lustre-iokit/mds-survey/mds-survey

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