5 LUSTRE=${LUSTRE:-$(cd $(dirname $0)/..; echo $PWD)}
6 . $LUSTRE/tests/test-framework.sh
8 . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
12 ALWAYS_EXCEPT="parallel_grouplock $PARALLEL_SCALE_EXCEPT"
17 cbench_DIR=${cbench_DIR:-""}
18 cbench_IDIRS=${cbench_IDIRS:-4}
19 cbench_RUNS=${cbench_RUNS:-4} # FIXME: wiki page requirements is 30, do we really need 30 ?
21 if [ "$SLOW" = "no" ]; then
29 METABENCH=${METABENCH:-$(which metabench 2> /dev/null || true)}
30 mbench_NFILES=${mbench_NFILES:-30400}
31 [ "$SLOW" = "no" ] && mbench_NFILES=10000
32 MACHINEFILE=${MACHINEFILE:-$TMP/$(basename $0 .sh).machines}
34 mbench_THREADS=${mbench_THREADS:-4}
39 SIMUL=${SIMUL:=$(which simul 2> /dev/null || true)}
41 simul_THREADS=${simul_THREADS:-2}
42 simul_REP=${simul_REP:-20}
43 [ "$SLOW" = "no" ] && simul_REP=2
48 cnt_DIR=${cnt_DIR:-""}
49 cnt_NRUN=${cnt_NRUN:-10}
50 [ "$SLOW" = "no" ] && cnt_NRUN=2
55 CASC_RW=${CASC_RW:-$(which cascading_rw 2> /dev/null || true)}
57 casc_THREADS=${casc_THREADS:-2}
58 casc_REP=${casc_REP:-300}
59 [ "$SLOW" = "no" ] && casc_REP=10
64 IOR=${IOR:-$(which IOR 2> /dev/null || true)}
66 ior_THREADS=${ior_THREADS:-2}
67 ior_blockSize=${ior_blockSize:-6} # Gb
68 ior_DURATION=${ior_DURATION:-30} # minutes
69 [ "$SLOW" = "no" ] && ior_DURATION=5
72 # write_append_truncate
75 write_THREADS=${write_THREADS:-8}
76 write_REP=${write_REP:-10000}
77 [ "$SLOW" = "no" ] && write_REP=100
82 WRITE_DISJOINT=${WRITE_DISJOINT:-$(which write_disjoint 2> /dev/null || true)}
84 wdisjoint_THREADS=${wdisjoint_THREADS:-4}
85 wdisjoint_REP=${wdisjoint_REP:-10000}
86 [ "$SLOW" = "no" ] && wdisjoint_REP=100
92 PARALLEL_GROUPLOCK=${PARALLEL_GROUPLOCK:-$(which parallel_grouplock 2> /dev/null || true)}
93 parallel_grouplock_MINTASKS=${parallel_grouplock_MINTASKS:-5}
96 check_and_setup_lustre
98 get_mpiuser_id $MPI_USER
99 MPI_RUNAS=${MPI_RUNAS:-"runas -u $MPI_USER_UID -g $MPI_USER_GID"}
100 $GSS_KRB5 && refresh_krb5_tgt $MPI_USER_UID $MPI_USER_GID $MPI_RUNAS
109 echo "${var}=${!var}"
111 [ -e $MACHINEFILE ] && cat $MACHINEFILE
115 # 5 min * cbench_RUNS
119 # compile dir kernel-1 680MB
120 # required space 680MB * cbench_IDIRS = ~7 Gb
122 test_compilebench() {
123 print_opts cbench_DIR cbench_IDIRS cbench_RUNS
125 [ x$cbench_DIR = x ] &&
126 { skip_env "compilebench not found" && return; }
128 [ -e $cbench_DIR/compilebench ] || \
129 { skip_env "No compilebench build" && return; }
131 local space=$(df -P $DIR | tail -n 1 | awk '{ print $4 }')
132 if [ $space -le $((680 * 1024 * cbench_IDIRS)) ]; then
133 cbench_IDIRS=$(( space / 680 / 1024))
134 [ $cbench_IDIRS = 0 ] && \
135 skip_env "Need free space atleast 680 Mb, have $space" && return
137 log free space=$space, reducing initial dirs to $cbench_IDIRS
140 # t-f _base needs to be modifyed to set properly tdir
141 # for new "test_foo" functions names
142 # local testdir=$DIR/$tdir
143 local testdir=$DIR/d0.compilebench
148 local cmd="./compilebench -D $testdir -i $cbench_IDIRS -r $cbench_RUNS --makej"
157 [ $rc = 0 ] || error "compilebench failed: $rc"
160 run_test compilebench "compilebench"
163 [ x$METABENCH = x ] &&
164 { skip_env "metabench not found" && return; }
166 local clients=$CLIENTS
167 [ -z $clients ] && clients=$(hostname)
169 num_clients=$(get_node_count ${clients//,/ })
172 # Need space estimation here.
174 generate_machine_file $clients $MACHINEFILE || return $?
176 print_opts METABENCH clients mbench_NFILES mbench_THREADS
178 local testdir=$DIR/d0.metabench
180 # mpi_run uses mpiuser
183 # -C Run the file creation tests.
184 # -S Run the file stat tests.
185 # -c nfile Number of files to be used in each test.
186 # -k Cleanup. Remove the test directories.
187 local cmd="$METABENCH -w $testdir -c $mbench_NFILES -C -S -k"
189 mpi_run -np $((num_clients * $mbench_THREADS)) -machinefile ${MACHINEFILE} $cmd
191 if [ $rc != 0 ] ; then
192 error "metabench failed! $rc"
196 run_test metabench "metabench"
199 if [ "$NFSCLIENT" ]; then
200 skip "skipped for NFSCLIENT mode"
205 { skip_env "simul not found" && return; }
207 local clients=$CLIENTS
208 [ -z $clients ] && clients=$(hostname)
210 local num_clients=$(get_node_count ${clients//,/ })
213 # Need space estimation here.
215 generate_machine_file $clients $MACHINEFILE || return $?
217 print_opts SIMUL clients simul_REP simul_THREADS
219 local testdir=$DIR/d0.simul
221 # mpi_run uses mpiuser
224 # -n # : repeat each test # times
225 # -N # : repeat the entire set of tests # times
227 local cmd="$SIMUL -d $testdir -n $simul_REP -N $simul_REP"
230 mpi_run -np $((num_clients * $simul_THREADS)) -machinefile ${MACHINEFILE} $cmd
233 if [ $rc != 0 ] ; then
234 error "simul failed! $rc"
238 run_test simul "simul"
240 test_connectathon() {
241 print_opts cnt_DIR cnt_NRUN
244 { skip_env "connectathon dir not found" && return; }
246 [ -e $cnt_DIR/runtests ] || \
247 { skip_env "No connectathon runtests found" && return; }
249 local testdir=$DIR/d0.connectathon
256 # cthon options (must be in this order)
258 # -N numpasses - will be passed to the runtests script. This argument
259 # is optional. It specifies the number of times to run
262 # One of these test types
267 # -a all of the above
269 # -f a quick functionality test
273 # Include lock tests unless we're running on nfsv4
274 local fstype=$(df -TP $testdir | awk 'NR==2 {print $2}')
275 echo "$testdir: $fstype"
276 if [[ $fstype != "nfs4" ]]; then
280 for test in $tests; do
281 local cmd="./runtests -N $cnt_NRUN $test -f $testdir"
287 [ $rc = 0 ] || error "connectathon failed: $rc"
293 run_test connectathon "connectathon"
297 { skip_env "IOR not found" && return; }
299 local clients=$CLIENTS
300 [ -z $clients ] && clients=$(hostname)
302 local num_clients=$(get_node_count ${clients//,/ })
304 local space=$(df -P $DIR | tail -n 1 | awk '{ print $4 }')
305 echo "+ $ior_blockSize * 1024 * 1024 * $num_clients * $ior_THREADS "
306 if [ $((space / 2)) -le $(( ior_blockSize * 1024 * 1024 * num_clients * ior_THREADS)) ]; then
307 echo "+ $space * 9/10 / 1024 / 1024 / $num_clients / $ior_THREADS"
308 ior_blockSize=$(( space /2 /1024 /1024 / num_clients / ior_THREADS ))
309 [ $ior_blockSize = 0 ] && \
310 skip_env "Need free space more than ($num_clients * $ior_THREADS )Gb: $((num_clients*ior_THREADS *1024 *1024*2)), have $space" && return
312 echo "free space=$space, Need: $num_clients x $ior_THREADS x $ior_blockSize Gb (blockSize reduced to $ior_blockSize Gb)"
315 generate_machine_file $clients $MACHINEFILE || return $?
317 print_opts IOR ior_THREADS ior_DURATION MACHINEFILE
319 local testdir=$DIR/d0.ior
321 # mpi_run uses mpiuser
323 if [ "$NFSCLIENT" ]; then
324 setstripe_nfsserver $testdir -c -1 ||
325 { error "setstripe on nfsserver failed" && return 1; }
327 $LFS setstripe $testdir -c -1 ||
328 { error "setstripe failed" && return 2; }
331 # -b N blockSize -- contiguous bytes to write per task (e.g.: 8, 4k, 2m, 1g)"
333 # -t N transferSize -- size of transfer in bytes (e.g.: 8, 4k, 2m, 1g)"
334 # -w writeFile -- write file"
335 # -r readFile -- read existing file"
336 # -T maxTimeDuration -- max time in minutes to run tests"
337 # -k keepFile -- keep testFile(s) on program exit
338 local cmd="$IOR -a POSIX -b ${ior_blockSize}g -o $testdir/iorData -t 2m -v -w -r -T $ior_DURATION -k"
341 mpi_run -np $((num_clients * $ior_THREADS)) -machinefile ${MACHINEFILE} $cmd
344 if [ $rc != 0 ] ; then
345 error "ior failed! $rc"
351 test_cascading_rw() {
352 if [ "$NFSCLIENT" ]; then
353 skip "skipped for NFSCLIENT mode"
358 { skip_env "cascading_rw not found" && return; }
360 local clients=$CLIENTS
361 [ -z $clients ] && clients=$(hostname)
363 num_clients=$(get_node_count ${clients//,/ })
366 # Need space estimation here.
368 generate_machine_file $clients $MACHINEFILE || return $?
370 print_opts CASC_RW clients casc_THREADS casc_REP MACHINEFILE
372 local testdir=$DIR/d0.cascading_rw
374 # mpi_run uses mpiuser
378 # -n: repeat test # times
380 local cmd="$CASC_RW -g -d $testdir -n $casc_REP"
383 mpi_run -np $((num_clients * $casc_THREADS)) -machinefile ${MACHINEFILE} $cmd
386 if [ $rc != 0 ] ; then
387 error "cascading_rw failed! $rc"
391 run_test cascading_rw "cascading_rw"
393 test_write_append_truncate() {
394 if [ "$NFSCLIENT" ]; then
395 skip "skipped for NFSCLIENT mode"
399 # location is lustre/tests dir
400 if ! which write_append_truncate > /dev/null 2>&1 ; then
401 skip_env "write_append_truncate not found"
405 local clients=$CLIENTS
406 [ -z $clients ] && clients=$(hostname)
408 local num_clients=$(get_node_count ${clients//,/ })
411 # Need space estimation here.
413 generate_machine_file $clients $MACHINEFILE || return $?
415 local testdir=$DIR/d0.write_append_truncate
416 local file=$testdir/f0.wat
418 print_opts clients write_REP write_THREADS MACHINEFILE
421 # mpi_run uses mpiuser
424 local cmd="write_append_truncate -n $write_REP $file"
427 mpi_run -np $((num_clients * $write_THREADS)) -machinefile ${MACHINEFILE} $cmd
430 if [ $rc != 0 ] ; then
431 error "write_append_truncate failed! $rc"
436 run_test write_append_truncate "write_append_truncate"
438 test_write_disjoint() {
439 if [ "$NFSCLIENT" ]; then
440 skip "skipped for NFSCLIENT mode"
444 [ x$WRITE_DISJOINT = x ] &&
445 { skip_env "write_disjoint not found" && return; }
447 local clients=$CLIENTS
448 [ -z $clients ] && clients=$(hostname)
450 local num_clients=$(get_node_count ${clients//,/ })
453 # Need space estimation here.
455 generate_machine_file $clients $MACHINEFILE || return $?
457 print_opts WRITE_DISJOINT clients wdisjoint_THREADS wdisjoint_REP MACHINEFILE
458 local testdir=$DIR/d0.write_disjoint
460 # mpi_run uses mpiuser
463 local cmd="$WRITE_DISJOINT -f $testdir/file -n $wdisjoint_REP"
466 mpi_run -np $((num_clients * $wdisjoint_THREADS)) -machinefile ${MACHINEFILE} $cmd
469 if [ $rc != 0 ] ; then
470 error "write_disjoint failed! $rc"
474 run_test write_disjoint "write_disjoint"
476 test_parallel_grouplock() {
477 if [ "$NFSCLIENT" ]; then
478 skip "skipped for NFSCLIENT mode"
482 [ x$PARALLEL_GROUPLOCK = x ] &&
483 { skip "PARALLEL_GROUPLOCK not found" && return; }
485 local clients=$CLIENTS
486 [ -z $clients ] && clients=$(hostname)
488 local num_clients=$(get_node_count ${clients//,/ })
490 generate_machine_file $clients $MACHINEFILE || return $?
492 print_opts clients parallel_grouplock_MINTASKS MACHINEFILE
494 local testdir=$DIR/d0.parallel_grouplock
496 # mpi_run uses mpiuser
499 do_nodes $clients "lctl set_param llite.*.max_rw_chunk=0" ||
500 error "set_param max_rw_chunk=0 failed "
505 for i in $(seq 12); do
507 local cmd="$PARALLEL_GROUPLOCK -g -v -d $testdir $subtest"
510 mpi_run -np $parallel_grouplock_MINTASKS -machinefile ${MACHINEFILE} $cmd
512 if [ $rc != 0 ] ; then
513 error_noexit "parallel_grouplock subtests $subtest failed! $rc"
515 echo "parallel_grouplock subtests $subtest PASS"
517 let status=$((status + rc))
518 # clear debug to collect one log per one test
519 do_nodes $(comma_list $(nodes_list)) lctl clear
521 [ $status -eq 0 ] || error "parallel_grouplock status: $status"
524 run_test parallel_grouplock "parallel_grouplock"
526 statahead_NUMMNTPTS=${statahead_NUMMNTPTS:-5}
527 statahead_NUMFILES=${statahead_NUMFILES:-500000}
529 cleanup_statahead () {
536 for i in $(seq 0 $num_mntpts);do
537 zconf_umount_clients $clients ${mntpt_root}$i ||
538 error_exit "Failed to umount lustre on ${mntpt_root}$i"
543 if [[ -n $NFSCLIENT ]]; then
544 skip "Statahead testing is not supported on NFS clients."
549 { skip_env "mdsrate not found" && return; }
551 local clients=$CLIENTS
552 [ -z $clients ] && clients=$(hostname)
554 local num_clients=$(get_node_count ${clients//,/ })
556 generate_machine_file $clients $MACHINEFILE || return $?
558 print_opts MDSRATE clients statahead_NUMMNTPTS statahead_NUMFILES
562 # do not use default "d[0-9]*" dir name
563 # to avoid of rm $statahead_NUMFILES (500k) files in t-f cleanup
565 local testdir=$DIR/$dir
567 # cleanup only if dir exists
568 # cleanup only $statahead_NUMFILES number of files
569 # ignore the other files created by someone else
571 mdsrate_cleanup $((num_clients * 32)) $MACHINEFILE $statahead_NUMFILES $testdir 'f%%d' --ignore
574 # mpi_run uses mpiuser
577 local num_files=$statahead_NUMFILES
579 local IFree=$(inodes_available)
580 if [ $IFree -lt $num_files ]; then
586 local cmd="${MDSRATE} ${MDSRATE_DEBUG} --mknod --dir $testdir --nfiles $num_files --filefmt 'f%%d'"
589 mpi_run -np $((num_clients * 32)) -machinefile ${MACHINEFILE} $cmd
592 if [ $rc != 0 ] ; then
593 error "mdsrate failed to create $rc"
597 local num_mntpts=$statahead_NUMMNTPTS
598 local mntpt_root=$TMP/mntpt/lustre
599 mntopts=${MNTOPTSTATAHEAD:-$MOUNTOPT}
601 echo "Mounting $num_mntpts lustre clients starts on $clients"
602 trap "cleanup_statahead $clients $mntpt_root $num_mntpts" EXIT ERR
603 for i in $(seq 0 $num_mntpts); do
604 zconf_mount_clients $clients ${mntpt_root}$i $mntopts ||
605 error_exit "Failed to mount lustre on ${mntpt_root}$i on $clients"
608 do_rpc_nodes $clients cancel_lru_locks mdc
610 do_rpc_nodes $clients do_ls $mntpt_root $num_mntpts $dir
612 mdsrate_cleanup $((num_clients * 32)) $MACHINEFILE $num_files $testdir 'f%%d' --ignore
614 # use rm instead of rmdir because of
615 # testdir could contain the files created by someone else,
616 # or by previous run where is num_files prev > num_files current
618 cleanup_statahead $clients $mntpt_root $num_mntpts
621 run_test statahead "statahead test, multiple clients"
623 equals_msg `basename $0`: test complete, cleaning up
624 check_and_cleanup_lustre
625 [ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG && grep -q FAIL $TESTSUITELOG && exit 1 || true