3 ######################################################################
6 # specify obd instances to exercise
7 # these can be either...
8 # obdfilter instances (set 'ost_names')
10 # echo_client instances (set 'client_names')
11 # ... use 'host:name' for obd instances on other nodes.
12 # allow these to be passed in via string...
14 # one can specify only case=disk or case=network or case=netdisk through
17 # Perquisite: For "disk" case and "netdisk" case you need to have lustre setup
18 # with one or more ost's. For "network" case you need to have all
19 # modules (those llmount.sh loades) loaded in kernel. And the
20 # 'lctl dl' output must be blank.
23 # case 1 (local disk):
24 # $ nobjhi=2 thrhi=2 size=1024 case=disk sh obdfilter-survey
25 # one can also run test with user defined targets as follows,
26 # $ nobjhi=2 thrhi=2 size=1024 targets="lustre-OST0000 lustre-OST0001 ..." sh obdfilter-survey
28 # $ nobjhi=2 thrhi=2 size=1024 targets="<name/ip_of_server>" case=network sh obdfilter-survey
29 # where, targets is name or ip address of system, which you want to
31 # case 3 (network and disk):
32 # $ nobjhi=2 thrhi=2 size=1024 case=netdisk sh obdfilter-survey
33 # one can also run test with user defined targets as follows,
34 # $ nobjhi=2 thrhi=2 size=1024 targets="<osc_name> ..." sh obdfilter-survey
35 #[ NOTE: It is advised to have automated login (passwordless entry) between server and
36 # client systems on which this test runs.]
39 source $(dirname $0)/libecho
41 # The following variables can be set in the environment, or on the
43 # result file prefix (date/time + hostname makes unique)
44 # NB ensure path to it exists
45 rslt_loc=${rslt_loc:-"/tmp"}
46 rslt=${rslt:-"$rslt_loc/obdfilter_survey_`date +%F@%R`_`uname -n`"}
48 # Set this true to check file contents
52 targets=${targets:-""}
56 # total size (MBytes) per obd instance
57 # large enough to avoid cache effects
58 # and to make test startup/shutdown overhead insignificant
61 # record size (KBytes) ( 7168 max)
65 # number of objects per OST
67 #was nobjhi=${nobjhi:-512}
70 # threads per OST (1024 max)
78 # create a set of objects, check there are 'n' contiguous ones and
79 # echo out the first or 'ERROR'
80 # parameter: 1. hostname
82 # 3. number of object to be created (specified by user)
89 remote_shell $host $lctl --device $devno create $nobj > $rfile 2>&1
95 # Count number of objects (lines containing " is object id "), and
96 # ensure that the objects numbers are sequential.
99 while read -u3 line; do
101 ( *' is object id '* )
103 if test $(( count += 1 )) -gt 1 ; then
104 (( $6 != prev + 1 )) && error=1
114 if [ $nobj -ne $count ]; then
115 echo "ERROR: $nobj != $count" >&2
118 elif [ $error -ne 0 ]; then
119 echo "ERROR: non contiguous objs found" >&2
127 # destroys all objects created in create_objects routine
128 # parameter: 3. start obj id.
135 remote_shell $host $lctl --device $devno destroy $obj0 $nobj > $rfile 2>&1
141 '/^Selected device [0-9]+$/ {n = 0; next}\
142 /error/ {n = -1; exit}\
144 /^[0-9]+\/[0-9]+ Total: [0-9]+\.[0-9]+\/second$/ {n++; v=strtonum($3); \
145 if (n == 1 || v < min) min = v;\
146 if (n == 1 || v > max) max = v;\
148 {if (n != 0) {n = -1; exit}}\
149 END {printf "%d %f %f\n", n, min, max}'
152 get_global_stats () {
154 awk < $rfile 'BEGIN {n = 0;}\
155 {n++; if (n == 1) {err = $1; min = $2; max = $3} else\
156 {if ($1 < err) err = $1;\
157 if ($2 < min) min = $2;\
158 if ($3 > max) max = $3}}\
159 END {if (n == 0) err = 0;\
160 printf "%d %f %f\n", err, min, max}'
163 # enable or disable data check.
164 # parameter: 1. read/write
166 # 'x' disables data check
173 *write*) echo "w$x";;
178 # for "echo_client + obdfilter" case, "prep + commit" mode should be used
179 # for "echo_client + osc" case, "BRW" mode should be used
188 if [ "$1" = "-n" ]; then
193 echo $minusn "$*" >> $rsltf
197 # Customisation variables
198 #####################################################################
199 # One can change variable values in this section as per requirements
201 if [ -n "$targets" ]; then
203 declare -a client_names
205 for name in $targets; do
206 if [ $case == "disk" ]; then
207 ost_names[$count]=$name
209 client_names[$count]=$name
215 # what tests to run (first must be write)
216 tests_str=${tests_str:-""}
217 if [ -n "$tests_str" ]; then
220 for name in $tests_str; do
225 #tests=(write rewrite read reread rewrite_again)
226 tests=(write rewrite read)
229 # restart from here iff all are defined
234 # machine's page size (K)
235 if [ -z "$PAGE_SIZE" ]; then
236 if which python >/dev/null; then
237 PAGE_SIZE=`echo 'import resource; print resource.getpagesize()/1024;' |python`
240 PAGE_SIZE=${PAGE_SIZE:-4}
242 # max buffer_mem (total_threads * buffer size)
243 # (to avoid lctl ENOMEM problems)
244 max_buffer_mem=$((1024 * 1024))
247 # Customisation variables ends here.
248 #####################################################################
249 # leave the rest of this alone unless you know what you're doing...
251 # check and insert obdecho module
252 if ! lsmod | grep obdecho > /dev/null; then
255 if [ ${#tests[@]} -eq 0 -o "${tests[0]}" != "write" ]; then
256 echo "tests: ${tests[@]}"
257 echo "First test must be 'write'" 1>&2
261 rsltf="${rslt}.summary"
262 workf="${rslt}.detail"
263 cmdsf="${rslt}.script"
264 vmstatf="${rslt}.vmstat"
268 # hide a little trick to unset this from the command line
269 if [ "$lustre_root" == " " ]; then
273 if [ -z "$lustre_root" ]; then
276 lctl=${lustre_root}/utils/lctl
279 # split out hostnames from client/ost names
281 for trgt in $targets; do
282 str=(`split_hostname $trgt`)
283 host_names[$ndevs]=${str[0]}
284 client_names[$ndevs]=${str[1]}
287 if [ $case == "disk" ]; then
288 if [ $rszhi -gt 1024 ]; then
289 echo "Test disk case support maximum 1024KB IO data" \
290 "(rszhi=$rszhi is too big) please use a smaller value."
293 for ((i = 0; i < $ndevs; i++)); do
294 ost_names[$i]=${client_names[$i]}
297 if [ $case == "netdisk" ]; then
298 if [ "$targets" ]; then
299 for ((i = 0; i < $ndevs; i++)); do
300 setup_osc_for_remote_ost ${host_names[$i]} \
301 ${client_names[$i]} $i
302 osc_name=${client_names[$i]}_osc
303 ec_using_osc $osc_name
304 cleanup_oscs="$cleanup_oscs $osc_name"
307 client_names_str=$($lctl dl | grep -v mdt | \
308 awk '{if ($2 == "UP" && $3 == "osc") {print $4} }')
310 for name in $client_names_str; do
311 client_names[$count]=`echo $name | sed 's/-osc-.*$//'`
315 host_names_str=$($lctl dl -t | grep -v mdt | \
316 awk '{if ($2 == "UP" && $3 == "osc") {print $7} }')
318 for name in $host_names_str; do
319 host_names[$count]=`echo $name | sed 's/@.*$//'`
323 for (( i = 0; i < $count; i++ )) do
324 setup_osc_for_remote_ost ${host_names[$i]} \
325 ${client_names[$i]} $i
326 osc_name=${client_names[$i]}_osc
327 ec_using_osc $osc_name
328 cleanup_oscs="$cleanup_oscs $osc_name"
332 echo_clients=$($lctl dl | grep echo_client | awk "{if (\$2 == \"UP\" && \$3 == \"echo_client\") {print \$4} }")
334 for name in $echo_clients; do
335 client_names[$cnt]=$name
336 host_names[$cnt]=localhost
339 ndevs=${#client_names[@]}
341 if [ $case == "network" ]; then
343 if [ -z "$server_nid" ]; then
344 echo "Specify hostname or ip-address of server"
347 # check for obdecho module on server
348 if ! dsh $server_nid root "lsmod | grep obdecho > /dev/null"; then
349 dsh $server_nid root "modprobe obdecho"
351 # Now do the server setup
352 setup_srv_obd $server_nid "echo_srv"
353 oss_on_srv=`dsh $server_nid root "$lctl dl | grep OSS" | awk '{ print $4 }'`
354 if [ -z $oss_on_srv ]; then
355 setup_OSS $server_nid
358 if ! dsh $server_nid root "$lctl dl | grep obdecho > /dev/null 2>&1"; then
359 echo "obdecho not setup on server"
362 if ! dsh $server_nid root "$lctl dl | grep ost > /dev/null 2>&1"; then
363 echo "ost not setup on server"
366 # Now start client setup
367 osc_names_str=$($lctl dl| grep osc | grep -v mdt | grep UP)
368 if [ -n "$osc_names_str" ]; then
369 echo "The existing setup must be cleaned";
372 ec_using_srv_nid $server_nid "echotmp" "echotmp_UUID"
373 client_names[0]="echotmp_ecc"
375 if [ -z "$targets" ]; then
376 if [ $case == "disk" ]; then
378 ndevs=${#ost_names[@]}
382 # disable portals debug and get obdecho loaded on all relevant hosts
383 unique_hosts=(`unique ${host_names[@]}`)
386 for host in ${unique_hosts[@]}; do
387 host_vmstatf=${vmstatf}_${host}
388 echo -n > $host_vmstatf
389 remote_shell $host "vmstat 5 >> $host_vmstatf" &> /dev/null &
391 vmstatpids[$pidcount]=$pid
392 pidcount=$((pidcount+1))
394 # get all the echo_client device numbers and names
395 for ((i=0; i < $ndevs; i++)); do
396 host=${host_names[$i]}
397 devno=(`get_ec_devno $host "${client_names[$i]}" "${ost_names[$i]}"`)
398 if ((${#devno[@]} != 3)); then
401 devnos[$i]=${devno[0]}
402 client_names[$i]=${devno[1]}
403 do_teardown_ec[$i]=${devno[2]}
405 if (($ndevs <= 0 || ${#host_names[@]} <= 0)); then
406 echo "no devices or hosts specified"
407 cleanup 0 $clean_srv_OSS $cleanup_oscs
409 # Buffers will be spread out among all hosts, so allow for that
410 max_buffer_mem=$(( ${max_buffer_mem} * ${#unique_hosts[@]} ))
411 print_summary "$(date) Obdfilter-survey for case=$case from $(hostname)"
412 for ((rsz = $rszlo; rsz <= $rszhi; rsz*=2)); do
413 for ((nobj = $nobjlo; nobj <= $nobjhi; nobj*=2)); do
414 for ((thr = $thrlo; thr <= $thrhi; thr*=2)); do
415 if ((thr % nobj)); then
419 if [ -n "$restart_rsz" -a\
420 -n "$restart_nobj" -a\
421 -n "$restart_thr" ]; then
422 if ((rsz < restart_rsz ||\
423 (rsz == restart_rsz &&\
424 (nobj < restart_nobj ||\
425 (nobj == restart_nobj &&\
426 thr < restart_thr))))); then
431 total_thr=$((ndevs*thr))
432 total_nobj=$((ndevs*nobj))
433 pages=$((rsz/PAGE_SIZE))
434 actual_rsz=$((pages*PAGE_SIZE))
435 count=$((size*1024/(actual_rsz*thr)))
436 actual_size=$((actual_rsz*count*thr))
437 total_size=$((actual_size*ndevs))
438 # show computed parameters
439 str=`printf 'ost %2d sz %8dK rsz %4dK obj %4d thr %4d ' \
440 $ndevs $total_size $actual_rsz $total_nobj $total_thr`
441 echo "=======================> $str" >> $workf
442 print_summary -n "$str"
443 if ((total_thr * actual_rsz > max_buffer_mem)); then
444 print_summary "Too much buffer space"
449 for ((idx = 0; idx < $ndevs; idx++)); do
450 host=${host_names[$idx]}
451 devno=${devnos[$idx]}
452 client_name="${host}:${client_names[$idx]}"
453 echo "=============> Create $nobj on $client_name" >> $workf
454 first_obj=`create_objects $host $devno $nobj $tmpf`
457 if [ $first_obj = "ERROR" ]; then
458 print_summary "created object #s on $client_name not contiguous"
461 first_objs[$idx]=$first_obj
464 for test in ${tests[@]}; do
466 for host in ${unique_hosts[@]}; do
467 echo "starting run for test: $test rsz: $rsz threads: $thr objects: $nobj" >> ${vmstatf}_${host}
469 print_summary -n "$test "
470 # create per-host script files
471 for host in ${unique_hosts[@]}; do
472 echo -n > ${cmdsf}_${host}
474 for ((idx = 0; idx < $ndevs; idx++)); do
475 host=${host_names[$idx]}
476 devno=${devnos[$idx]}
478 first_obj=${first_objs[$idx]}
479 thr_per_obj=$((${thr}/${nobj}))
480 echo >> ${cmdsf}_${host} \
481 "$lctl > $tmpfi 2>&1 \\
482 --threads $thr -$snap $devno \\
483 test_brw $count `testname2type $test` q $pages \\
484 ${thr_per_obj}t${first_obj} `testcase2mode $pages` &"
487 for host in ${unique_hosts[@]}; do
488 echo "wait" >> ${cmdsf}_${host}
489 pidarray[$pidcount]=0
490 pidcount=$((pidcount+1))
492 # timed run of all the per-host script files
495 for host in ${unique_hosts[@]}; do
496 remote_shell $host bash < ${cmdsf}_${host} &
497 pidarray[$pidcount]=$!
498 pidcount=$((pidcount+1))
501 for host in ${unique_hosts[@]}; do
502 wait ${pidarray[$pidcount]}
503 pidcount=$((pidcount+1))
507 # clean up per-host script files
508 for host in ${unique_hosts[@]}; do
512 # compute bandwidth from total data / elapsed time
513 str=`awk "BEGIN {printf \"%7.2f \",\
514 $total_size / (( $t1 - $t0 ) * 1024)}"`
515 print_summary -n "$str"
516 # collect/check individual OST stats
518 for ((idx = 0; idx < $ndevs; idx++)); do
519 client_name="${host_names[$idx]}:${client_names[$idx]}"
521 echo "=============> $test $client_name" >> $workf
522 host="${host_names[$idx]}"
523 remote_shell $host cat $tmpfi > ${tmpfi}_local
524 cat ${tmpfi}_local >> $workf
525 get_stats ${tmpfi}_local >> $tmpf
526 rm -f $tmpfi ${tmpfi}_local
528 # compute/display global min/max stats
529 echo "=============> $test global" >> $workf
531 stats=(`get_global_stats $tmpf`)
533 if ((stats[0] <= 0)); then
534 if ((stats[0] < 0)); then
535 str=`printf "%17s " ERROR`
537 str=`printf "%17s " SHORT`
540 str=`awk "BEGIN {printf \"[%7.2f,%7.2f] \",\
541 (${stats[1]} * $actual_rsz)/1024,\
542 (${stats[2]} * $actual_rsz)/1024; exit}"`
544 print_summary -n "$str"
547 # destroy objects we created
548 for ((idx = 0; idx < $ndevs; idx++)); do
549 host=${host_names[$idx]}
550 devno=${devnos[$idx]}
551 client_name="${host}:${client_names[$idx]}"
552 first_obj=${first_objs[$idx]}
553 echo "=============> Destroy $nobj on $client_name" >> $workf
554 destroy_objects $host $devno $first_obj $nobj $tmpf
561 cleanup 0 $clean_srv_OSS $cleanup_oscs