X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Ftests%2Flustre-rsync-test.sh;h=9c8658dba1038ba4dcb22c2b55bed166bb1b000c;hb=7b2dcd2c348d1b3da59095260edbd31383baea30;hp=b091763421f1217c925f2bd02ccf2c65efcc0078;hpb=3edfe6f6a322df4de44332c45f9eb95ac7db1e51;p=fs%2Flustre-release.git

diff --git a/lustre/tests/lustre-rsync-test.sh b/lustre/tests/lustre-rsync-test.sh
index b091763..9c8658d 100644
--- a/lustre/tests/lustre-rsync-test.sh
+++ b/lustre/tests/lustre-rsync-test.sh
@@ -12,8 +12,8 @@ SRCDIR=`dirname $0`
 export PATH=$PWD/$SRCDIR:$SRCDIR:$PWD/$SRCDIR/../utils:$PATH:/sbin
 
 ONLY=${ONLY:-"$*"}
-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!
 
 [ "$SLOW" = "no" ] && EXCEPT_SLOW=""
@@ -38,6 +38,13 @@ check_and_setup_lustre
 DIR=${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
 
@@ -45,44 +52,40 @@ export LRSYNC=${LRSYNC:-"$LUSTRE/utils/lustre_rsync"}
 [ ! -f "$LRSYNC" ] && export LRSYNC=$(which lustre_rsync)
 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 -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.
@@ -121,8 +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)
@@ -164,7 +199,7 @@ test_1() {
 
     # 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
@@ -178,39 +213,29 @@ test_1() {
 	echo "Replication #2"
 	$LRSYNC -l $LREPL_LOG -D $LRSYNC_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
-
-    # 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
 	check_diff $DIR/$tdir $TGT2/$tdir
 
-    fini_changelog
-    cleanup_src_tgt
-    return $RC
+	fini_changelog
+	cleanup_src_tgt
 }
-run_test 1 "Simple Replication"
+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
@@ -243,6 +268,7 @@ test_1a() { # LU-5005
 	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-*
@@ -273,7 +299,6 @@ test_2a() {
 }
 run_test 2a "Replicate files created by dbench."
 
-
 # Test 2b - Replicate files changed by dbench.
 test_2b() {
 	local child_pid
@@ -289,7 +314,7 @@ test_2b() {
 
 	echo PIDs: $child_pid
 	echo Stopping dbench
-	$KILL -SIGSTOP $child_pid
+	stop_procs $child_pid
 
 	local LRSYNC_LOG=$(generate_logname "lrsync_log")
 	echo Starting replication
@@ -302,7 +327,7 @@ test_2b() {
     sleep 10
 
     echo Stopping dbench
-    $KILL -SIGSTOP $child_pid
+	stop_procs $child_pid
 
 	echo Starting replication
 	$LRSYNC -l $LREPL_LOG -D $LRSYNC_LOG
@@ -380,7 +405,6 @@ test_3a() {
 }
 run_test 3a "Replicate files created by createmany"
 
-
 # Test 3b - Replicate files created by writemany
 test_3b() {
 	init_src
@@ -429,20 +453,17 @@ run_test 3c "Replicate files created by createmany/unlinkmany"
 
 # Test 4 - Replicate files created by iozone
 test_4() {
-    which iozone > /dev/null 2>&1
-    if [ $? -ne 0 ]; then
-	skip "iozone not found. Skipping test"
-	return
-    fi
+	which iozone > /dev/null 2>&1
+	[ $? -ne 0 ] && skip "iozone not found"
 
-    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)
-    $KILL -SIGSTOP $child_pid
+	END_RUN_FILE=${DIR}/$tdir/run LOAD_PID_FILE=${DIR}/$tdir/pid \
+		MOUNT=${DIR}/$tdir run_iozone.sh &
+	sleep 30
+	child_pid=$(pgrep iozone)
+	stop_procs $child_pid
 
 	local LRSYNC_LOG=$(generate_logname "lrsync_log")
 	# Replicate the changes to $TGT
@@ -534,17 +555,18 @@ run_test 5b "Kill / restart lustre_rsync"
 
 # Test 6 - lustre_rsync large no of hard links
 test_6() {
-    init_src
-    init_changelog
+	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
+	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
@@ -553,27 +575,30 @@ test_6() {
 	check_diff $DIR/$tdir $TGT/$tdir
 	check_diff $DIR/$tdir $TGT2/$tdir
 
-    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, $count2"
-    fi
-    fini_changelog
-    cleanup_src_tgt
-    return 0
+	local count1=$(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
+
+	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
+
+	init_src
+	mkdir -p ${DIR}/tgt/$tdir
+	init_changelog
 
-    local NUMFILES=100
-    lfs setstripe -c $OSTCOUNT $DIR/$tdir
-    createmany -o $DIR/$tdir/$tfile $NUMFILES
+	$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
@@ -586,9 +611,9 @@ test_7() {
 	check_diff ${DIR}/$tdir $DIR/tgt/$tdir
 
 	local i=0
-	while [ $i -lt $NUMFILES ];
+	while [ $i -lt $numfiles ];
 	do
-		local count=$(lfs getstripe $DIR/tgt/$tdir/${tfile}$i | \
+		local count=$($LFS getstripe $DIR/tgt/$tdir/${tfile}$i |
 			      awk '/stripe_count/ {print $2}')
 		if [ $count -ne $OSTCOUNT ]; then
 			error "Stripe size not replicated"