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 # instead of case=disk one can also run test as follows
26 # $ nobjhi=2 thrhi=2 size=1024 OSTS="lustre-OST0000 lustre-OST0001 ..." sh obdfilter-survey
28 # $ nobjhi=2 thrhi=2 size=1024 server_nid="<name/nid_of_server>" case=network sh obdfilter-survey
29 # where, server_nid 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 # instead of case="netdisk" one can also run test as follows
34 # $ nobjhi=2 thrhi=2 size=1024 ECHO_CLIENTS="ECHO_<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.]
41 # create a set of objects, check there are 'n' contiguous ones and
42 # return the first or 'ERROR'
43 # parameter: 1. hostname
45 # 3. number of object to be created (specified by user)
52 remote_shell $host $lctl --device $devno create $nobj > $rfile 2>&1
58 echo "$line" | grep -q 'is object id'
62 if [ $first -eq 0 ]; then
63 first=$(echo $line | awk '{print $6}')
64 first=$(printf "%d" $first)
68 obj=$(echo $line | awk '{print $6}')
69 obj=$(printf "%d" $obj)
70 diff=$((obj - (prev+1)))
71 if [ $diff -ne 0 ]; then
78 if [ $nobj -ne $count ]; then
79 echo "ERROR: $nobj != $count" >&2
82 elif [ $error -ne 0 ]; then
83 echo "ERROR: non contiguous objs found" >&2
89 # destroys all objects created in create_objects routine
90 # parameter: 3. start obj id.
97 remote_shell $host $lctl --device $devno destroy $obj0 $nobj > $rfile 2>&1
103 '/^Selected device [0-9]+$/ {n = 0; next}\
104 /error/ {n = -1; exit}\
105 /^[0-9]+\/[0-9]+ Total: [0-9]+\.[0-9]+\/second$/ {n++; v=strtonum($3); \
106 if (n == 1 || v < min) min = v;\
107 if (n == 1 || v > max) max = v;\
109 {if (n != 0) {n = -1; exit}}\
110 END {printf "%d %f %f\n", n, min, max}'
113 get_global_stats () {
115 awk < $rfile 'BEGIN {n = 0;}\
116 {n++; if (n == 1) {err = $1; min = $2; max = $3} else\
117 {if ($1 < err) err = $1;\
118 if ($2 < min) min = $2;\
119 if ($3 > max) max = $3}}\
120 END {if (n == 0) err = 0;\
121 printf "%d %f %f\n", err, min, max}'
124 # enable or disable data check.
125 # parameter: 1. read/write
127 # 'x' disables data check
134 *write*) echo "w$x";;
140 if [ "$1" = "-n" ]; then
145 echo $minusn "$*" >> $rsltf
149 # Customisation variables
150 #####################################################################
151 # One can change variable values in this section as per requirements
154 server_nid=${server_nid:-""}
156 if [ -n "$OSTS" ]; then
159 for name in $OSTS; do
160 ost_names[$count]=$name
164 ECHO_CLIENTS=${ECHO_CLIENTS:-""}
165 if [ -n "$ECHO_CLIENTS" ]; then
166 # make sure we unset ost_names so that our client_names get noticed...
168 declare -a client_names
170 for name in $ECHO_CLIENTS; do
171 client_names[$count]=$name
176 # result file prefix (date/time + hostname makes unique)
177 # NB ensure path to it exists
178 rslt_loc=${rslt_loc:-"/tmp"}
179 rslt=${rslt:-"$rslt_loc/obdfilter_survey_`date +%F@%R`_`uname -n`"}
181 # what tests to run (first must be write)
182 tests_str=${tests_str:-""}
183 if [ -n "$tests_str" ]; then
186 for name in $tests_str; do
191 #tests=(write rewrite read reread rewrite_again)
192 tests=(write rewrite read)
195 # Set this true to check file contents
198 # total size (MBytes) per obd instance
199 # large enough to avoid cache effects
200 # and to make test startup/shutdown overhead insignificant
203 # record size (KBytes)
207 # number of objects per OST
209 #was nobjhi=${nobjhi:-512}
212 # threads per OST (1024 max)
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 if [ ${#tests[@]} -eq 0 -o "${tests[0]}" != "write" ]; then
239 echo "tests: ${tests[@]}"
240 echo "First test must be 'write'" 1>&2
244 rsltf="${rslt}.summary"
245 workf="${rslt}.detail"
246 cmdsf="${rslt}.script"
247 vmstatf="${rslt}.vmstat"
251 # hide a little trick to unset this from the command line
252 if [ "$lustre_root" == " " ]; then
256 if [ -z "$lustre_root" ]; then
259 lctl=${lustre_root}/utils/lctl
262 if [ $case == "network" ]; then
263 if [ -z "$server_nid" ]; then
264 echo "Specify the server NID"
267 osc_names_string=`dsh $server_nid root "lctl dl"`
269 for name in $osc_names_str; do
273 if [ $count != 0 ]; then
274 echo "The existing setup must be cleaned";
277 # Now do the server setup
278 setup_srv_obd $server_nid "ost_testfs"
279 op_string=`dsh $server_nid root "lctl dl"`
283 for name in $op_string; do
284 if [ "$name" = "obdecho" ]; then
287 if [ "$name" = "ost" ]; then
292 if (( $obdecho == 0 || $ost == 0 )); then
293 echo "Server setup not done properly"
296 # Now start client setup
297 osc_names_str=$(lctl dl)
298 if [ -n "$osc_names_str" ]; then
299 echo "The existing setup must be cleaned";
302 ec_using_srv_nid $server_nid "osc_testfs" "test_obdfs"
303 declare -a client_names
304 client_names[0]="ECHO_osc_testfs"
307 if [ -z "$ECHO_CLIENTS" ]; then
308 if [ $case == "netdisk" ]; then
311 osc_names_str=$(lctl dl |grep osc | awk "{if (\$2 == \"UP\" && \$3 == \"osc\") {print \$4} }")
313 for name in $osc_names_str; do
314 osc_names[$count]=$name
317 osc_uuid_str=$(lctl dl |grep osc | awk "{if (\$2 == \"UP\" && \$3 == \"osc\") {print \$5} }")
319 for uuid in $osc_uuid_str; do
320 osc_uuids[$count]=$uuid
323 for (( i = 0 ; i < $count; i++ ))
325 ec_using_osc ${osc_names[$i]} ${osc_uuids[$i]}
327 ECHO_CLIENTS=$(lctl dl | grep echo_client | awk "{if (\$2 == \"UP\" && \$3 == \"echo_client\") {print \$4} }")
329 for name in $ECHO_CLIENTS; do
330 client_names[$cnt]=$name
336 if [ -z "$OSTS" ]; then
337 if [ $case == "disk" ]; then
342 # split out hostnames from client/ost names
343 ndevs=${#client_names[@]}
344 if ((ndevs != 0)); then
345 if ((${#ost_names[@]} != 0)); then
346 echo "Please specify client_names or ost_names, but not both" 1>&2
349 for ((i = 0; i < ndevs; i++)); do
350 str=(`split_hostname ${client_names[$i]}`)
351 host_names[$i]=${str[0]}
352 client_names[$i]=${str[1]}
355 ndevs=${#ost_names[@]}
356 if ((ndevs == 0)); then
357 echo "Please specify either client_names or ost_names" 1>&2
360 for ((i = 0; i < ndevs; i++)); do
361 str=(`split_hostname ${ost_names[$i]}`)
362 host_names[$i]=${str[0]}
363 ost_names[$i]=${str[1]}
368 # disable portals debug and get obdecho loaded on all relevant hosts
369 unique_hosts=(`unique ${host_names[@]}`)
371 for host in ${unique_hosts[@]}; do
372 host_vmstatf=${vmstatf}_${host}
373 echo -n > $host_vmstatf
374 remote_shell $host "vmstat 5 >> $host_vmstatf" &
376 vmstatpids[$pidcount]=$pid
377 pidcount=$((pidcount+1))
378 do_unload_obdecho[$host]=0
379 if obdecho_loaded $host; then
383 if obdecho_loaded $host; then
384 do_unload_obdecho[$host]=1
387 echo "Can't load obdecho on $host" 1>&2
391 # get all the echo_client device numbers and names
392 for ((i=0; i < ndevs; i++)); do
393 host=${host_names[$i]}
394 devno=(`get_ec_devno $host "${client_names[$i]}" "${ost_names[$i]}"`)
395 if ((${#devno[@]} != 3)); then
398 devnos[$i]=${devno[0]}
399 client_names[$i]=${devno[1]}
400 do_teardown_ec[$i]=${devno[2]}
402 for ((rsz = $rszlo; rsz <= $rszhi; rsz*=2)); do
403 for ((nobj = $nobjlo; nobj <= $nobjhi; nobj*=2)); do
404 for ((thr = $thrlo; thr <= $thrhi; thr*=2)); do
405 if ((thr % nobj)); then
409 if [ -n "$restart_rsz" -a\
410 -n "$restart_nobj" -a\
411 -n "$restart_thr" ]; then
412 if ((rsz < restart_rsz ||\
413 (rsz == restart_rsz &&\
414 (nobj < restart_nobj ||\
415 (nobj == restart_nobj &&\
416 thr < restart_thr))))); then
421 total_thr=$((ndevs*thr))
422 total_nobj=$((ndevs*nobj))
423 pages=$((rsz/PAGE_SIZE))
424 actual_rsz=$((pages*PAGE_SIZE))
425 count=$((size*1024/(actual_rsz*thr)))
426 actual_size=$((actual_rsz*count*thr))
427 total_size=$((actual_size*ndevs))
428 # show computed parameters
429 str=`printf 'ost %2d sz %8dK rsz %4d obj %4d thr %4d ' \
430 $ndevs $total_size $actual_rsz $total_nobj $total_thr`
431 echo "=======================> $str" >> $workf
432 print_summary -n "$str"
433 if ((total_thr * actual_rsz > max_buffer_mem)); then
434 print_summary "Too much buffer space"
439 for ((idx = 0; idx < ndevs; idx++)); do
440 host=${host_names[$idx]}
441 devno=${devnos[$idx]}
442 client_name="${host}:${client_names[$idx]}"
443 echo "=============> Create $nobj on $client_name" >> $workf
444 first_obj=`create_objects $host $devno $nobj $tmpf`
447 if [ $first_obj = "ERROR" ]; then
448 print_summary "created object #s on $client_name not contiguous"
451 first_objs[$idx]=$first_obj
454 for test in ${tests[@]}; do
456 for host in ${unique_hosts[@]}; do
457 echo "starting run for test: $test rsz: $rsz threads: $thr objects: $nobj" >> ${vmstatf}_${host}
459 print_summary -n "$test "
460 # create per-host script files
461 for host in ${unique_hosts[@]}; do
462 echo -n > ${cmdsf}_${host}
464 for ((idx = 0; idx < ndevs; idx++)); do
465 host=${host_names[$idx]}
466 devno=${devnos[$idx]}
468 first_obj=${first_objs[$idx]}
469 thr_per_obj=$((${thr}/${nobj}))
470 echo >> ${cmdsf}_${host} \
471 "$lctl > $tmpfi 2>&1 \\
472 --threads $thr -$snap $devno \\
473 test_brw $count `testname2type $test` q $pages ${thr_per_obj}t${first_obj} &"
476 for host in ${unique_hosts[@]}; do
477 echo "wait" >> ${cmdsf}_${host}
478 pidarray[$pidcount]=0
479 pidcount=$((pidcount+1))
481 # timed run of all the per-host script files
484 for host in ${unique_hosts[@]}; do
485 remote_shell $host bash < ${cmdsf}_${host} &
486 pidarray[$pidcount]=$!
487 pidcount=$((pidcount+1))
490 for host in ${unique_hosts[@]}; do
491 wait ${pidarray[$pidcount]}
492 pidcount=$((pidcount+1))
496 # clean up per-host script files
497 for host in ${unique_hosts[@]}; do
500 # compute bandwidth from total data / elapsed time
501 str=`awk "BEGIN {printf \"%7.2f \",\
502 $total_size / (( $t1 - $t0 ) * 1024)}"`
503 print_summary -n "$str"
504 # collect/check individual OST stats
506 for ((idx = 0; idx < ndevs; idx++)); do
507 client_name="${host_names[$idx]}:${client_names[$idx]}"
509 echo "=============> $test $client_name" >> $workf
510 host="${host_names[$idx]}"
511 remote_shell $host cat $tmpfi >> $workf
512 get_stats $tmpfi >> $tmpf
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