X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Ftests%2Ffunctions.sh;h=0bc4fa0e279482a5e86176c2830dfbe2c40a2a1d;hp=9e6976d26807b13d5c73cc83d0efc958ea3ba0aa;hb=c394068ee148595711c661651368b91373bcc78a;hpb=4ac9f1ad9ca68de0df7ce0d6451c29a5ff3be1b9

diff --git a/lustre/tests/functions.sh b/lustre/tests/functions.sh
index 9e6976d..0bc4fa0 100644
--- a/lustre/tests/functions.sh
+++ b/lustre/tests/functions.sh
@@ -1,4 +1,6 @@
 #!/bin/bash
+# -*- mode: Bash; tab-width: 4; indent-tabs-mode: t; -*-
+# vim:shiftwidth=4:softtabstop=4:tabstop=4:
 
 # Simple function used by run_*.sh scripts
 
@@ -252,6 +254,19 @@ short_hostname() {
   echo $(sed 's/\..*//' <<< $1)
 }
 
+###
+# short_nodename
+#
+# Find remote nodename, stripped of any domain, etc.
+# 'hostname -s' is easy, but not implemented on all systems
+short_nodename() {
+	local rname=$(do_node $1 "uname -n" || echo -1)
+	if [[ "$rname" = "-1" ]]; then
+		rname=$1
+	fi
+	echo $(short_hostname $rname)
+}
+
 print_opts () {
     local var
 
@@ -265,18 +280,13 @@ print_opts () {
 }
 
 run_compilebench() {
-
-# Takes:
-# 5 min * cbench_RUNS
-#        SLOW=no     10 mins
-#        SLOW=yes    50 mins
-# Space estimation:
-#        compile dir kernel-1 680MB
-#        required space       680MB * cbench_IDIRS = ~7 Gb
+	# Space estimation:
+	# compile dir kernel-0	~1GB
+	# required space	~1GB * cbench_IDIRS
 
     cbench_DIR=${cbench_DIR:-""}
-    cbench_IDIRS=${cbench_IDIRS:-4}
-    cbench_RUNS=${cbench_RUNS:-4}
+    cbench_IDIRS=${cbench_IDIRS:-2}
+    cbench_RUNS=${cbench_RUNS:-2}
 
     print_opts cbench_DIR cbench_IDIRS cbench_RUNS
 
@@ -286,14 +296,16 @@ run_compilebench() {
     [ -e $cbench_DIR/compilebench ] || \
         { skip_env "No compilebench build" && return; }
 
-    local space=$(df -P $DIR | tail -n 1 | awk '{ print $4 }')
-    if [ $space -le $((680 * 1024 * cbench_IDIRS)) ]; then
-        cbench_IDIRS=$(( space / 680 / 1024))
-        [ $cbench_IDIRS = 0 ] && \
-            skip_env "Need free space atleast 680 Mb, have $space" && return
+	local space=$(df -P $DIR | tail -n 1 | awk '{ print $4 }')
+	if [[ $space -le $((1024 * 1024 * cbench_IDIRS)) ]]; then
+		cbench_IDIRS=$((space / 1024 / 1024))
+		[[ $cbench_IDIRS -eq 0 ]] &&
+			skip_env "Need free space at least 1GB, have $space" &&
+			return
+
+		echo "free space=$space, reducing initial dirs to $cbench_IDIRS"
+	fi
 
-        log free space=$space, reducing initial dirs to $cbench_IDIRS
-    fi
     # FIXME:
     # t-f _base needs to be modifyed to set properly tdir
     # for new "test_foo" functions names
@@ -344,14 +356,15 @@ run_metabench() {
     local cmd="$METABENCH -w $testdir -c $mbench_NFILES -C -S -k"
     echo "+ $cmd"
 
-    # find out if we need to use srun by checking $SRUN_PARTITION
-    if [ "$SRUN_PARTITION" ]; then
-        $SRUN $SRUN_OPTIONS -D $testdir -w $clients -N $num_clients \
-            -n $((num_clients * mbench_THREADS)) -p $SRUN_PARTITION -- $cmd
-    else
-        mpi_run -np $((num_clients * $mbench_THREADS)) \
-            -machinefile ${MACHINEFILE} $cmd
-    fi
+	# find out if we need to use srun by checking $SRUN_PARTITION
+	if [ "$SRUN_PARTITION" ]; then
+		$SRUN $SRUN_OPTIONS -D $testdir -w $clients -N $num_clients \
+			-n $((num_clients * mbench_THREADS)) \
+			-p $SRUN_PARTITION -- $cmd
+	else
+		mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \
+			-np $((num_clients * $mbench_THREADS)) $cmd
+	fi
 
     local rc=$?
     if [ $rc != 0 ] ; then
@@ -390,15 +403,16 @@ run_simul() {
 
     local cmd="$SIMUL -d $testdir -n $simul_REP -N $simul_REP"
 
-    echo "+ $cmd"
-    # find out if we need to use srun by checking $SRUN_PARTITION
-    if [ "$SRUN_PARTITION" ]; then
-        $SRUN $SRUN_OPTIONS -D $testdir -w $clients -N $num_clients \
-            -n $((num_clients * simul_THREADS)) -p $SRUN_PARTITION -- $cmd
-    else
-        mpi_run -np $((num_clients * simul_THREADS)) \
-            -machinefile ${MACHINEFILE} $cmd
-    fi
+	echo "+ $cmd"
+	# find out if we need to use srun by checking $SRUN_PARTITION
+	if [ "$SRUN_PARTITION" ]; then
+		$SRUN $SRUN_OPTIONS -D $testdir -w $clients -N $num_clients \
+			-n $((num_clients * simul_THREADS)) -p $SRUN_PARTITION \
+			-- $cmd
+	else
+		mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \
+			-np $((num_clients * simul_THREADS)) $cmd
+	fi
 
     local rc=$?
     if [ $rc != 0 ] ; then
@@ -445,15 +459,16 @@ run_mdtest() {
     local cmd="$MDTEST -d $testdir -i $mdtest_iteration -n $mdtest_nFiles"
     [ $type = "fpp" ] && cmd="$cmd -u"
 
-    echo "+ $cmd"
-    # find out if we need to use srun by checking $SRUN_PARTITION
-    if [ "$SRUN_PARTITION" ]; then
-        $SRUN $SRUN_OPTIONS -D $testdir -w $clients -N $num_clients \
-            -n $((num_clients * mdtest_THREADS)) -p $SRUN_PARTITION -- $cmd
-    else
-        mpi_run -np $((num_clients * mdtest_THREADS)) \
-            -machinefile ${MACHINEFILE} $cmd
-    fi
+	echo "+ $cmd"
+	# find out if we need to use srun by checking $SRUN_PARTITION
+	if [ "$SRUN_PARTITION" ]; then
+		$SRUN $SRUN_OPTIONS -D $testdir -w $clients -N $num_clients \
+			-n $((num_clients * mdtest_THREADS)) \
+			-p $SRUN_PARTITION -- $cmd
+	else
+		mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \
+			-np $((num_clients * mdtest_THREADS)) $cmd
+	fi
 
     local rc=$?
     if [ $rc != 0 ] ; then
@@ -578,15 +593,16 @@ run_ior() {
          -t $ior_xferSize -v -w -r -i $ior_iteration -T $ior_DURATION -k"
     [ $type = "fpp" ] && cmd="$cmd -F"
 
-    echo "+ $cmd"
-    # find out if we need to use srun by checking $SRUN_PARTITION
-    if [ "$SRUN_PARTITION" ]; then
-        $SRUN $SRUN_OPTIONS -D $testdir -w $clients -N $num_clients \
-            -n $((num_clients * ior_THREADS)) -p $SRUN_PARTITION -- $cmd
-    else
-        mpi_run -np $((num_clients * $ior_THREADS)) \
-            -machinefile ${MACHINEFILE} $cmd
-    fi
+	echo "+ $cmd"
+	# find out if we need to use srun by checking $SRUN_PARTITION
+	if [ "$SRUN_PARTITION" ]; then
+		$SRUN $SRUN_OPTIONS -D $testdir -w $clients -N $num_clients \
+			-n $((num_clients * ior_THREADS)) -p $SRUN_PARTITION \
+			-- $cmd
+	else
+		mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \
+			-np $((num_clients * $ior_THREADS)) $cmd
+	fi
 
     local rc=$?
     if [ $rc != 0 ] ; then
@@ -631,15 +647,16 @@ run_mib() {
     local cmd="$MIB -t $testdir -s $mib_xferSize -l $mib_xferLimit \
         -L $mib_timeLimit -HI -p mib.$(date +%Y%m%d%H%M%S)"
 
-    echo "+ $cmd"
-    # find out if we need to use srun by checking $SRUN_PARTITION
-    if [ "$SRUN_PARTITION" ]; then
-        $SRUN $SRUN_OPTIONS -D $testdir -w $clients -N $num_clients \
-            -n $((num_clients * mib_THREADS)) -p $SRUN_PARTITION -- $cmd
-    else
-        mpi_run -np $((num_clients * mib_THREADS)) \
-            -machinefile ${MACHINEFILE} $cmd
-    fi
+	echo "+ $cmd"
+	# find out if we need to use srun by checking $SRUN_PARTITION
+	if [ "$SRUN_PARTITION" ]; then
+		$SRUN $SRUN_OPTIONS -D $testdir -w $clients -N $num_clients \
+			-n $((num_clients * mib_THREADS)) -p $SRUN_PARTITION \
+			-- $cmd
+	else
+		mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \
+			-np $((num_clients * mib_THREADS)) $cmd
+	fi
 
     local rc=$?
     if [ $rc != 0 ] ; then
@@ -678,9 +695,9 @@ run_cascading_rw() {
 
     local cmd="$CASC_RW -g -d $testdir -n $casc_REP"
 
-    echo "+ $cmd"
-    mpi_run -np $((num_clients * $casc_THREADS)) \
-        -machinefile ${MACHINEFILE} $cmd
+	echo "+ $cmd"
+	mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \
+		-np $((num_clients * $casc_THREADS)) $cmd
 
     local rc=$?
     if [ $rc != 0 ] ; then
@@ -720,9 +737,9 @@ run_write_append_truncate() {
 
     local cmd="write_append_truncate -n $write_REP $file"
 
-    echo "+ $cmd"
-    mpi_run -np $((num_clients * $write_THREADS)) \
-        -machinefile ${MACHINEFILE} $cmd
+	echo "+ $cmd"
+	mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \
+		-np $((num_clients * $write_THREADS)) $cmd
 
     local rc=$?
     if [ $rc != 0 ] ; then
@@ -734,8 +751,8 @@ run_write_append_truncate() {
 
 run_write_disjoint() {
 
-    WRITE_DISJOINT=${WRITE_DISJOINT:-\
-        $(which write_disjoint 2> /dev/null || true)}
+    WRITE_DISJOINT=${WRITE_DISJOINT:-$(which write_disjoint \
+        2> /dev/null || true)}
     # threads per client
     wdisjoint_THREADS=${wdisjoint_THREADS:-4}
     wdisjoint_REP=${wdisjoint_REP:-10000}
@@ -760,9 +777,9 @@ run_write_disjoint() {
 
     local cmd="$WRITE_DISJOINT -f $testdir/file -n $wdisjoint_REP"
 
-    echo "+ $cmd"
-    mpi_run -np $((num_clients * $wdisjoint_THREADS)) \
-        -machinefile ${MACHINEFILE} $cmd
+	echo "+ $cmd"
+	mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \
+		-np $((num_clients * $wdisjoint_THREADS)) $cmd
 
     local rc=$?
     if [ $rc != 0 ] ; then
@@ -773,8 +790,8 @@ run_write_disjoint() {
 
 run_parallel_grouplock() {
 
-    PARALLEL_GROUPLOCK=${PARALLEL_GROUPLOCK:-\
-        $(which parallel_grouplock 2> /dev/null || true)}
+    PARALLEL_GROUPLOCK=${PARALLEL_GROUPLOCK:-$(which parallel_grouplock \
+        2> /dev/null || true)}
     parallel_grouplock_MINTASKS=${parallel_grouplock_MINTASKS:-5}
 
     if [ "$NFSCLIENT" ]; then
@@ -798,25 +815,26 @@ run_parallel_grouplock() {
     local cmd
     local status=0
     local subtest
-    for i in $(seq 12); do
-        subtest="-t $i"
-        local cmd="$PARALLEL_GROUPLOCK -g -v -d $testdir $subtest"
-        echo "+ $cmd"
-
-        mpi_run -np $parallel_grouplock_MINTASKS \
-            -machinefile ${MACHINEFILE} $cmd
-        local rc=$?
-        if [ $rc != 0 ] ; then
-            error_noexit "parallel_grouplock subtests $subtest failed! $rc"
-        else
-            echo "parallel_grouplock subtests $subtest PASS"
-        fi
-        let status=$((status + rc))
-        # clear debug to collect one log per one test
-        do_nodes $(comma_list $(nodes_list)) lctl clear
-     done
-    [ $status -eq 0 ] || error "parallel_grouplock status: $status"
-    rm -rf $testdir
+	for i in $(seq 12); do
+		subtest="-t $i"
+		local cmd="$PARALLEL_GROUPLOCK -g -v -d $testdir $subtest"
+		echo "+ $cmd"
+
+		mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \
+			-np $parallel_grouplock_MINTASKS $cmd
+		local rc=$?
+		if [ $rc != 0 ] ; then
+			error_noexit "parallel_grouplock subtests $subtest " \
+				     "failed! $rc"
+		else
+			echo "parallel_grouplock subtests $subtest PASS"
+		fi
+		let status=$((status + rc))
+		# clear debug to collect one log per one test
+		do_nodes $(comma_list $(nodes_list)) lctl clear
+	done
+	[ $status -eq 0 ] || error "parallel_grouplock status: $status"
+	rm -rf $testdir
 }
 
 cleanup_statahead () {
@@ -879,7 +897,8 @@ run_statahead () {
     local cmd="$cmd1 $cmd2"
     echo "+ $cmd"
 
-    mpi_run -np $((num_clients * 32)) -machinefile ${MACHINEFILE} $cmd
+	mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \
+		-np $((num_clients * 32)) $cmd
 
     local rc=$?
     if [ $rc != 0 ] ; then