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
51 # total size (MBytes) per obd instance
52 # large enough to avoid cache effects
53 # and to make test startup/shutdown overhead insignificant
56 # record size (KBytes) ( 7168 max)
60 # number of objects per OST
62 #was nobjhi=${nobjhi:-512}
65 # threads per OST (1024 max)
73 # create a set of objects, check there are 'n' contiguous ones and
74 # echo out the first or 'ERROR'
75 # parameter: 1. hostname
77 # 3. number of object to be created (specified by user)
84 remote_shell $host $lctl --device $devno create $nobj > $rfile 2>&1
90 # Count number of objects (lines containing " is object id "), and
91 # ensure that the objects numbers are sequential.
94 while read -u3 line; do
96 ( *' is object id '* )
98 if test $(( count += 1 )) -gt 1 ; then
99 (( $6 != prev + 1 )) && error=1
109 if [ $nobj -ne $count ]; then
110 echo "ERROR: $nobj != $count" >&2
113 elif [ $error -ne 0 ]; then
114 echo "ERROR: non contiguous objs found" >&2
122 # destroys all objects created in create_objects routine
123 # parameter: 3. start obj id.
130 remote_shell $host $lctl --device $devno destroy $obj0 $nobj > $rfile 2>&1
136 '/^Selected device [0-9]+$/ {n = 0; next}\
137 /error/ {n = -1; exit}\
138 /^[0-9]+\/[0-9]+ Total: [0-9]+\.[0-9]+\/second$/ {n++; v=strtonum($3); \
139 if (n == 1 || v < min) min = v;\
140 if (n == 1 || v > max) max = v;\
142 {if (n != 0) {n = -1; exit}}\
143 END {printf "%d %f %f\n", n, min, max}'
146 get_global_stats () {
148 awk < $rfile 'BEGIN {n = 0;}\
149 {n++; if (n == 1) {err = $1; min = $2; max = $3} else\
150 {if ($1 < err) err = $1;\
151 if ($2 < min) min = $2;\
152 if ($3 > max) max = $3}}\
153 END {if (n == 0) err = 0;\
154 printf "%d %f %f\n", err, min, max}'
157 # enable or disable data check.
158 # parameter: 1. read/write
160 # 'x' disables data check
167 *write*) echo "w$x";;
173 if [ "$1" = "-n" ]; then
178 echo $minusn "$*" >> $rsltf
182 # Customisation variables
183 #####################################################################
184 # One can change variable values in this section as per requirements
186 targets=${targets:-""}
188 if [ -n "$targets" ]; then
190 declare -a client_names
192 for name in $targets; do
193 if [ $case == "disk" ]; then
194 ost_names[$count]=$name
196 client_names[$count]=$name
202 # what tests to run (first must be write)
203 tests_str=${tests_str:-""}
204 if [ -n "$tests_str" ]; then
207 for name in $tests_str; do
212 #tests=(write rewrite read reread rewrite_again)
213 tests=(write rewrite read)
216 # restart from here iff all are defined
221 # machine's page size (K)
222 if [ -z "$PAGE_SIZE" ]; then
223 if which python >/dev/null; then
224 PAGE_SIZE=`echo 'import resource; print resource.getpagesize()/1024;' |python`
227 PAGE_SIZE=${PAGE_SIZE:-4}
229 # max buffer_mem (total_threads * buffer size)
230 # (to avoid lctl ENOMEM problems)
231 max_buffer_mem=$((1024 * 1024))
234 # Customisation variables ends here.
235 #####################################################################
236 # leave the rest of this alone unless you know what you're doing...
238 # check and insert obdecho module
239 if ! lsmod | grep obdecho > /dev/null; then
242 if [ ${#tests[@]} -eq 0 -o "${tests[0]}" != "write" ]; then
243 echo "tests: ${tests[@]}"
244 echo "First test must be 'write'" 1>&2
248 rsltf="${rslt}.summary"
249 workf="${rslt}.detail"
250 cmdsf="${rslt}.script"
251 vmstatf="${rslt}.vmstat"
255 # hide a little trick to unset this from the command line
256 if [ "$lustre_root" == " " ]; then
260 if [ -z "$lustre_root" ]; then
263 lctl=${lustre_root}/utils/lctl
266 # split out hostnames from client/ost names
268 for trgt in $targets; do
269 str=(`split_hostname $trgt`)
270 host_names[$ndevs]=${str[0]}
271 client_names[$ndevs]=${str[1]}
274 if [ $case == "disk" ]; then
275 if [ $rszhi -gt 1024 ]; then
276 echo "Test disk case support maximum 1024KB IO data" \
277 "(rszhi=$rszhi is too big) please use a smaller value."
280 for ((i = 0; i < $ndevs; i++)); do
281 ost_names[$i]=${client_names[$i]}
284 if [ $case == "netdisk" ]; then
285 if [ "$targets" ]; then
286 for ((i = 0; i < $ndevs; i++)); do
287 setup_osc_for_remote_ost ${host_names[$i]} \
288 ${client_names[$i]} $i
289 osc_name=${client_names[$i]}_osc
290 ec_using_osc $osc_name
291 cleanup_oscs="$cleanup_oscs $osc_name"
294 client_names_str=$($lctl dl | grep -v mdt | \
295 awk '{if ($2 == "UP" && $3 == "osc") {print $4} }')
297 for name in $client_names_str; do
298 client_names[$count]=`echo $name | sed 's/-osc-.*$//'`
302 host_names_str=$($lctl dl -t | grep -v mdt | \
303 awk '{if ($2 == "UP" && $3 == "osc") {print $7} }')
305 for name in $host_names_str; do
306 host_names[$count]=`echo $name | sed 's/@.*$//'`
310 for (( i = 0; i < $count; i++ )) do
311 setup_osc_for_remote_ost ${host_names[$i]} \
312 ${client_names[$i]} $i
313 osc_name=${client_names[$i]}_osc
314 ec_using_osc $osc_name
315 cleanup_oscs="$cleanup_oscs $osc_name"
319 echo_clients=$($lctl dl | grep echo_client | awk "{if (\$2 == \"UP\" && \$3 == \"echo_client\") {print \$4} }")
321 for name in $echo_clients; do
322 client_names[$cnt]=$name
323 host_names[$cnt]=localhost
326 ndevs=${#client_names[@]}
328 if [ $case == "network" ]; then
330 if [ -z "$server_nid" ]; then
331 echo "Specify hostname or ip-address of server"
334 # check for obdecho module on server
335 if ! dsh $server_nid root "lsmod | grep obdecho > /dev/null"; then
336 dsh $server_nid root "modprobe obdecho"
338 # Now do the server setup
339 setup_srv_obd $server_nid "echo_srv"
340 oss_on_srv=`dsh $server_nid root "$lctl dl | grep OSS" | awk '{ print $4 }'`
341 if [ -z $oss_on_srv ]; then
342 setup_OSS $server_nid
345 if ! dsh $server_nid root "$lctl dl | grep obdecho > /dev/null 2>&1"; then
346 echo "obdecho not setup on server"
349 if ! dsh $server_nid root "$lctl dl | grep ost > /dev/null 2>&1"; then
350 echo "ost not setup on server"
353 # Now start client setup
354 osc_names_str=$($lctl dl| grep osc | grep -v mdt | grep UP)
355 if [ -n "$osc_names_str" ]; then
356 echo "The existing setup must be cleaned";
359 ec_using_srv_nid $server_nid "echotmp" "echotmp_UUID"
360 client_names[0]="echotmp_ecc"
362 if [ -z "$targets" ]; then
363 if [ $case == "disk" ]; then
365 ndevs=${#ost_names[@]}
369 # disable portals debug and get obdecho loaded on all relevant hosts
370 unique_hosts=(`unique ${host_names[@]}`)
373 for host in ${unique_hosts[@]}; do
374 host_vmstatf=${vmstatf}_${host}
375 echo -n > $host_vmstatf
376 remote_shell $host "vmstat 5 >> $host_vmstatf" &> /dev/null &
378 vmstatpids[$pidcount]=$pid
379 pidcount=$((pidcount+1))
381 # get all the echo_client device numbers and names
382 for ((i=0; i < $ndevs; i++)); do
383 host=${host_names[$i]}
384 devno=(`get_ec_devno $host "${client_names[$i]}" "${ost_names[$i]}"`)
385 if ((${#devno[@]} != 3)); then
388 devnos[$i]=${devno[0]}
389 client_names[$i]=${devno[1]}
390 do_teardown_ec[$i]=${devno[2]}
392 if (($ndevs <= 0 || ${#host_names[@]} <= 0)); then
393 echo "no devices or hosts specified"
394 cleanup 0 $clean_srv_OSS $cleanup_oscs
396 # Buffers will be spread out among all hosts, so allow for that
397 max_buffer_mem=$(( ${max_buffer_mem} * ${#unique_hosts[@]} ))
398 print_summary "$(date) Obdfilter-survey for case=$case from $(hostname)"
399 for ((rsz = $rszlo; rsz <= $rszhi; rsz*=2)); do
400 for ((nobj = $nobjlo; nobj <= $nobjhi; nobj*=2)); do
401 for ((thr = $thrlo; thr <= $thrhi; thr*=2)); do
402 if ((thr % nobj)); then
406 if [ -n "$restart_rsz" -a\
407 -n "$restart_nobj" -a\
408 -n "$restart_thr" ]; then
409 if ((rsz < restart_rsz ||\
410 (rsz == restart_rsz &&\
411 (nobj < restart_nobj ||\
412 (nobj == restart_nobj &&\
413 thr < restart_thr))))); then
418 total_thr=$((ndevs*thr))
419 total_nobj=$((ndevs*nobj))
420 pages=$((rsz/PAGE_SIZE))
421 actual_rsz=$((pages*PAGE_SIZE))
422 count=$((size*1024/(actual_rsz*thr)))
423 actual_size=$((actual_rsz*count*thr))
424 total_size=$((actual_size*ndevs))
425 # show computed parameters
426 str=`printf 'ost %2d sz %8dK rsz %4dK obj %4d thr %4d ' \
427 $ndevs $total_size $actual_rsz $total_nobj $total_thr`
428 echo "=======================> $str" >> $workf
429 print_summary -n "$str"
430 if ((total_thr * actual_rsz > max_buffer_mem)); then
431 print_summary "Too much buffer space"
436 for ((idx = 0; idx < $ndevs; idx++)); do
437 host=${host_names[$idx]}
438 devno=${devnos[$idx]}
439 client_name="${host}:${client_names[$idx]}"
440 echo "=============> Create $nobj on $client_name" >> $workf
441 first_obj=`create_objects $host $devno $nobj $tmpf`
444 if [ $first_obj = "ERROR" ]; then
445 print_summary "created object #s on $client_name not contiguous"
448 first_objs[$idx]=$first_obj
451 for test in ${tests[@]}; do
453 for host in ${unique_hosts[@]}; do
454 echo "starting run for test: $test rsz: $rsz threads: $thr objects: $nobj" >> ${vmstatf}_${host}
456 print_summary -n "$test "
457 # create per-host script files
458 for host in ${unique_hosts[@]}; do
459 echo -n > ${cmdsf}_${host}
461 for ((idx = 0; idx < $ndevs; idx++)); do
462 host=${host_names[$idx]}
463 devno=${devnos[$idx]}
465 first_obj=${first_objs[$idx]}
466 thr_per_obj=$((${thr}/${nobj}))
467 echo >> ${cmdsf}_${host} \
468 "$lctl > $tmpfi 2>&1 \\
469 --threads $thr -$snap $devno \\
470 test_brw $count `testname2type $test` q $pages ${thr_per_obj}t${first_obj} &"
473 for host in ${unique_hosts[@]}; do
474 echo "wait" >> ${cmdsf}_${host}
475 pidarray[$pidcount]=0
476 pidcount=$((pidcount+1))
478 # timed run of all the per-host script files
481 for host in ${unique_hosts[@]}; do
482 remote_shell $host bash < ${cmdsf}_${host} &
483 pidarray[$pidcount]=$!
484 pidcount=$((pidcount+1))
487 for host in ${unique_hosts[@]}; do
488 wait ${pidarray[$pidcount]}
489 pidcount=$((pidcount+1))
493 # clean up per-host script files
494 for host in ${unique_hosts[@]}; do
498 # compute bandwidth from total data / elapsed time
499 str=`awk "BEGIN {printf \"%7.2f \",\
500 $total_size / (( $t1 - $t0 ) * 1024)}"`
501 print_summary -n "$str"
502 # collect/check individual OST stats
504 for ((idx = 0; idx < $ndevs; idx++)); do
505 client_name="${host_names[$idx]}:${client_names[$idx]}"
507 echo "=============> $test $client_name" >> $workf
508 host="${host_names[$idx]}"
509 remote_shell $host cat $tmpfi > ${tmpfi}_local
510 cat ${tmpfi}_local >> $workf
511 get_stats ${tmpfi}_local >> $tmpf
512 rm -f $tmpfi ${tmpfi}_local
514 # compute/display global min/max stats
515 echo "=============> $test global" >> $workf
517 stats=(`get_global_stats $tmpf`)
519 if ((stats[0] <= 0)); then
520 if ((stats[0] < 0)); then
521 str=`printf "%17s " ERROR`
523 str=`printf "%17s " SHORT`
526 str=`awk "BEGIN {printf \"[%7.2f,%7.2f] \",\
527 (${stats[1]} * $actual_rsz)/1024,\
528 (${stats[2]} * $actual_rsz)/1024; exit}"`
530 print_summary -n "$str"
533 # destroy objects we created
534 for ((idx = 0; idx < $ndevs; idx++)); do
535 host=${host_names[$idx]}
536 devno=${devnos[$idx]}
537 client_name="${host}:${client_names[$idx]}"
538 first_obj=${first_objs[$idx]}
539 echo "=============> Destroy $nobj on $client_name" >> $workf
540 destroy_objects $host $devno $first_obj $nobj $tmpf
547 cleanup 0 $clean_srv_OSS $cleanup_oscs