lustre/tests/dom-performance.sh

   1 #!/bin/bash
   2 #
   3 # Run select tests by setting ONLY, or as arguments to the script.
   4 # Skip specific tests by setting EXCEPT.
   5 #
   6
   7 set -e
   8
   9 ONLY=${ONLY:-"$*"}
  10 ALWAYS_EXCEPT=${ALWAYS_EXCEPT:-"$DOM_PERFORMANCE_EXCEPT"}
  11 # UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT!
  12
  13 [ "$SLOW" = "no" ] && EXCEPT_SLOW=""
  14
  15 LUSTRE=${LUSTRE:-$(cd $(dirname $0)/..; echo $PWD)}
  16
  17 . $LUSTRE/tests/test-framework.sh
  18
  19 init_test_env $@
  20 . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
  21 init_logging
  22
  23 SAVED_FAIL_ON_ERROR=$FAIL_ON_ERROR
  24 FAIL_ON_ERROR=false
  25
  26 SAVED_DEBUG=$($LCTL get_param -n debug 2> /dev/null)
  27
  28
  29 . $LUSTRE/tests/functions.sh
  30 build_test_filter
  31 check_and_setup_lustre
  32
  33 # if MACHINEFILE set and exists -- use it
  34 MACHINEFILE=${MACHINEFILE:-$TMP/$(basename $0 .sh)-$(hostname).machines}
  35 clients=${CLIENTS:-$HOSTNAME}
  36 generate_machine_file $clients $MACHINEFILE ||
  37         error "Failed to generate machine file"
  38
  39 DP_DIO=${DP_DIO:-"no"}
  40
  41 DOM_SIZE=${DOM_SIZE:-"1M"}
  42 DP_OSC="mdc"
  43
  44 DP_NORM=$DIR/dp_norm
  45 DP_DOM=$DIR/dp_dom
  46 DP_DOM_DNE=$DIR/dp_dne
  47 DP_STATS=${DP_STATS:-"no"}
  48
  49 if $DO_CLEANUP; then
  50         rm -rf $DIR/*
  51 else
  52         rm -rf $DP_NORM $DP_DOM $DP_DOM_DNE
  53 fi
  54
  55 # total number of files
  56 DP_FNUM=${DP_FNUM:-16384}
  57 # number of threads
  58 DP_NUM=${DP_NUM:-4}
  59
  60 # 1 stripe for normal files
  61 mkdir -p $DP_NORM
  62 $LFS setstripe -c 2 $DP_NORM ||
  63         error "Cannot create test directory for ordinary files"
  64
  65 if [[ $MDSCOUNT -gt 1 ]] ; then
  66         $LFS setdirstripe -i 0 -c $MDSCOUNT $DP_DOM_DNE ||
  67                 error_noexit "Cannot create striped directory"
  68         $LFS setstripe -E ${DOM_SIZE} -L mdt -E EOF $DP_DOM_DNE ||
  69                 error_noexit "Cannot create test directory for dom files"
  70 fi
  71
  72 mkdir -p $DP_DOM
  73 $LFS setstripe -E ${DOM_SIZE} -L mdt -E EOF $DP_DOM ||
  74         error "Cannot create test directory for dom files"
  75
  76 dp_clear_stats() {
  77         local cli=$1
  78
  79         $LCTL set_param -n osc.*.stats=0
  80         $LCTL set_param -n mdc.*.stats=0
  81         $LCTL set_param -n ${cli}.*.${cli}_stats=0
  82         $LCTL set_param -n ${cli}.*.rpc_stats=0
  83         $LCTL set_param -n llite.*.read_ahead_stats=0
  84         $LCTL set_param -n llite.*.unstable_stats=0
  85 }
  86
  87 dp_collect_stats() {
  88         local cli=$1
  89
  90         sync;sync
  91         echo ----- MDC RPCs: $(calc_stats mdc.*.stats req_active)
  92         echo ----- OSC RPCs: $(calc_stats osc.*.stats req_active)
  93
  94         if [ "x$DP_STATS" != "xyes" ] ; then
  95                 return 0
  96         fi
  97
  98         $LCTL get_param ${cli}.*.${cli}_stats
  99         $LCTL get_param ${cli}.*.rpc_stats
 100         # for OSC get both OSC and MDC stats
 101         if [ $cli == "osc" ] ; then
 102                 $LCTL get_param mdc.*.stats
 103         fi
 104         $LCTL get_param ${cli}.*.stats
 105         $LCTL get_param ${cli}.*.unstable_stats
 106         $LCTL get_param ${cli}.*.${cli}_cached_mb
 107         $LCTL get_param llite.*.read_ahead_stats
 108 }
 109
 110 dp_setup_test() {
 111         local cli=$1
 112
 113         cancel_lru_locks $cli
 114         ### drop all debug except critical
 115         $LCTL set_param -n debug="error warning console emerg"
 116         dp_clear_stats $cli
 117 }
 118
 119 dp_run_cmd() {
 120         local cmd=$1
 121         local cmdlog=$TMP/dp_cmd.log
 122         local rc
 123
 124         dp_setup_test $DP_OSC
 125         if ! grep -qw "$MOUNT" /proc/mounts ; then
 126                 echo "!!!!! Lustre is not mounted !!!!!, aborting"
 127                 return 0
 128         fi
 129
 130         echo "## $cmd" | awk '{ if (NR==1) {gsub(/[ \t\r\n]+/, " "); \
 131                                 gsub(/\|.*$/, ""); print }}'
 132         echo "## $(date +'%F %H:%M:%S'): START"
 133         eval $cmd 2>&1 | tee $cmdlog || true
 134
 135         rc=${PIPESTATUS[0]}
 136         if [ $rc -eq 0 ] && grep -q "p4_error:" $cmdlog ; then
 137                 rc=1
 138         fi
 139
 140         dp_collect_stats $DP_OSC
 141         remount_client $DIR > /dev/null
 142         return $rc
 143 }
 144
 145 run_MDtest() {
 146         if ! which mdtest > /dev/null 2>&1 ; then
 147                 skip_env "Mdtest is not installed, skipping"
 148         fi
 149
 150         local mdtest=$(which mdtest)
 151
 152         local TDIR=${1:-$MOUNT}
 153         local th_num=$((DP_FNUM * 2 / DP_NUM))
 154         local bsizes="8192"
 155
 156         chmod 0777 $TDIR
 157
 158         [ "$SLOW" = "yes" ] && bsizes="4096 32768"
 159
 160         for bsize in $bsizes ; do
 161                 dp_run_cmd "mpi_run -np $DP_NUM $mdtest -i 3 -I $th_num -F \
 162                         -z 1 -b 1 -L -u -w $bsize -R -d $TDIR"
 163                 if [ ${PIPESTATUS[0]} != 0 ]; then
 164                         error "MDtest failed, aborting"
 165                 fi
 166         done
 167
 168         rm -rf $TDIR/*
 169         return 0
 170 }
 171
 172 run_SmallIO() {
 173         local TDIR=${1:-$DIR}
 174         local count=$DP_FNUM
 175
 176         local MIN=$((count * 16))
 177         local mdssize=$(mdssize_from_index $TDIR 0)
 178         [ $mdssize -le $MIN ] && count=$((mdssize / 16))
 179
 180         dp_run_cmd "createmany -o $TDIR/file- $count | grep 'total:'"
 181         if [ ${PIPESTATUS[0]} != 0 ]; then
 182                 error "File creation failed, aborting"
 183         fi
 184
 185         dp_run_cmd "statmany -s $TDIR/file- $count $((count * 5)) |
 186                 grep 'total:'"
 187         if [ ${PIPESTATUS[0]} != 0 ]; then
 188                 error "File stat failed, aborting"
 189         fi
 190
 191         for opc in w a r ; do
 192                 dp_run_cmd "smalliomany -${opc} $TDIR/file- $count 300 |
 193                         grep 'total:'"
 194                 if [ ${PIPESTATUS[0]} != 0 ]; then
 195                         error "SmallIO -${opc} failed, aborting"
 196                 fi
 197
 198         done
 199
 200         dp_run_cmd "unlinkmany $TDIR/file- $count | grep 'total:'"
 201         if [ ${PIPESTATUS[0]} != 0 ]; then
 202                 error "SmallIO failed, aborting"
 203         fi
 204
 205         return 0
 206 }
 207
 208 run_IOR() {
 209         if ! which IOR > /dev/null 2>&1 ; then
 210                 skip_env "IOR is not installed, skipping"
 211         fi
 212
 213         local IOR=$(which IOR)
 214         local iter=$((DP_FNUM / DP_NUM))
 215         local direct=""
 216
 217         if [ "x$DP_DIO" == "xyes" ] ; then
 218                 direct="-B"
 219         fi
 220
 221         local TDIR=${1:-$MOUNT}
 222
 223         chmod 0777 $TDIR
 224
 225         local bsizes="8"
 226         [ "$SLOW" = "yes" ] && bsizes="4 32"
 227
 228         for bsize in $bsizes ; do
 229                 segments=$((128 / bsize))
 230
 231                 dp_run_cmd "mpi_run -np $DP_NUM $IOR \
 232                         -a POSIX -b ${bsize}K -t ${bsize}K -o $TDIR/ -k \
 233                         -s $segments -w -r -i $iter -F -E -z -m -Z $direct" |
 234                         awk '($1 !~ /^(write|read|access)$/) || NF>12 {print}'
 235                 if [ ${PIPESTATUS[0]} != 0 ]; then
 236                         error "IOR write test for ${bsize}K failed, aborting"
 237                 fi
 238
 239                 # check READ performance only (no cache)
 240                 dp_run_cmd "mpi_run -np $DP_NUM $IOR \
 241                         -a POSIX -b ${bsize}K -t ${bsize}K -o $TDIR/ -X 42\
 242                         -s $segments -r -i $iter -F -E -z -m -Z $direct" |
 243                         awk '($1 !~ /^(read|access|remove)$/) || NF>12 {print}'
 244                 if [ ${PIPESTATUS[0]} != 0 ]; then
 245                         error "IOR read test for ${bsize}K failed, aborting"
 246                 fi
 247
 248         done
 249         rm -rf $TDIR/*
 250         return 0
 251 }
 252
 253 run_Dbench() {
 254         if ! which dbench > /dev/null 2>&1 ; then
 255                 skip_env "Dbench is not installed, skipping"
 256         fi
 257
 258         local TDIR=${1:-$MOUNT}
 259
 260         if [ "x$DP_DOM_DNE" == "x$TDIR" ] ; then
 261                 echo "dbench uses subdirs, skipping for DNE dir"
 262                 return 0
 263         fi
 264
 265         dp_run_cmd "dbench -D $TDIR $DP_NUM | egrep -v 'warmup|execute'"
 266         if [ ${PIPESTATUS[0]} != 0 ]; then
 267                 error "Dbench failed, aborting"
 268         fi
 269
 270         rm -rf $TDIR/*
 271         return 0
 272 }
 273
 274 run_FIO() {
 275         # https://github.com/axboe/fio/archive/fio-2.8.zip
 276         if ! which fio > /dev/null 2>&1 ; then
 277                 skip_env "No FIO installed, skipping"
 278         fi
 279
 280         local fnum=128 # per thread
 281         local total=$((fnum * DP_NUM)) # files in all threads
 282         local loops=$((DP_FNUM / total)) # number of loops
 283         local direct=""
 284         local output=""
 285
 286         if [ $loops -eq 0 ] ; then
 287                 loops=1
 288         fi
 289
 290         if [ "x$DP_DIO" == "xyes" ] ; then
 291                 direct="--direct=1"
 292         else
 293                 direct="--buffered=1 --bs_unaligned=1"
 294         fi
 295
 296         if [ "x$DP_STATS" != "xyes" ] ; then
 297                 output="--minimal"
 298         fi
 299
 300         local TDIR=${1:-$MOUNT}
 301         base_cmd="fio --name=smallio --ioengine=posixaio $output \
 302                   --iodepth=$((DP_NUM * 4)) --directory=$TDIR \
 303                   --nrfiles=$fnum --openfiles=10000 \
 304                   --numjobs=$DP_NUM --filesize=64k --lockfile=readwrite"
 305
 306         dp_run_cmd "$base_cmd --create_only=1" > /dev/null
 307         if [ ${PIPESTATUS[0]} != 0 ]; then
 308                 error "FIO file creation failed, aborting"
 309         fi
 310
 311         local bsizes="8"
 312         [ "$SLOW" = "yes" ] && bsizes="4 32"
 313
 314         for bsize in $bsizes ; do
 315                 local write_cmd="$base_cmd --bs=${bsize}k --rw=randwrite \
 316                         $direct --file_service_type=random --randrepeat=1 \
 317                          --norandommap --group_reporting=1 --loops=$loops"
 318                 if [ "x$DP_STATS" != "xyes" ] ; then
 319                         dp_run_cmd "$write_cmd | awk -F\; '{printf \"WRITE: \
 320                                 BW %dKiB/sec, IOPS %d, lat (%d/%d/%d)usec\n\", \
 321                                 \$48, \$49, \$53, \$57, \$81}'"
 322                 else
 323                         dp_run_cmd "$write_cmd"
 324                 fi
 325                 if [ ${PIPESTATUS[0]} != 0 ]; then
 326                         error "FIO write test with ${bsize}k failed, aborting"
 327                 fi
 328
 329                 local read_cmd="$base_cmd --bs=${bsize}k --rw=randread \
 330                         $direct --file_service_type=random --randrepeat=1 \
 331                          --norandommap --group_reporting=1 --loops=$loops"
 332                 if [ "x$DP_STATS" != "xyes" ] ; then
 333                         dp_run_cmd "$read_cmd | awk -F\; '{printf \"READ : \
 334                                 BW %dKiB/sec, IOPS %d, lat (%d/%d/%d)usec\n\", \
 335                                 \$7, \$8, \$12, \$16, \$40}'"
 336                 else
 337                         dp_run_cmd "$read_cmd"
 338                 fi
 339                 if [ ${PIPESTATUS[0]} != 0 ]; then
 340                         error "FIO read test with ${bsize}k failed, aborting"
 341                 fi
 342         done
 343         rm -rf $TDIR/*
 344         return 0
 345 }
 346
 347 run_compbench() {
 348         local compilebench
 349         if [ x$cbench_DIR = x ]; then
 350                 compilebench=$(which compilebench 2> /dev/null)
 351         else
 352                 cd $cbench_DIR
 353                 [ -x compilebench ] ||
 354                         skip_env "compilebench is missing in $cbench_DIR"
 355                 compilebench=compilebench
 356         fi
 357
 358         [ x$compilebench != x ] ||
 359                 skip_env "Compilebench is not installed, skipping"
 360
 361         local TDIR=${1:-$MOUNT}
 362
 363         dp_run_cmd "$compilebench -D $TDIR -i 2 -r 2 --makej"
 364         if [ ${PIPESTATUS[0]} != 0 ]; then
 365                 error "Compilebench failed, aborting"
 366         fi
 367
 368         rm -rf $TDIR/*
 369 }
 370
 371 dp_test_run() {
 372         local test=$1
 373         local facets=$(get_facets MDS)
 374         local nodes=$(comma_list $(mdts_nodes))
 375         local p="$TMP/$TESTSUITE-$TESTNAME.parameters"
 376
 377         save_lustre_params $facets "mdt.*.dom_lock" >> $p
 378
 379         printf "\n##### $test: DoM files\n"
 380         do_nodes $nodes "lctl set_param -n mdt.*.dom_lock=1"
 381         DP_OSC="mdc"
 382         run_${test} $DP_DOM
 383
 384         if [ -d $DP_DOM_DNE ] ; then
 385                 printf "\n##### $test: DoM files + DNE\n"
 386                 DP_OSC="mdc"
 387                 run_${test} $DP_DOM_DNE
 388         fi
 389
 390         printf "\n##### $test: OST files\n"
 391         DP_OSC="osc"
 392         run_${test} $DP_NORM
 393
 394         restore_lustre_params < $p
 395         rm -f $p
 396 }
 397
 398 test_smallio() {
 399         dp_test_run SmallIO
 400 }
 401 run_test smallio "Performance comparision: smallio"
 402
 403 test_mdtest() {
 404         dp_test_run MDtest
 405 }
 406 run_test mdtest "Performance comparision: mdtest"
 407
 408 test_IOR() {
 409         dp_test_run IOR
 410 }
 411 run_test IOR "Performance comparision: IOR"
 412
 413 test_dbench() {
 414         dp_test_run Dbench
 415 }
 416 run_test dbench "Performance comparision: dbench"
 417
 418 test_fio() {
 419         dp_test_run FIO
 420 }
 421 run_test fio "Performance comparision: FIO"
 422
 423 test_compbench() {
 424         dp_test_run compbench
 425 }
 426 run_test compbench "Performance comparision: compilebench"
 427
 428 FAIL_ON_ERROR=$SAVED_FAIL_ON_ERROR
 429 $LCTL set_param -n debug="$SAVED_DEBUG"
 430
 431 complete $SECONDS
 432 check_and_cleanup_lustre
 433 exit_status