Viewing: parallel-scale-cifs.sh
#!/usr/bin/bash
# Requires the pre-configured samba machine
# RPMS required are :
# server:
# samba
# samba-common
# cifs-utils
# clients:
# samba-client
# samba-common
# cifs-utils
#set -vx
LUSTRE=${LUSTRE:-$(dirname $0)/..}
. $LUSTRE/tests/test-framework.sh
init_test_env "$@"
init_logging
. $LUSTRE/tests/setup-cifs.sh
# lustre client used as samba server (default is mds node)
LUSTRE_CLIENT_SMBSRV=${LUSTRE_CLIENT_SMBSRV:-$(facet_active_host $SINGLEMDS)}
SMBSHARE=${SMBSHARE:-lustretest}
SMBUSER=${SMBUSER:-root}
SMBPASSWD=${SMBPASSWD:-lustre}
SMBSRVMNTPT=${SMBSRVMNTPT:-$MOUNT}
SMBCLIMNTPT=${SMBCLIMNTPT:-$MOUNT}
SMBCLIENTS=${SMBCLIENTS:-$CLIENTS}
SMBCLIENTS=$(exclude_items_from_list $SMBCLIENTS $LUSTRE_CLIENT_SMBSRV)
[ -z "$SMBCLIENTS" ] &&
skip_env "need at least two nodes: samba server and samba client"
do_nodes $SMBCLIENTS modinfo cifs | grep dummy > /dev/null &&
skip_env "OFED installation caused CIFS to break in RHEL8.4 mlnx 5.4"
check_and_setup_lustre
# first unmount all the lustre clients
cleanup_mount $MOUNT
# set CONFIGURE_SMB=false to skip smb config
CONFIGURE_SMB=${CONFIGURE_SMB:-true}
# store smb status to restart smb service if it was running initially
SMBSTATUS=0
smb_status $LUSTRE_CLIENT_SMBSRV || SMBSTATUS=$?
SMBCONFTMP=$(do_node $LUSTRE_CLIENT_SMBSRV "mktemp -t smb.conf.XXX")
cleanup_exit() {
trap 0
cleanup
check_and_cleanup_lustre
exit
}
cleanup() {
cleanup_cifs $LUSTRE_CLIENT_SMBSRV $SMBCLIMNTPT $SMBCLIENTS ||
error_noexit false "failed to cleanup cifs"
zconf_umount $LUSTRE_CLIENT_SMBSRV $SMBSRVMNTPT force ||
error_noexit false "failed to umount lustre on $LUSTRE_CLIENT_SMBSRV"
# restore lustre mount
restore_mount $MOUNT ||
error_noexit false "failed to mount lustre"
$CONFIGURE_SMB && restore_config_smb $LUSTRE_CLIENT_SMBSRV $SMBCONFTMP
[[ $SMBSTATUS -eq 0 ]] &&
do_node $LUSTRE_CLIENT_SMBSRV "service smb start"
unset CIFSCLIENT
}
$CONFIGURE_SMB && configure_smb $LUSTRE_CLIENT_SMBSRV $SMBSHARE $SMBUSER \
$SMBPASSWD $SMBSRVMNTPT $SMBCONFTMP ||
echo -e "\nSkipping smb config ..."
trap cleanup_exit EXIT SIGHUP SIGINT
# mount lustre client on smb server
zconf_mount $LUSTRE_CLIENT_SMBSRV $SMBSRVMNTPT ||
error "mount lustre on $LUSTRE_CLIENT_SMBSRV failed"
# setup the cifs
setup_cifs $LUSTRE_CLIENT_SMBSRV $SMBSHARE $SMBCLIMNTPT $SMBUSER \
$SMBPASSWD $SMBCLIENTS ||
error false "setup cifs failed"
CIFSCLIENT=yes
FAIL_ON_ERROR=false
# compilbench
# Run short iteration in cifs mode
cbench_IDIRS=${cbench_IDIRS:-2}
cbench_RUNS=${cbench_RUNS:-2}
# source the common file after all parameters are set to take effect
. $LUSTRE/tests/functions.sh
build_test_filter
check_prog_output() {
local clients=$1
local file=$2
local str=$3
do_nodes $clients grep -q \\\"$str\\\" $file 2>/dev/null
}
wait_prog_output() {
local clients=$1
local file=$2
local str=$3
local time=$4
local start_ts=$(date +%s)
local elapsed
while ! check_prog_output $clients $file "$str"; do
elapsed=$(($(date +%s) - start_ts))
if [ $elapsed -gt $time ]; then
return 1
fi
sleep 1
done
}
test_compilebench() {
run_compilebench $SMBCLIMNTPT
}
run_test compilebench "compilebench on cifs clients"
test_dbench() {
local clients=$SMBCLIENTS
local duration=${DBENCH_DURATION:-300}
local nproc=${DBENCH_NPROC:-1}
local delay=${dbench_STARTDELAY:-120}
local log=$TMP/dbench.log
local pid=""
local cmd="rundbench $nproc -t $duration"
echo "Using: $cmd"
do_nodesv $clients "set -x; MISSING_DBENCH_OK=$MISSING_DBENCH_OK \
PATH=\$PATH DBENCH_LIB=$DBENCH_LIB \
TESTSUITE=$TESTSUITE TESTNAME=$TESTNAME \
DIR=$SMBCLIMNTPT/$tdir/\\\$(hostname) \
LCTL=$LCTL $cmd 2>&1 | tee $log; \
exit \\\${PIPESTATUS[0]}" &
pid=$!
# check that dbench is started on all clients after
# $dbench_STARTDELAY: the dbench log on each client
# is to be started for this moment and contain "dbench PID";
if ! wait_prog_output $clients $log "dbench PID" $delay; then
kill -s TERM $pid
killall_process $clients dbench
error "dbench failed to start on $clients!"
fi
log "Started rundbench load pid=$pid ..."
wait $pid || error "rundbench load on $clients failed!"
}
run_test dbench "dbench on cifs clients"
test_fsx() {
local clients=$SMBCLIENTS
local seed=${fsx_SEED:-$RANDOM}
local size=${fsx_SIZE:-1024}
local numop=${fsx_NUMOP:-100000}
local delay=${fsx_STARTDELAY:-120}
local log=$TMP/fsx.log
local pid=""
local nclients=$(get_node_count ${clients//,/ })
local space=$(df -P $SMBCLIMNTPT | tail -n 1 | awk '{ print $4 }')
[ $space -lt $((size * nclients)) ] && size=$((space * 3 / 4 / nclients))
check_set_fallocate
local cmd="$FSX -c 50 -p 500 -S $seed -P $TMP -l $size -N $numop "
echo "Using: $cmd"
do_nodesv $clients "set -x; \
PATH=\$PATH \
$cmd $SMBCLIMNTPT/f0.fsx_\\\$(hostname) 2>&1 | tee $log; \
exit \\\${PIPESTATUS[0]}" &
pid=$!
# check that fsx is started on all clients after
# $fsx_STARTDELAY: the fsx log on each client
# is to be started for this moment and contain "Seed set";
if ! wait_prog_output $clients $log "Seed set" $delay; then
kill -s TERM $pid
killall_process $clients fsx
error "fsx failed to start on $clients!"
fi
log "Started fsx load pid=$pid ..."
wait $pid || error "fsx load on $clients failed!"
}
run_test fsx "fsx on cifs clients"
test_iozone() {
local clients=$SMBCLIENTS
local size=${iozone_SIZE:-262144} # 256m
local delay=${iozone_STARTDELAY:-120}
local log=$TMP/iozone.log
local pid=""
local nclients=$(get_node_count ${clients//,/ })
local space=$(df -P $SMBCLIMNTPT | tail -n 1 | awk '{ print $4 }')
[[ $((size * nclients)) -gt $((space * 3 / 4)) ]] &&
size=$((space * 3 / 4 / nclients))
do_node $LUSTRE_CLIENT_SMBSRV "mkdir $SMBSRVMNTPT/$tdir
lfs setstripe -c -1 $SMBSRVMNTPT/$tdir"
log "free space: $space Kb, using $size size, $nclients number of clients"
local cmd="iozone -a -e -+d -s $size "
echo "Using: $cmd"
do_nodesv $clients "set -x; \
PATH=\$PATH \
$cmd -f $SMBCLIMNTPT/$tdir/f0.iozone_\\\$(hostname) \
2>&1 | tee $log; exit \\\${PIPESTATUS[0]}" &
pid=$!
# check that iozone is started on all clients after
# $iozone_STARTDELAY: the iozone log on each client
# is to be started for this moment and contain "Command line used";
if ! wait_prog_output $clients $log "Command line used" $delay; then
kill -s TERM $pid
killall_process $clients iozone
error "iozone failed to start on $clients!"
fi
log "Started iozone load pid=$pid ..."
wait $pid
rc=$?
log "Processing iozone log"
do_nodesv $clients "tail -1 $log | grep -q complete" || rc=2
do_node $LUSTRE_CLIENT_SMBSRV "rm -rf $SMBSRVMNTPT/$tdir"
[ $rc -eq 0 ] || error "iozone load on $clients failed! rc=$rc"
}
run_test iozone "iozone on cifs clients"
complete_test $SECONDS
exit_status
