $(ps -p "$pids" -o state=); do + if [[ "$state" != T ]]; then + return 1 + fi + done + + return 0 +} + +# Send SIGSTOP to PIDs and wait up to 60 seconds for them to show a +# stopped process state. +stop_procs() { + local pids="$*" + local end + + $KILL -SIGSTOP $pids + end=$((SECONDS + 60)) + while ((SECONDS < end)); do + if procs_are_stopped $pids; then + return 0 + fi + + sleep 1 + done + + return 1 +} + +# Test 1A - test basic operations +test_1A() { # was test_1 init_src init_changelog local xattr=$(check_xattr $TGT/foo) @@ -155,13 +191,15 @@ test_1() { # Device files #mknod $DIR/$tdir/dev1 b 8 1 - # Replicate - echo "Replication #1" - $LRSYNC -s $DIR -t $TGT -t $TGT2 -m $MDT0 -u $CL_USER -l $LREPL_LOG + # Replicate + local LRSYNC_LOG=$(generate_logname "lrsync_log") + echo "Replication #1" + $LRSYNC -s $DIR -t $TGT -t $TGT2 -m $MDT0 -u $CL_USER -l $LREPL_LOG \ + -D $LRSYNC_LOG # Set attributes chmod 000 $DIR/$tdir/d2/file3 - chown nobody:nobody $DIR/$tdir/d2/file3 + chown nobody:$GROUP $DIR/$tdir/d2/file3 # Set xattrs if [[ "$xattr" != "no" ]]; then @@ -172,102 +210,139 @@ test_1() { setfattr -n user.foo -v $value $DIR/$tdir/file5 fi - echo "Replication #2" - $LRSYNC -l $LREPL_LOG - - if [[ "$xattr" != "no" ]]; then - local xval1=$(get_xattr_value user.foo $TGT/$tdir/file5) - local xval2=$(get_xattr_value user.foo $TGT2/$tdir/file5) - fi - - RC=0 + echo "Replication #2" + $LRSYNC -l $LREPL_LOG -D $LRSYNC_LOG - # fid2path and path2fid aren't implemented for block devices - #if [[ ! -b $TGT/$tdir/dev1 ]] || [[ ! -b $TGT2/$tdir/dev1 ]]; then - # ls -l $DIR/$tdir/dev1 $TGT/$tdir/dev1 $TGT2/$tdir/dev1 - # error "Error replicating block devices" - # RC=1 + if [[ "$xattr" != "no" ]]; then + local xval1=$(get_xattr_value user.foo $TGT/$tdir/file5) + local xval2=$(get_xattr_value user.foo $TGT2/$tdir/file5) - if [[ "$xattr" != "no" ]] && - [[ "$xval1" != "$value" || "$xval2" != "$value" ]]; then - error "Error in replicating xattrs." - RC=1 - fi + if [[ "$xval1" != "$value" || "$xval2" != "$value" ]]; then + error "Error in replicating xattrs." + fi + fi - # Use diff to compare the source and the destination - check_diff $DIR/$tdir $TGT/$tdir 1 - check_diff $DIR/$tdir $TGT2/$tdir 1 + # Use diff to compare the source and the destination + check_diff $DIR/$tdir $TGT/$tdir + check_diff $DIR/$tdir $TGT2/$tdir - fini_changelog - cleanup_src_tgt - return $RC + fini_changelog + cleanup_src_tgt +} +run_test 1A "Simple Replication" + +# Test 1a - test create/delete operations in ROOT directory +test_1a() { # LU-5005 + rm -rf $TGT/root-* 2> /dev/null + rm -rf $DIR/root-* 2> /dev/null + init_src + init_changelog + + # Directory create + mkdir $DIR/root-dir + + # File create + touch $DIR/root-file + touch $DIR/root-file2 + + # File rename + mv $DIR/root-file2 $DIR/root-file3 + + # File and directory delete + touch $DIR/root-file4 + mkdir $DIR/root-dir1 + rm $DIR/root-file4 + rm -rf $DIR/root-dir1 + + # Replicate + local LRSYNC_LOG=$(generate_logname "lrsync_log") + echo "Replication" + $LRSYNC -s $DIR -t $TGT -m $MDT0 -u $CL_USER -l $LREPL_LOG \ + -D $LRSYNC_LOG + + # Verify + stat $TGT/root-dir || error "Dir create not replicated" + stat $TGT/root-file || error "File create not replicated" + stat $TGT/root-file2 && error "Rename not replicated (src)" + stat $TGT/root-file3 || error "Rename not replicated (tgt)" + stat $TGT/root-dir1 && error "Dir delete not replicated" + stat $TGT/root-file4 && error "File delete not replicated" + + cleanup_src_tgt + fini_changelog + rm -fr $TGT/root-* + rm -fr $DIR/root-* + return 0 } -run_test 1 "Simple Replication" +run_test 1a "Replicate create/delete operations in ROOT directory" -# Test 2a - Replicate files created by dbench +# Test 2a - Replicate files created by dbench test_2a() { - [ "$SLOW" = "no" ] && skip "Skipping slow test" && return - init_src - init_changelog + init_src + init_changelog - # Run dbench - sh rundbench -C -D $DIR/$tdir 2 -t $DBENCH_TIME || error "dbench failed!" + # Run dbench + sh rundbench -C -D $DIR/$tdir 2 -t $DBENCH_TIME || error "dbench failed" - # Replicate the changes to $TGT - $LRSYNC -s $DIR -t $TGT -t $TGT2 -m $MDT0 -u $CL_USER -l $LREPL_LOG + local LRSYNC_LOG=$(generate_logname "lrsync_log") + # Replicate the changes to $TGT + $LRSYNC -s $DIR -t $TGT -t $TGT2 -m $MDT0 -u $CL_USER -l $LREPL_LOG \ + -D $LRSYNC_LOG - # Use diff to compare the source and the destination - check_diff $DIR/$tdir $TGT/$tdir 2a - check_diff $DIR/$tdir $TGT2/$tdir 2a + # Use diff to compare the source and the destination + check_diff $DIR/$tdir $TGT/$tdir + check_diff $DIR/$tdir $TGT2/$tdir - fini_changelog - cleanup_src_tgt - return 0 + fini_changelog + cleanup_src_tgt + return 0 } run_test 2a "Replicate files created by dbench." - # Test 2b - Replicate files changed by dbench. test_2b() { - [ "$SLOW" = "no" ] && skip "Skipping slow test" && return - - init_src - init_changelog - - # Run dbench - sh rundbench -C -D $DIR/$tdir 2 -t $DBENCH_TIME & - sleep 20 - - local child_pid=$(pgrep dbench) - echo PIDs: $child_pid - echo Stopping dbench - $KILL -SIGSTOP $child_pid - - echo Starting replication - $LRSYNC -s $DIR -t $TGT -t $TGT2 -m $MDT0 -u $CL_USER -l $LREPL_LOG - check_diff $DIR/$tdir $TGT/$tdir 2b + local child_pid + init_src + init_changelog + + # Run dbench + sh rundbench -C -D $DIR/$tdir 2 -t $DBENCH_TIME & + # wait for dbench to start + wait_for_function 'child_pid=$(pgrep dbench)' 360 + # let dbench run for a bit + sleep 10 + + echo PIDs: $child_pid + echo Stopping dbench + stop_procs $child_pid + + local LRSYNC_LOG=$(generate_logname "lrsync_log") + echo Starting replication + $LRSYNC -s $DIR -t $TGT -t $TGT2 -m $MDT0 -u $CL_USER -l $LREPL_LOG \ + -D $LRSYNC_LOG + check_diff $DIR/$tdir $TGT/$tdir echo Resuming dbench $KILL -SIGCONT $child_pid sleep 10 echo Stopping dbench - $KILL -SIGSTOP $child_pid + stop_procs $child_pid - echo Starting replication - $LRSYNC -l $LREPL_LOG - check_diff $DIR/$tdir $TGT/$tdir 2b + echo Starting replication + $LRSYNC -l $LREPL_LOG -D $LRSYNC_LOG + check_diff $DIR/$tdir $TGT/$tdir echo "Wait for dbench to finish" $KILL -SIGCONT $child_pid wait - # Replicate the changes to $TGT - echo Starting replication - $LRSYNC -l $LREPL_LOG + # Replicate the changes to $TGT + echo Starting replication + $LRSYNC -l $LREPL_LOG -D $LRSYNC_LOG - check_diff $DIR/$tdir $TGT/$tdir 2b - check_diff $DIR/$tdir $TGT2/$tdir 2b + check_diff $DIR/$tdir $TGT/$tdir + check_diff $DIR/$tdir $TGT2/$tdir fini_changelog cleanup_src_tgt @@ -275,129 +350,127 @@ test_2b() { } run_test 2b "Replicate files changed by dbench." -# Test 2c - Replicate files while dbench is running +# Test 2c - Replicate files while dbench is running test_2c() { - [ "$SLOW" = "no" ] && skip "Skipping slow test" && return - init_src - init_changelog - - # Run dbench - sh rundbench -C -D $DIR/$tdir 2 -t $DBENCH_TIME & - - # Replicate the changes to $TGT - sleep 10 # give dbench a headstart - local quit=0 - while [ $quit -le 1 ]; - do - echo "Running lustre_rsync" - $LRSYNC -s $DIR -t $TGT -t $TGT2 -m ${mds1_svc} -u $CL_USER -l $LREPL_LOG - sleep 5 - pgrep dbench - if [ $? -ne 0 ]; then - quit=$(expr $quit + 1) - fi - done + init_src + init_changelog + + # Run dbench + sh rundbench -C -D $DIR/$tdir 2 -t $DBENCH_TIME & + + local LRSYNC_LOG=$(generate_logname "lrsync_log") + # Replicate the changes to $TGT + sleep 10 # give dbench a headstart + local quit=0 + while [ $quit -le 1 ]; + do + echo "Running lustre_rsync" + $LRSYNC -s $DIR -t $TGT -t $TGT2 -m ${mds1_svc} -u $CL_USER \ + -l $LREPL_LOG -D $LRSYNC_LOG + sleep 5 + pgrep dbench + if [ $? -ne 0 ]; then + quit=$(expr $quit + 1) + fi + done - # Use diff to compare the source and the destination - check_diff $DIR/$tdir $TGT/$tdir 2c - check_diff $DIR/$tdir $TGT2/$tdir 2c + # Use diff to compare the source and the destination + check_diff $DIR/$tdir $TGT/$tdir + check_diff $DIR/$tdir $TGT2/$tdir - fini_changelog - cleanup_src_tgt - return 0 + fini_changelog + cleanup_src_tgt + return 0 } run_test 2c "Replicate files while dbench is running." # Test 3a - Replicate files created by createmany test_3a() { - [ "$SLOW" = "no" ] && skip "Skipping slow test" && return - - init_src - init_changelog - - local numfiles=1000 - createmany -o $DIR/$tdir/$tfile $numfiles || error "createmany failed!" - - # Replicate the changes to $TGT - $LRSYNC -s $DIR -t $TGT -t $TGT2 -m $MDT0 -u $CL_USER -l $LREPL_LOG - check_diff $DIR/$tdir $TGT/$tdir 3a - check_diff $DIR/$tdir $TGT2/$tdir 3a - - fini_changelog - cleanup_src_tgt - return 0 + init_src + init_changelog + + local numfiles=1000 + createmany -o $DIR/$tdir/$tfile $numfiles || error "createmany failed" + + local LRSYNC_LOG=$(generate_logname "lrsync_log") + # Replicate the changes to $TGT + $LRSYNC -s $DIR -t $TGT -t $TGT2 -m $MDT0 -u $CL_USER -l $LREPL_LOG \ + -D $LRSYNC_LOG + check_diff $DIR/$tdir $TGT/$tdir + check_diff $DIR/$tdir $TGT2/$tdir + + fini_changelog + cleanup_src_tgt + return 0 } run_test 3a "Replicate files created by createmany" - # Test 3b - Replicate files created by writemany test_3b() { - [ "$SLOW" = "no" ] && skip "Skipping slow test" && return + init_src + init_changelog - init_src - init_changelog + local time=60 + local threads=5 + writemany -q -a $DIR/$tdir/$tfile $time $threads || + error "writemany failed" - local time=60 - local threads=5 - writemany -q -a $DIR/$tdir/$tfile $time $threads || error "writemany failed!" + local LRSYNC_LOG=$(generate_logname "lrsync_log") + # Replicate the changes to $TGT + $LRSYNC -s $DIR -t $TGT -t $TGT2 -m $MDT0 -u $CL_USER -l $LREPL_LOG \ + -D $LRSYNC_LOG - # Replicate the changes to $TGT - $LRSYNC -s $DIR -t $TGT -t $TGT2 -m $MDT0 -u $CL_USER -l $LREPL_LOG + check_diff $DIR/$tdir $TGT/$tdir + check_diff $DIR/$tdir $TGT2/$tdir - check_diff $DIR/$tdir $TGT/$tdir 3b - check_diff $DIR/$tdir $TGT2/$tdir 3b - - fini_changelog - cleanup_src_tgt - return 0 + fini_changelog + cleanup_src_tgt + return 0 } run_test 3b "Replicate files created by writemany" # Test 3c - Replicate files created by createmany/unlinkmany test_3c() { - [ "$SLOW" = "no" ] && skip "Skipping slow test" && return - - init_src - init_changelog - - local numfiles=1000 - createmany -o $DIR/$tdir/$tfile $numfiles || error "createmany failed!" - unlinkmany $DIR/$tdir/$tfile $numfiles || error "unlinkmany failed!" - - # Replicate the changes to $TGT - $LRSYNC -s $DIR -t $TGT -t $TGT2 -m $MDT0 -u $CL_USER -l $LREPL_LOG - check_diff $DIR/$tdir $TGT/$tdir 3c - check_diff $DIR/$tdir $TGT2/$tdir 3c - - fini_changelog - cleanup_src_tgt - return 0 + init_src + init_changelog + + local numfiles=1000 + createmany -o $DIR/$tdir/$tfile $numfiles || error "createmany failed" + unlinkmany $DIR/$tdir/$tfile $numfiles || error "unlinkmany failed" + + local LRSYNC_LOG=$(generate_logname "lrsync_log") + # Replicate the changes to $TGT + $LRSYNC -s $DIR -t $TGT -t $TGT2 -m $MDT0 -u $CL_USER -l $LREPL_LOG \ + -D $LRSYNC_LOG + check_diff $DIR/$tdir $TGT/$tdir + check_diff $DIR/$tdir $TGT2/$tdir + + fini_changelog + cleanup_src_tgt + return 0 } run_test 3c "Replicate files created by createmany/unlinkmany" # Test 4 - Replicate files created by iozone test_4() { - [ "$SLOW" = "no" ] && skip "Skipping slow test" && return + which iozone > /dev/null 2>&1 + [ $? -ne 0 ] && skip "iozone not found" - which iozone > /dev/null 2>&1 - if [ $? -ne 0 ]; then - skip "iozone not found. Skipping test" - return - fi + init_src + init_changelog - init_src - init_changelog + END_RUN_FILE=${DIR}/$tdir/run LOAD_PID_FILE=${DIR}/$tdir/pid \ + MOUNT=${DIR}/$tdir run_iozone.sh & + sleep 30 + child_pid=$(pgrep iozone) + stop_procs $child_pid - END_RUN_FILE=${DIR}/$tdir/run LOAD_PID_FILE=${DIR}/$tdir/pid \ - MOUNT=${DIR}/$tdir run_iozone.sh & - sleep 30 - child_pid=$(pgrep iozone) - $KILL -SIGSTOP $child_pid - - # Replicate the changes to $TGT - $LRSYNC -s $DIR -t $TGT -t $TGT2 -m $MDT0 -u $CL_USER -l $LREPL_LOG - check_diff $DIR/$tdir $TGT/$tdir 4 - check_diff $DIR/$tdir $TGT2/$tdir 4 + local LRSYNC_LOG=$(generate_logname "lrsync_log") + # Replicate the changes to $TGT + $LRSYNC -s $DIR -t $TGT -t $TGT2 -m $MDT0 -u $CL_USER -l $LREPL_LOG \ + -D $LRSYNC_LOG + check_diff $DIR/$tdir $TGT/$tdir + check_diff $DIR/$tdir $TGT2/$tdir $KILL -SIGCONT $child_pid sleep 60 @@ -416,9 +489,9 @@ test_4() { sleep 1; done - $LRSYNC -l $LREPL_LOG - check_diff $DIR/$tdir $TGT/$tdir 4 - check_diff $DIR/$tdir $TGT2/$tdir 4 + $LRSYNC -l $LREPL_LOG -D $LRSYNC_LOG + check_diff $DIR/$tdir $TGT/$tdir + check_diff $DIR/$tdir $TGT2/$tdir fini_changelog cleanup_src_tgt @@ -428,121 +501,128 @@ run_test 4 "Replicate files created by iozone" # Test 5a - Stop / start lustre_rsync test_5a() { - [ "$SLOW" = "no" ] && skip "Skipping slow test" && return - - init_src - init_changelog - - NUMTEST=2000 - createmany -o $DIR/$tdir/$tfile $NUMTEST - - # Replicate the changes to $TGT - - $LRSYNC -s $DIR -t $TGT -t $TGT2 -m $MDT0 -u $CL_USER -l $LREPL_LOG & - local child_pid=$! - sleep 30 - $KILL -SIGHUP $child_pid - wait - $LRSYNC -l $LREPL_LOG - - check_diff $DIR/$tdir $TGT/$tdir 5a - check_diff $DIR/$tdir $TGT2/$tdir 5a - - fini_changelog - cleanup_src_tgt - return 0 + init_src + init_changelog + + NUMTEST=2000 + createmany -o $DIR/$tdir/$tfile $NUMTEST + + # Replicate the changes to $TGT + local LRSYNC_LOG=$(generate_logname "lrsync_log") + $LRSYNC -s $DIR -t $TGT -t $TGT2 -m $MDT0 -u $CL_USER -l $LREPL_LOG \ + -D $LRSYNC_LOG & + local child_pid=$! + sleep 30 + $KILL -SIGHUP $child_pid + wait + $LRSYNC -l $LREPL_LOG -D $LRSYNC_LOG + + check_diff $DIR/$tdir $TGT/$tdir + check_diff $DIR/$tdir $TGT2/$tdir + + fini_changelog + cleanup_src_tgt + return 0 } run_test 5a "Stop / start lustre_rsync" # Test 5b - Kill / restart lustre_rsync test_5b() { - [ "$SLOW" = "no" ] && skip "Skipping slow test" && return - - init_src - init_changelog - - NUMTEST=2000 - createmany -o $DIR/$tdir/$tfile $NUMTEST - - # Replicate the changes to $TGT - - $LRSYNC -s $DIR -t $TGT -t $TGT2 -m $MDT0 -u $CL_USER -l $LREPL_LOG & - local child_pid=$! - sleep 30 - $KILL -SIGKILL $child_pid - wait - $LRSYNC -l $LREPL_LOG - - check_diff $DIR/$tdir $TGT/$tdir 5b - check_diff $DIR/$tdir $TGT2/$tdir 5b - - fini_changelog - cleanup_src_tgt - return 0 + init_src + init_changelog + + NUMTEST=2000 + createmany -o $DIR/$tdir/$tfile $NUMTEST + + # Replicate the changes to $TGT + local LRSYNC_LOG=$(generate_logname "lrsync_log") + $LRSYNC -s $DIR -t $TGT -t $TGT2 -m $MDT0 -u $CL_USER -l $LREPL_LOG \ + -D $LRSYNC_LOG & + local child_pid=$! + sleep 30 + $KILL -SIGKILL $child_pid + wait + $LRSYNC -l $LREPL_LOG -D $LRSYNC_LOG + + check_diff $DIR/$tdir $TGT/$tdir + check_diff $DIR/$tdir $TGT2/$tdir + + fini_changelog + cleanup_src_tgt + return 0 } run_test 5b "Kill / restart lustre_rsync" # Test 6 - lustre_rsync large no of hard links test_6() { - init_src - init_changelog - - local NUMLINKS=128 - touch $DIR/$tdir/link0 - local i=1 - while [ $i -lt $NUMLINKS ]; - do - ln $DIR/$tdir/link0 $DIR/$tdir/link${i} - i=$(expr $i + 1) - done - - # Replicate the changes to $TGT - $LRSYNC -s $DIR -t $TGT -t $TGT2 -m $MDT0 -u $CL_USER -l $LREPL_LOG - check_diff $DIR/$tdir $TGT/$tdir 6 - check_diff $DIR/$tdir $TGT2/$tdir 6 + init_src + init_changelog + + local num_links=128 + local i + + touch $DIR/$tdir/link0 + for ((i = 1; i < num_links - 1; i++)); do + ln $DIR/$tdir/link0 $DIR/$tdir/link$i + done + # create an extra hard link of src name ending with dest name + ln $DIR/$tdir/link0 $DIR/$tdir/ink0 + + local LRSYNC_LOG=$(generate_logname "lrsync_log") + # Replicate the changes to $TGT + $LRSYNC -s $DIR -t $TGT -t $TGT2 -m $MDT0 -u $CL_USER -l $LREPL_LOG \ + -D $LRSYNC_LOG + check_diff $DIR/$tdir $TGT/$tdir + check_diff $DIR/$tdir $TGT2/$tdir + + local count1=$(stat --format=%h $TGT/$tdir/link0) + local count2=$(stat --format=%h $TGT2/$tdir/link0) + if ((count1 != num_links || count2 != num_links)); then + ls -l $TGT/$tdir/link0 $TGT2/$tdir/link0 + error "Incorrect no of hard links found $count1, $count2" + fi - local count1=$(ls -l $TGT/$tdir/link0 | sed -r 's/ +/ /g' | cut -f 2 -d ' ') - local count2=$(ls -l $TGT/$tdir/link0 | sed -r 's/ +/ /g' | cut -f 2 -d ' ') - if [[ $count1 -ne $NUMLINKS ]] || [[ $count2 -ne $NUMLINKS ]]; then - ls -l $TGT/$tdir/link0 $TGT2/$tdir/link0 - error "Incorrect no of hard links found $count1, $count2" - fi - fini_changelog - cleanup_src_tgt - return 0 + fini_changelog + cleanup_src_tgt + return 0 } run_test 6 "lustre_rsync large no of hard links" # Test 7 - lustre_rsync stripesize test_7() { - init_src - mkdir -p ${DIR}/tgt/$tdir - init_changelog - - local NUMFILES=100 - lfs setstripe -c $OSTCOUNT $DIR/$tdir - createmany -o $DIR/$tdir/$tfile $NUMFILES - - # To simulate replication to another lustre filesystem, replicate - # the changes to $DIR/tgt. We can't turn off the changelogs - # while we are registered, so lustre_rsync better not try to - # replicate the replication steps. It seems ok :) - - $LRSYNC -s $DIR -t $DIR/tgt -m $MDT0 -u $CL_USER -l $LREPL_LOG - check_diff ${DIR}/$tdir $DIR/tgt/$tdir 7 - - local i=0 - while [ $i -lt $NUMFILES ]; - do - local count=$(lfs getstripe $DIR/tgt/$tdir/${tfile}$i | awk '/stripe_count/ {print $2}') - if [ $count -ne $OSTCOUNT ]; then - error "Stripe size not replicated" - fi - i=$(expr $i + 1) - done - fini_changelog - cleanup_src_tgt - return 0 + local numfiles=100 + + init_src + mkdir -p ${DIR}/tgt/$tdir + init_changelog + + $LFS setstripe -c $OSTCOUNT $DIR/$tdir || + error "$LFS setstripe failed" + createmany -o $DIR/$tdir/$tfile $numfiles + + # To simulate replication to another lustre filesystem, replicate + # the changes to $DIR/tgt. We can't turn off the changelogs + # while we are registered, so lustre_rsync better not try to + # replicate the replication steps. It seems ok :) + + local LRSYNC_LOG=$(generate_logname "lrsync_log") + $LRSYNC -s $DIR -t $DIR/tgt -m $MDT0 -u $CL_USER -l $LREPL_LOG \ + -D $LRSYNC_LOG + check_diff ${DIR}/$tdir $DIR/tgt/$tdir + + local i=0 + while [ $i -lt $numfiles ]; + do + local count=$($LFS getstripe $DIR/tgt/$tdir/${tfile}$i | + awk '/stripe_count/ {print $2}') + if [ $count -ne $OSTCOUNT ]; then + error "Stripe size not replicated" + fi + i=$(expr $i + 1) + done + fini_changelog + cleanup_src_tgt + return 0 } run_test 7 "lustre_rsync stripesize" @@ -565,9 +645,11 @@ test_8() { mv $DIR/$tdir/d$i $DIR/$tdir/d0$i done - $LRSYNC -s $DIR -t $TGT -m $MDT0 -u $CL_USER -l $LREPL_LOG + local LRSYNC_LOG=$(generate_logname "lrsync_log") + $LRSYNC -s $DIR -t $TGT -m $MDT0 -u $CL_USER -l $LREPL_LOG \ + -D $LRSYNC_LOG - check_diff ${DIR}/$tdir $TGT/$tdir 8 + check_diff ${DIR}/$tdir $TGT/$tdir fini_changelog cleanup_src_tgt @@ -582,15 +664,18 @@ test_9() { mkdir $DIR/$tdir/foo touch $DIR/$tdir/foo/a1 - $LRSYNC -s $DIR -t $TGT -m $MDT0 -u $CL_USER -l $LREPL_LOG + local LRSYNC_LOG=$(generate_logname "lrsync_log") + $LRSYNC -s $DIR -t $TGT -m $MDT0 -u $CL_USER -l $LREPL_LOG \ + -D $LRSYNC_LOG - check_diff ${DIR}/$tdir $TGT/$tdir 9 + check_diff ${DIR}/$tdir $TGT/$tdir - rm -rf $DIR/$tdir/foo + rm -rf $DIR/$tdir/foo - $LRSYNC -s $DIR -t $TGT -m $MDT0 -u $CL_USER -l $LREPL_LOG + $LRSYNC -s $DIR -t $TGT -m $MDT0 -u $CL_USER -l $LREPL_LOG \ + -D $LRSYNC_LOG - check_diff ${DIR}/$tdir $TGT/$tdir 9 + check_diff ${DIR}/$tdir $TGT/$tdir fini_changelog cleanup_src_tgt @@ -599,6 +684,6 @@ test_9() { run_test 9 "Replicate recursive directory removal" cd $ORIG_PWD -complete $(basename $0) $SECONDS +complete $SECONDS check_and_cleanup_lustre exit_status