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]=$name
176 # result file prefix (date/time + hostname makes unique)
177 # NB ensure path to it exists
178 rslt=${rslt:-"/tmp/obdfilter_survey_`date +%F@%R`_`uname -n`"}
180 # what tests to run (first must be write)
181 tests_str=${tests_str:-""}
182 if [ -n "$tests_str" ]; then
185 for name in $tests_str; do
190 #tests=(write rewrite read reread rewrite_again)
191 tests=(write rewrite read)
194 # Set this true to check file contents
197 # total size (MBytes) per obd instance
198 # large enough to avoid cache effects
199 # and to make test startup/shutdown overhead insignificant
202 # record size (KBytes)
206 # number of objects per OST
208 #was nobjhi=${nobjhi:-512}
211 # threads per OST (1024 max)
215 # restart from here iff all are defined
220 # machine's page size (K)
221 if [ -z "$PAGE_SIZE" ]; then
222 if which python >/dev/null; then
223 PAGE_SIZE=`echo 'import resource; print resource.getpagesize()/1024;' |python`
226 PAGE_SIZE=${PAGE_SIZE:-4}
228 # max buffer_mem (total_threads * buffer size)
229 # (to avoid lctl ENOMEM problems)
230 max_buffer_mem=$((1024 * 1024))
233 # Customisation variables ends here.
234 #####################################################################
235 # leave the rest of this alone unless you know what you're doing...
237 if [ ${#tests[@]} -eq 0 -o "${tests[0]}" != "write" ]; then
238 echo "tests: ${tests[@]}"
239 echo "First test must be 'write'" 1>&2
243 rsltf="${rslt}.summary"
244 workf="${rslt}.detail"
245 cmdsf="${rslt}.script"
246 vmstatf="${rslt}.vmstat"
250 # hide a little trick to unset this from the command line
251 if [ "$lustre_root" == " " ]; then
255 if [ -z "$lustre_root" ]; then
258 lctl=${lustre_root}/utils/lctl
261 if [ $case == "network" ]; then
262 if [ -z "$server_nid" ]; then
263 echo "Specify the server NID"
266 osc_names_string=`ssh root@"$server_nid" lctl dl`
268 for name in $osc_names_str; do
272 if [ $count != 0 ]; then
273 echo "The existing setup must be cleaned";
276 # Now do the server setup
277 setup_srv_obd $server_nid "ost_testfs"
278 op_string=`ssh root@"$server_nid" lctl dl`
282 for name in $op_string; do
283 if [ "$name" = "obdecho" ]; then
286 if [ "$name" = "ost" ]; then
291 if (( $obdecho == 0 || $ost == 0 )); then
292 echo "Server setup not done properly"
295 # Now start client setup
296 osc_names_str=$(lctl dl)
297 if [ -n "$osc_names_str" ]; then
298 echo "The existing setup must be cleaned";
301 ec_using_srv_nid $server_nid "osc_testfs" "test_obdfs"
302 declare -a client_names
303 client_names[0]="ECHO_osc_testfs"
306 if [ -z "$ECHO_CLIENTS" ]; then
307 if [ $case == "netdisk" ]; then
310 osc_names_str=$(lctl dl |grep osc | awk "{if (\$2 == \"UP\" && \$3 == \"osc\") {print \$4} }")
312 for name in $osc_names_str; do
313 osc_names[$count]=$name
316 osc_uuid_str=$(lctl dl |grep osc | awk "{if (\$2 == \"UP\" && \$3 == \"osc\") {print \$5} }")
318 for uuid in $osc_uuid_str; do
319 osc_uuids[$count]=$uuid
322 for (( i = 0 ; i < $count; i++ ))
324 ec_using_osc ${osc_names[$i]} ${osc_uuids[$i]}
326 ECHO_CLIENTS=$(lctl dl | grep echo_client | awk "{if (\$2 == \"UP\" && \$3 == \"echo_client\") {print \$4} }")
328 for name in $ECHO_CLIENTS; do
329 client_names[$cnt]=$name
335 if [ -z "$OSTS" ]; then
336 if [ $case == "disk" ]; then
341 # split out hostnames from client/ost names
342 ndevs=${#client_names[@]}
343 if ((ndevs != 0)); then
344 if ((${#ost_names[@]} != 0)); then
345 echo "Please specify client_names or ost_names, but not both" 1>&2
348 for ((i = 0; i < ndevs; i++)); do
349 str=(`split_hostname ${client_names[$i]}`)
350 host_names[$i]=${str[0]}
351 client_names[$i]=${str[1]}
354 ndevs=${#ost_names[@]}
355 if ((ndevs == 0)); then
356 echo "Please specify either client_names or ost_names" 1>&2
359 for ((i = 0; i < ndevs; i++)); do
360 str=(`split_hostname ${ost_names[$i]}`)
361 host_names[$i]=${str[0]}
362 ost_names[$i]=${str[1]}
367 # disable portals debug and get obdecho loaded on all relevant hosts
368 unique_hosts=(`unique ${host_names[@]}`)
370 for host in ${unique_hosts[@]}; do
371 host_vmstatf=${vmstatf}_${host}
372 echo -n > $host_vmstatf
373 remote_shell $host "vmstat 5 >> $host_vmstatf" &
375 vmstatpids[$pidcount]=$pid
376 pidcount=$((pidcount+1))
377 do_unload_obdecho[$host]=0
378 if obdecho_loaded $host; then
382 if obdecho_loaded $host; then
383 do_unload_obdecho[$host]=1
386 echo "Can't load obdecho on $host" 1>&2
390 # get all the echo_client device numbers and names
391 for ((i=0; i < ndevs; i++)); do
392 host=${host_names[$i]}
393 devno=(`get_ec_devno $host "${client_names[$i]}" "${ost_names[$i]}"`)
394 if ((${#devno[@]} != 3)); then
397 devnos[$i]=${devno[0]}
398 client_names[$i]=${devno[1]}
399 do_teardown_ec[$i]=${devno[2]}
401 for ((rsz = $rszlo; rsz <= $rszhi; rsz*=2)); do
402 for ((nobj = $nobjlo; nobj <= $nobjhi; nobj*=2)); do
403 for ((thr = $thrlo; thr <= $thrhi; thr*=2)); do
404 if ((thr % nobj)); then
408 if [ -n "$restart_rsz" -a\
409 -n "$restart_nobj" -a\
410 -n "$restart_thr" ]; then
411 if ((rsz < restart_rsz ||\
412 (rsz == restart_rsz &&\
413 (nobj < restart_nobj ||\
414 (nobj == restart_nobj &&\
415 thr < restart_thr))))); then
420 total_thr=$((ndevs*thr))
421 total_nobj=$((ndevs*nobj))
422 pages=$((rsz/PAGE_SIZE))
423 actual_rsz=$((pages*PAGE_SIZE))
424 count=$((size*1024/(actual_rsz*thr)))
425 actual_size=$((actual_rsz*count*thr))
426 total_size=$((actual_size*ndevs))
427 # show computed parameters
428 str=`printf 'ost %2d sz %8dK rsz %4d obj %4d thr %4d ' \
429 $ndevs $total_size $actual_rsz $total_nobj $total_thr`
430 echo "=======================> $str" >> $workf
431 print_summary -n "$str"
432 if ((total_thr * actual_rsz > max_buffer_mem)); then
433 print_summary "Too much buffer space"
438 for ((idx = 0; idx < ndevs; idx++)); do
439 host=${host_names[$idx]}
440 devno=${devnos[$idx]}
441 client_name="${host}:${client_names[$idx]}"
442 echo "=============> Create $nobj on $client_name" >> $workf
443 first_obj=`create_objects $host $devno $nobj $tmpf`
446 if [ $first_obj = "ERROR" ]; then
447 print_summary "created object #s on $client_name not contiguous"
450 first_objs[$idx]=$first_obj
453 for test in ${tests[@]}; do
455 for host in ${unique_hosts[@]}; do
456 echo "starting run for test: $test rsz: $rsz threads: $thr objects: $nobj" >> ${vmstatf}_${host}
458 print_summary -n "$test "
459 # create per-host script files
460 for host in ${unique_hosts[@]}; do
461 echo -n > ${cmdsf}_${host}
463 for ((idx = 0; idx < ndevs; idx++)); do
464 host=${host_names[$idx]}
465 devno=${devnos[$idx]}
467 first_obj=${first_objs[$idx]}
468 thr_per_obj=$((${thr}/${nobj}))
469 echo >> ${cmdsf}_${host} \
470 "$lctl > $tmpfi 2>&1 \\
471 --threads $thr -$snap $devno \\
472 test_brw $count `testname2type $test` q $pages ${thr_per_obj}t${first_obj} &"
475 for host in ${unique_hosts[@]}; do
476 echo "wait" >> ${cmdsf}_${host}
477 pidarray[$pidcount]=0
478 pidcount=$((pidcount+1))
480 # timed run of all the per-host script files
483 for host in ${unique_hosts[@]}; do
484 # brutal hack to deal with a non-shared /tmp
485 scp -q ${cmdsf}_${host} ${host}:/tmp > /dev/null
486 remote_shell $host bash ${cmdsf}_${host} &
487 pidarray[$pidcount]=$!
488 pidcount=$((pidcount+1))
491 for host in ${unique_hosts[@]}; do
492 wait ${pidarray[$pidcount]}
493 pidcount=$((pidcount+1))
497 # clean up per-host script files
498 for host in ${unique_hosts[@]}; do
501 # compute bandwidth from total data / elapsed time
502 str=`awk "BEGIN {printf \"%7.2f \",\
503 $total_size / (( $t1 - $t0 ) * 1024)}"`
504 print_summary -n "$str"
505 # collect/check individual OST stats
507 for ((idx = 0; idx < ndevs; idx++)); do
508 client_name="${host_names[$idx]}:${client_names[$idx]}"
510 echo "=============> $test $client_name" >> $workf
511 host="${host_names[$idx]}"
512 scp -q ${host}:$tmpfi $tmpfi > /dev/null
514 get_stats $tmpfi >> $tmpf
517 # compute/display global min/max stats
518 echo "=============> $test global" >> $workf
520 stats=(`get_global_stats $tmpf`)
522 if ((stats[0] <= 0)); then
523 if ((stats[0] < 0)); then
524 str=`printf "%17s " ERROR`
526 str=`printf "%17s " SHORT`
529 str=`awk "BEGIN {printf \"[%7.2f,%7.2f] \",\
530 (${stats[1]} * $actual_rsz)/1024,\
531 (${stats[2]} * $actual_rsz)/1024; exit}"`
533 print_summary -n "$str"
536 # destroy objects we created
537 for ((idx = 0; idx < ndevs; idx++)); do
538 host=${host_names[$idx]}
539 devno=${devnos[$idx]}
540 client_name="${host}:${client_names[$idx]}"
541 first_obj=${first_objs[$idx]}
542 echo "=============> Destroy $nobj on $client_name" >> $workf
543 destroy_objects $host $devno $first_obj $nobj $tmpf