lustre-iokit/mds-survey/mds-survey

   1 #!/bin/bash
   2 # SPDX-License-Identifier: GPL-2.0
   3
   4 #
   5 # This file is part of Lustre, http://www.lustre.org/
   6 #
   7
   8 ######################################################################
   9 # customize per survey
  10
  11 # Prerequisite: For "stripe_count > 0" you need to have ost setup and mounted.
  12 #
  13 # How to run test:
  14 # case 1 (stripe_count=0 default):
  15 #  $ thrhi=8 dir_count=4 sh mds-survey
  16 #  one can also run test with user defined targets as follows,
  17 #  $ thrhi=8 dir_count=4 file_count=50000 targets="lustre-MDT0000" sh mds-survey
  18 # case 2 (stripe_count > 0, must have ost mounted):
  19 #  $ thrhi=8 dir_count=4 file_count=50000 stripe_count=2
  20 #  targets="lustre-MDT0000" sh mds-survey
  21 # [ NOTE: It is advised to have automated login (passwordless entry) on server ]
  22
  23 # include library
  24 source $(dirname $0)/iokit-libecho
  25
  26 # Customisation variables
  27 #####################################################################
  28 # One can change variable values in this section as per requirements
  29 # The following variables can be set in the environment, or on the
  30 # command line
  31 # result file prefix (date/time + hostname makes unique)
  32 # NB ensure path to it exists
  33 rslt_loc=${rslt_loc:-"/tmp"}
  34 rslt=${rslt:-"$rslt_loc/mds_survey_`date +%F@%R`_`uname -n`"}
  35
  36 # min and max thread count
  37 thrlo=${thrlo:-4}
  38 thrhi=${thrhi:-32}
  39
  40 # number of directories to test
  41 dir_count=${dir_count:-$thrlo}
  42 # number of files per thread
  43 file_count=${file_count:-100000}
  44
  45 targets=${targets:-""}
  46 stripe_count=${stripe_count:-0}
  47 # what tests to run (first must be create, and last must be destroy)
  48 # default=(create lookup md_getattr setxattr destroy)
  49 tests_str=${tests_str:-"create lookup md_getattr setxattr destroy"}
  50
  51 # start number for each thread
  52 start_number=${start_number:-2}
  53
  54 # layer to be tested
  55 layer=${layer:-"mdd"}
  56 # Customisation variables ends here.
  57 #####################################################################
  58 # leave the rest of this alone unless you know what you're doing...
  59 export LC_ALL=POSIX
  60 basedir="tests"
  61 mdtbasedir="MDT%04x-"
  62
  63 create_directories () {
  64         local host=$1
  65         local devno=$2
  66         local ndir=$3
  67         local rfile=$4
  68         local mdtidx=$5
  69         local dir_stripes=$6
  70         local idx
  71
  72         for ((idx = 0; idx < $ndir; idx++)); do
  73                 if (( idx == 0 )); then
  74                         dirname="$(printf "${mdtbasedir}" $mdtidx)${basedir}"
  75                 else
  76                         dirname="$(printf "${mdtbasedir}" $mdtidx)${basedir}${idx}"
  77                 fi
  78                 remote_shell $host $LCTL --device $devno test_mkdir /$dirname \
  79                         -c $dir_stripes --stripe_index $mdtidx > $rfile 2>&1
  80                 while read line; do
  81                         echo "$line" | grep -q 'error: test_mkdir'
  82                         if [ $?  -eq 0 ]; then
  83                                 cat $rfile >&2
  84                                 echo "ERROR: fail test_mkdir" >&2
  85                                 echo "ERROR"
  86                                 return
  87                         fi
  88                 done < $rfile
  89         done
  90         echo $basedir
  91 }
  92
  93 destroy_directories () {
  94         local host=$1
  95         local devno=$2
  96         local ndir=$3
  97         local rfile=$4
  98         local mdtidx=$5
  99         local idx
 100
 101         for ((idx = 0; idx < $ndir; idx++)); do
 102                 if (( idx == 0 )); then
 103                         dirname="$(printf "${mdtbasedir}" $mdtidx)${basedir}"
 104                 else
 105                         dirname="$(printf "${mdtbasedir}" $mdtidx)${basedir}${idx}"
 106                 fi
 107                 remote_shell $host $LCTL --device $devno test_rmdir /$dirname > $rfile 2>&1
 108         done
 109 }
 110
 111 get_stats () {
 112         local rfile=$1
 113
 114         gawk < $rfile                                                   \
 115         '/starting/ {                                                   \
 116                 n = 0; next;                                            \
 117         }                                                               \
 118         /error/ {                                                       \
 119                 n = -1;                                                 \
 120                 exit;                                                   \
 121         }                                                               \
 122         /^Total: total [0-9]+ threads [0-9]+ sec [0-9\.]+ [0-9]+\.[0-9]+\/second$/ { \
 123                 ave = strtonum($8);                                     \
 124                 n++;                                                    \
 125                 next;                                                   \
 126         }                                                               \
 127         /^[0-9]+\/[0-9]+ Total: [0-9]+\.[0-9]+\/second$/ {              \
 128                 n++; v = strtonum($3);                                  \
 129                 if (n == 1 || v < min) min = v;                         \
 130                 if (n == 1 || v > max) max = v;                         \
 131                 next;                                                   \
 132         }                                                               \
 133         {                                                               \
 134                 if (n != 0) {                                           \
 135                         n = -1;                                         \
 136                         exit;                                           \
 137                 }                                                       \
 138         }                                                               \
 139         END {                                                           \
 140                 if (n == 1) {                                           \
 141                         min = ave;                                      \
 142                         max = ave;                                      \
 143                 }                                                       \
 144                 printf "%d %f %f %f\n", n, ave, min, max                \
 145         }'
 146 }
 147
 148 get_global_stats () {
 149         local rfile=$1
 150
 151         awk < $rfile                                                    \
 152         'BEGIN {                                                        \
 153                 n = 0;                                                  \
 154         }                                                               \
 155         {                                                               \
 156                 n++;                                                    \
 157                 if (n == 1) {                                           \
 158                         err = $1;                                       \
 159                         ave = $2;                                       \
 160                         min = $3;                                       \
 161                         max = $4;                                       \
 162                 } else {                                                \
 163                         if ($1 < err)                                   \
 164                                 err = $1;                               \
 165                         ave += $2;                                      \
 166                         if ($3 < min)                                   \
 167                                 min = $3;                               \
 168                         if ($4 > max)                                   \
 169                                 max = $4;                               \
 170                 }                                                       \
 171         }                                                               \
 172         END {                                                           \
 173                 if (n == 0)                                             \
 174                         err = 0;                                        \
 175                 printf "%d %f %f %f\n", err, ave/n, min, max            \
 176         }'
 177 }
 178
 179 print_summary () {
 180         if [ "$1" = "-n" ]; then
 181                 minusn=$1; shift
 182         else
 183                 minusn=""
 184         fi
 185         echo $minusn "$*" >> $rsltf
 186         echo $minusn "$*"
 187 }
 188
 189 declare -a tests
 190 count=0
 191 for name in $tests_str; do
 192         tests[$count]=$name
 193         count=$((count+1))
 194 done
 195
 196 # hide a little trick to unset this from the command line
 197 if [ "$lustre_root" == " " ]; then
 198         unset lustre_root
 199 fi
 200
 201 # find where 'lctl' binary is installed on this system
 202 if [[ -x "$LCTL" ]]; then       # full pathname specified
 203         : # echo "LCTL=$LCTL"
 204 elif [[ -n "$lustre_root" && -x "$lustre_root/utils/lctl" ]]; then
 205         LCTL=${lustre_root}/utils/lctl
 206 elif [[ -n "$LUSTRE" && -x "$LUSTRE/utils/lctl" ]]; then
 207         LCTL=$LUSTRE/utils/lctl
 208 else                            # hope that it is in the PATH
 209         LCTL=${LCTL:-lctl}
 210 fi
 211 [[ -n "$(which $LCTL)" ]] || { echo "error: lctl not found"; exit 99; }
 212
 213 declare -a client_names
 214 declare -a host_names
 215 declare -a client_indexes
 216 if [ -z "$targets" ]; then
 217         targets=$($LCTL device_list | awk "{if (\$2 == \"UP\" && \
 218                                                \$3 == \"mdt\") {print \$4} }")
 219         if [ -z "$targets" ]; then
 220                 echo "Can't find any MDT to test.  Please set targets=..."
 221                 exit 1
 222         fi
 223 fi
 224
 225 # split out hostnames from mdt names
 226 ndevs=0
 227 for trgt in $targets; do
 228         str=($(split_hostname $trgt))
 229         host_names[$ndevs]=${str[0]}
 230         client_names[$ndevs]=${str[1]}
 231         client_indexes[$ndevs]=0x$(echo ${str[1]} |
 232                 sed 's/.*MDT\([0-9a-f][0-9a-f][0-9a-f][0-9a-f]\).*/\1/')
 233         ndevs=$((ndevs+1))
 234 done
 235
 236 # check for ost
 237 if (( $stripe_count > 0 )); then
 238         for ((i=0; i < $ndevs; i++)); do
 239                 host=${host_names[$i]}
 240                 obd=$(remote_shell $host $LCTL device_list |
 241                       awk "{ if (\$2 == \"UP\" &&
 242                                  (\$3 == \"osc\" || \$3 == \"osp\"))
 243                                 { print \$4 } }")
 244                 if [ -z "$obd" ]; then
 245                         echo "Need obdfilter to test stripe_count"
 246                         exit 1
 247                 fi
 248         done
 249 fi
 250
 251 # check and insert obdecho module
 252 if ! lsmod | grep obdecho > /dev/null; then
 253         modprobe obdecho
 254 fi
 255 count=${#tests[@]}
 256 if [ $count -eq 0 -o "${tests[0]}" != "create" -o "${tests[(($count - 1))]}" != "destroy" ]; then
 257         echo "tests: ${tests[@]}"
 258         echo "First test must be 'create', and last test must be 'destroy'" 1>&2
 259         exit 1
 260 fi
 261
 262 rsltf="${rslt}.summary"
 263 workf="${rslt}.detail"
 264 cmdsf="${rslt}.script"
 265 vmstatf="${rslt}.vmstat"
 266 echo -n > $rsltf
 267 echo -n > $workf
 268
 269 # get vmstat started
 270 # disable portals debug and get obdecho loaded on all relevant hosts
 271 unique_hosts=($(unique ${host_names[@]}))
 272 load_obdechos
 273 pidcount=0
 274 for host in ${unique_hosts[@]}; do
 275         host_vmstatf=${vmstatf}_${host}
 276         echo -n > $host_vmstatf
 277         remote_shell $host "vmstat 5 >> $host_vmstatf" &> /dev/null &
 278         pid=$!
 279         vmstatpids[$pidcount]=$pid
 280         pidcount=$((pidcount+1))
 281 done
 282 # get all the echo_client device numbers and names
 283 for ((i=0; i < $ndevs; i++)); do
 284         host=${host_names[$i]}
 285         devno=($(get_ec_devno $host "${client_names[$i]}" "${client_names[$i]}" "mdt" $layer))
 286         if ((${#devno[@]} != 3)); then
 287                 exit 1
 288         fi
 289         devnos[$i]=${devno[0]}
 290         client_names[$i]=${devno[1]}
 291         do_teardown_ec[$i]=${devno[2]}
 292 done
 293 if (($ndevs <= 0 || ${#host_names[@]} <= 0)); then
 294         echo "no devices or hosts specified"
 295         cleanup 0
 296 fi
 297 print_summary "$(date) $0 from $(hostname)"
 298 # create directories
 299 tmpf="${workf}_tmp"
 300 for ((idx = 0; idx < $ndevs; idx++)); do
 301         host=${host_names[$idx]}
 302         devno=${devnos[$idx]}
 303         client_name="${host}:${client_names[$idx]}"
 304         mdtidx=${client_indexes[$idx]}
 305         echo "=======> Create $dir_count directories on $client_name" >> $workf
 306         destroy_directories $host $devno $dir_count $tmpf $mdtidx
 307         ret=$(create_directories $host $devno $dir_count $tmpf $mdtidx $ndevs)
 308         cat $tmpf >> $workf
 309         rm $tmpf
 310         if [ $ret = "ERROR" ]; then
 311                 print_summary "created directories on $client_name failed"
 312                 cleanup 1
 313         fi
 314 done
 315
 316 snap=1
 317 status=0
 318 for ((thr = $thrlo; thr <= $thrhi; thr*=2)); do
 319         thr_per_dir=$((${thr}/${dir_count}))
 320         # skip if no enough thread
 321         if (( thr_per_dir <= 0 )); then
 322                 continue
 323         fi
 324         file_count_per_thread=$((${file_count}/${thr}))
 325         str=$(printf 'mdt %1d file %7d dir %4d thr %4d ' \
 326               $ndevs $file_count $dir_count $thr)
 327         echo "=======> $str" >> $workf
 328         print_summary -n "$str"
 329         # run tests
 330         for test in ${tests[@]}; do
 331                 declare -a pidarray
 332                 for host in ${unique_hosts[@]}; do
 333                         echo "starting run for config: $config test: $test " \
 334                              "file: $file_count threads: $thr " \
 335                              "directories: $dir_count" >> ${vmstatf}_${host}
 336                 done
 337                 print_summary -n "$test "
 338                 # create per-host script files
 339                 for host in ${unique_hosts[@]}; do
 340                         echo -n > ${cmdsf}_${host}
 341                 done
 342                 for ((idx = 0; idx < $ndevs; idx++)); do
 343                         host=${host_names[$idx]}
 344                         devno=${devnos[$idx]}
 345                         dirname="$(printf "${mdtbasedir}" ${client_indexes[$idx]})$basedir"
 346                         tmpfi="${tmpf}_$idx"
 347                         [ "$test" = "create" ] && test="create -c $stripe_count"
 348                         echo >> ${cmdsf}_${host}                        \
 349                                 "$LCTL > $tmpfi 2>&1                    \
 350                                 --threads $thr -$snap $devno test_$test \
 351                                 -d /$dirname -D $dir_count              \
 352                                 -b $start_number -n $file_count_per_thread"
 353                 done
 354                 pidcount=0
 355                 for host in ${unique_hosts[@]}; do
 356                         echo "wait" >> ${cmdsf}_${host}
 357                         pidarray[$pidcount]=0
 358                         pidcount=$((pidcount+1))
 359                 done
 360                 pidcount=0
 361                 for host in ${unique_hosts[@]}; do
 362                         remote_shell $host bash < ${cmdsf}_${host} &
 363                         pidarray[$pidcount]=$!
 364                         pidcount=$((pidcount+1))
 365                 done
 366                 pidcount=0
 367                 for host in ${unique_hosts[@]}; do
 368                         wait ${pidarray[$pidcount]}
 369                         pidcount=$((pidcount+1))
 370                 done
 371                 #wait
 372                 # clean up per-host script files
 373                 for host in ${unique_hosts[@]}; do
 374                         rm ${cmdsf}_${host}
 375                 done
 376
 377                 # collect/check individual MDT stats
 378                 echo -n > $tmpf
 379                 for ((idx = 0; idx < $ndevs; idx++)); do
 380                         client_name="${host_names[$idx]}:${client_names[$idx]}"
 381                         tmpfi="${tmpf}_$idx"
 382                         echo "=============> $test $client_name" >> $workf
 383                         host="${host_names[$idx]}"
 384                         remote_shell $host cat $tmpfi > ${tmpfi}_local
 385                         cat ${tmpfi}_local >> $workf
 386                         get_stats ${tmpfi}_local >> $tmpf
 387                         rm -f $tmpfi ${tmpfi}_local
 388                 done
 389                 # compute/display global min/max stats
 390                 echo "=============> $test global" >> $workf
 391                 cat $tmpf >> $workf
 392                 stats=($(get_global_stats $tmpf))
 393                 rm $tmpf
 394                 if ((stats[0] <= 0)); then
 395                         str=$(printf "%17s " ERROR)
 396                         status=1
 397                 else
 398                         str=$(awk "BEGIN {printf \"%7.2f [ %7.2f, %7.2f] \", \
 399                               ${stats[1]}, ${stats[2]}, ${stats[3]}; exit}")
 400                 fi
 401                 print_summary -n "$str"
 402         done
 403         print_summary ""
 404 done
 405
 406 # destroy directories
 407 tmpf="${workf}_tmp"
 408 for ((idx = 0; idx < $ndevs; idx++)); do
 409         host=${host_names[$idx]}
 410         devno=${devnos[$idx]}
 411         mdtidx=${client_indexes[$idx]}
 412         client_name="${host}:${client_names[$idx]}"
 413         echo "====> Destroy $dir_count directories on $client_name" >> $workf
 414         destroy_directories $host $devno $dir_count $tmpf $mdtidx
 415 done
 416
 417 cleanup $status
 418 exit $status