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}\
139 /^[0-9]+\/[0-9]+ Total: [0-9]+\.[0-9]+\/second$/ {n++; v=strtonum($3); \
140 if (n == 1 || v < min) min = v;\
141 if (n == 1 || v > max) max = v;\
143 {if (n != 0) {n = -1; exit}}\
144 END {printf "%d %f %f\n", n, min, max}'
147 get_global_stats () {
149 awk < $rfile 'BEGIN {n = 0;}\
150 {n++; if (n == 1) {err = $1; min = $2; max = $3} else\
151 {if ($1 < err) err = $1;\
152 if ($2 < min) min = $2;\
153 if ($3 > max) max = $3}}\
154 END {if (n == 0) err = 0;\
155 printf "%d %f %f\n", err, min, max}'
158 # enable or disable data check.
159 # parameter: 1. read/write
161 # 'x' disables data check
168 *write*) echo "w$x";;
174 if [ "$1" = "-n" ]; then
179 echo $minusn "$*" >> $rsltf
183 # Customisation variables
184 #####################################################################
185 # One can change variable values in this section as per requirements
187 targets=${targets:-""}
189 if [ -n "$targets" ]; then
191 declare -a client_names
193 for name in $targets; do
194 if [ $case == "disk" ]; then
195 ost_names[$count]=$name
197 client_names[$count]=$name
203 # what tests to run (first must be write)
204 tests_str=${tests_str:-""}
205 if [ -n "$tests_str" ]; then
208 for name in $tests_str; do
213 #tests=(write rewrite read reread rewrite_again)
214 tests=(write rewrite read)
217 # restart from here iff all are defined
222 # machine's page size (K)
223 if [ -z "$PAGE_SIZE" ]; then
224 if which python >/dev/null; then
225 PAGE_SIZE=`echo 'import resource; print resource.getpagesize()/1024;' |python`
228 PAGE_SIZE=${PAGE_SIZE:-4}
230 # max buffer_mem (total_threads * buffer size)
231 # (to avoid lctl ENOMEM problems)
232 max_buffer_mem=$((1024 * 1024))
235 # Customisation variables ends here.
236 #####################################################################
237 # leave the rest of this alone unless you know what you're doing...
239 # check and insert obdecho module
240 if ! lsmod | grep obdecho > /dev/null; then
243 if [ ${#tests[@]} -eq 0 -o "${tests[0]}" != "write" ]; then
244 echo "tests: ${tests[@]}"
245 echo "First test must be 'write'" 1>&2
249 rsltf="${rslt}.summary"
250 workf="${rslt}.detail"
251 cmdsf="${rslt}.script"
252 vmstatf="${rslt}.vmstat"
256 # hide a little trick to unset this from the command line
257 if [ "$lustre_root" == " " ]; then
261 if [ -z "$lustre_root" ]; then
264 lctl=${lustre_root}/utils/lctl
267 # split out hostnames from client/ost names
269 for trgt in $targets; do
270 str=(`split_hostname $trgt`)
271 host_names[$ndevs]=${str[0]}
272 client_names[$ndevs]=${str[1]}
275 if [ $case == "disk" ]; then
276 if [ $rszhi -gt 1024 ]; then
277 echo "Test disk case support maximum 1024KB IO data" \
278 "(rszhi=$rszhi is too big) please use a smaller value."
281 for ((i = 0; i < $ndevs; i++)); do
282 ost_names[$i]=${client_names[$i]}
285 if [ $case == "netdisk" ]; then
286 if [ "$targets" ]; then
287 for ((i = 0; i < $ndevs; i++)); do
288 setup_osc_for_remote_ost ${host_names[$i]} \
289 ${client_names[$i]} $i
290 osc_name=${client_names[$i]}_osc
291 ec_using_osc $osc_name
292 cleanup_oscs="$cleanup_oscs $osc_name"
295 client_names_str=$($lctl dl | grep -v mdt | \
296 awk '{if ($2 == "UP" && $3 == "osc") {print $4} }')
298 for name in $client_names_str; do
299 client_names[$count]=`echo $name | sed 's/-osc-.*$//'`
303 host_names_str=$($lctl dl -t | grep -v mdt | \
304 awk '{if ($2 == "UP" && $3 == "osc") {print $7} }')
306 for name in $host_names_str; do
307 host_names[$count]=`echo $name | sed 's/@.*$//'`
311 for (( i = 0; i < $count; i++ )) do
312 setup_osc_for_remote_ost ${host_names[$i]} \
313 ${client_names[$i]} $i
314 osc_name=${client_names[$i]}_osc
315 ec_using_osc $osc_name
316 cleanup_oscs="$cleanup_oscs $osc_name"
320 echo_clients=$($lctl dl | grep echo_client | awk "{if (\$2 == \"UP\" && \$3 == \"echo_client\") {print \$4} }")
322 for name in $echo_clients; do
323 client_names[$cnt]=$name
324 host_names[$cnt]=localhost
327 ndevs=${#client_names[@]}
329 if [ $case == "network" ]; then
331 if [ -z "$server_nid" ]; then
332 echo "Specify hostname or ip-address of server"
335 # check for obdecho module on server
336 if ! dsh $server_nid root "lsmod | grep obdecho > /dev/null"; then
337 dsh $server_nid root "modprobe obdecho"
339 # Now do the server setup
340 setup_srv_obd $server_nid "echo_srv"
341 oss_on_srv=`dsh $server_nid root "$lctl dl | grep OSS" | awk '{ print $4 }'`
342 if [ -z $oss_on_srv ]; then
343 setup_OSS $server_nid
346 if ! dsh $server_nid root "$lctl dl | grep obdecho > /dev/null 2>&1"; then
347 echo "obdecho not setup on server"
350 if ! dsh $server_nid root "$lctl dl | grep ost > /dev/null 2>&1"; then
351 echo "ost not setup on server"
354 # Now start client setup
355 osc_names_str=$($lctl dl| grep osc | grep -v mdt | grep UP)
356 if [ -n "$osc_names_str" ]; then
357 echo "The existing setup must be cleaned";
360 ec_using_srv_nid $server_nid "echotmp" "echotmp_UUID"
361 client_names[0]="echotmp_ecc"
363 if [ -z "$targets" ]; then
364 if [ $case == "disk" ]; then
366 ndevs=${#ost_names[@]}
370 # disable portals debug and get obdecho loaded on all relevant hosts
371 unique_hosts=(`unique ${host_names[@]}`)
374 for host in ${unique_hosts[@]}; do
375 host_vmstatf=${vmstatf}_${host}
376 echo -n > $host_vmstatf
377 remote_shell $host "vmstat 5 >> $host_vmstatf" &> /dev/null &
379 vmstatpids[$pidcount]=$pid
380 pidcount=$((pidcount+1))
382 # get all the echo_client device numbers and names
383 for ((i=0; i < $ndevs; i++)); do
384 host=${host_names[$i]}
385 devno=(`get_ec_devno $host "${client_names[$i]}" "${ost_names[$i]}"`)
386 if ((${#devno[@]} != 3)); then
389 devnos[$i]=${devno[0]}
390 client_names[$i]=${devno[1]}
391 do_teardown_ec[$i]=${devno[2]}
393 if (($ndevs <= 0 || ${#host_names[@]} <= 0)); then
394 echo "no devices or hosts specified"
395 cleanup 0 $clean_srv_OSS $cleanup_oscs
397 # Buffers will be spread out among all hosts, so allow for that
398 max_buffer_mem=$(( ${max_buffer_mem} * ${#unique_hosts[@]} ))
399 print_summary "$(date) Obdfilter-survey for case=$case from $(hostname)"
400 for ((rsz = $rszlo; rsz <= $rszhi; rsz*=2)); do
401 for ((nobj = $nobjlo; nobj <= $nobjhi; nobj*=2)); do
402 for ((thr = $thrlo; thr <= $thrhi; thr*=2)); do
403 if ((thr % nobj)); then
407 if [ -n "$restart_rsz" -a\
408 -n "$restart_nobj" -a\
409 -n "$restart_thr" ]; then
410 if ((rsz < restart_rsz ||\
411 (rsz == restart_rsz &&\
412 (nobj < restart_nobj ||\
413 (nobj == restart_nobj &&\
414 thr < restart_thr))))); then
419 total_thr=$((ndevs*thr))
420 total_nobj=$((ndevs*nobj))
421 pages=$((rsz/PAGE_SIZE))
422 actual_rsz=$((pages*PAGE_SIZE))
423 count=$((size*1024/(actual_rsz*thr)))
424 actual_size=$((actual_rsz*count*thr))
425 total_size=$((actual_size*ndevs))
426 # show computed parameters
427 str=`printf 'ost %2d sz %8dK rsz %4dK obj %4d thr %4d ' \
428 $ndevs $total_size $actual_rsz $total_nobj $total_thr`
429 echo "=======================> $str" >> $workf
430 print_summary -n "$str"
431 if ((total_thr * actual_rsz > max_buffer_mem)); then
432 print_summary "Too much buffer space"
437 for ((idx = 0; idx < $ndevs; idx++)); do
438 host=${host_names[$idx]}
439 devno=${devnos[$idx]}
440 client_name="${host}:${client_names[$idx]}"
441 echo "=============> Create $nobj on $client_name" >> $workf
442 first_obj=`create_objects $host $devno $nobj $tmpf`
445 if [ $first_obj = "ERROR" ]; then
446 print_summary "created object #s on $client_name not contiguous"
449 first_objs[$idx]=$first_obj
452 for test in ${tests[@]}; do
454 for host in ${unique_hosts[@]}; do
455 echo "starting run for test: $test rsz: $rsz threads: $thr objects: $nobj" >> ${vmstatf}_${host}
457 print_summary -n "$test "
458 # create per-host script files
459 for host in ${unique_hosts[@]}; do
460 echo -n > ${cmdsf}_${host}
462 for ((idx = 0; idx < $ndevs; idx++)); do
463 host=${host_names[$idx]}
464 devno=${devnos[$idx]}
466 first_obj=${first_objs[$idx]}
467 thr_per_obj=$((${thr}/${nobj}))
468 echo >> ${cmdsf}_${host} \
469 "$lctl > $tmpfi 2>&1 \\
470 --threads $thr -$snap $devno \\
471 test_brw $count `testname2type $test` q $pages ${thr_per_obj}t${first_obj} &"
474 for host in ${unique_hosts[@]}; do
475 echo "wait" >> ${cmdsf}_${host}
476 pidarray[$pidcount]=0
477 pidcount=$((pidcount+1))
479 # timed run of all the per-host script files
482 for host in ${unique_hosts[@]}; do
483 remote_shell $host bash < ${cmdsf}_${host} &
484 pidarray[$pidcount]=$!
485 pidcount=$((pidcount+1))
488 for host in ${unique_hosts[@]}; do
489 wait ${pidarray[$pidcount]}
490 pidcount=$((pidcount+1))
494 # clean up per-host script files
495 for host in ${unique_hosts[@]}; do
499 # compute bandwidth from total data / elapsed time
500 str=`awk "BEGIN {printf \"%7.2f \",\
501 $total_size / (( $t1 - $t0 ) * 1024)}"`
502 print_summary -n "$str"
503 # collect/check individual OST stats
505 for ((idx = 0; idx < $ndevs; idx++)); do
506 client_name="${host_names[$idx]}:${client_names[$idx]}"
508 echo "=============> $test $client_name" >> $workf
509 host="${host_names[$idx]}"
510 remote_shell $host cat $tmpfi > ${tmpfi}_local
511 cat ${tmpfi}_local >> $workf
512 get_stats ${tmpfi}_local >> $tmpf
513 rm -f $tmpfi ${tmpfi}_local
515 # compute/display global min/max stats
516 echo "=============> $test global" >> $workf
518 stats=(`get_global_stats $tmpf`)
520 if ((stats[0] <= 0)); then
521 if ((stats[0] < 0)); then
522 str=`printf "%17s " ERROR`
524 str=`printf "%17s " SHORT`
527 str=`awk "BEGIN {printf \"[%7.2f,%7.2f] \",\
528 (${stats[1]} * $actual_rsz)/1024,\
529 (${stats[2]} * $actual_rsz)/1024; exit}"`
531 print_summary -n "$str"
534 # destroy objects we created
535 for ((idx = 0; idx < $ndevs; idx++)); do
536 host=${host_names[$idx]}
537 devno=${devnos[$idx]}
538 client_name="${host}:${client_names[$idx]}"
539 first_obj=${first_objs[$idx]}
540 echo "=============> Destroy $nobj on $client_name" >> $workf
541 destroy_objects $host $devno $first_obj $nobj $tmpf
548 cleanup 0 $clean_srv_OSS $cleanup_oscs