3 # Simple function used by run_*.sh scripts
8 if [ -z "${!name}" ]; then
9 echo "$0: $name must be set"
13 [ $failed ] && exit 1 || true
16 # lrepl - Lustre test Read-Eval-Print Loop.
18 # This function implements a REPL for the Lustre test framework. It
19 # doesn't exec an actual shell because the user may want to inspect
20 # variables and use functions from the test framework.
27 This is an interactive read-eval-print loop interactive shell
28 simulation that you can use to debug failing tests. You can
29 enter most bash command lines (see notes below).
31 Use this REPL to inspect variables, set them, call test
32 framework shell functions, etcetera.
34 'exit' or EOF to exit this shell.
36 set \$retcode to 0 to cause the assertion failure that
37 triggered this REPL to be ignored.
40 do_facet ost1 lctl get_param ost.*.ost.threads_*
41 do_rpc_nodes \$OSTNODES unload_modules
44 All but the last line of multi-line statements or blocks
45 must end in a backslash.
47 "Here documents" are not supported.
49 History is not supported, but command-line editing is.
53 # Prompt escapes don't work in read -p, sadly.
54 prompt=":test_${testnum:-UNKNOWN}:$(uname -n):$(basename $PWD)% "
56 # We use read -r to get close to a shell experience
57 while read -e -r -p "$prompt" rawline; do
60 # Don't want to exit-exit, just exit the REPL
62 # We need to handle continuations, and read -r doesn't do
63 # that for us. Yet we need read -r.
65 # We also use case/esac to compare lines read to "*\\"
66 # because [ "$line" = *\\ ] and variants of that don't work.
68 while read -e -r -p '> ' rawline
70 line="$line"$'\n'"$rawline"
72 # We could check for here documents by matching
73 # against *<<*, but who cares.
86 # Finally! Time to eval.
90 echo $'\n\tExiting interactive shell...\n'
94 # lassert - Lustre test framework assert
96 # Arguments: failure code, failure message, expression/statement
98 # lassert evaluates the expression given, and, if false, calls
99 # error() to trigger test failure. If REPL_ON_LASSERT is true then
100 # lassert will call lrepl() to give the user an interactive shell.
101 # If the REPL sets retcode=0 then the assertion failure will be
108 echo "checking $* ($(eval echo \""$*"\"))..."
109 eval "$@" && return 0;
111 if ${REPL_ON_LASSERT:-false}; then
112 echo "Assertion $retcode failed: $* (expanded: $(eval echo \""$*"\"))
117 error "Assertion $retcode failed: $* (expanded: $(eval echo \""$*"\"))
122 # setmodopts- set module options for subsequent calls to load_modules
124 # Usage: setmodopts module_name new_value [var_in_which_to_save_old_value]
125 # setmodopts -a module_name new_value [var_in_which_to_save_old_value]
127 # In the second usage the new value is appended to the old.
131 if [ "$1" = -a ]; then
136 local _var=MODOPTS_$1
141 # Dynamic naming of variables is a pain in bash. In ksh93 we could
142 # write "nameref opts_var=${modname}_MODOPTS" then assign directly
143 # to opts_var. Associative arrays would also help, alternatively.
144 # Alas, we're stuck with eval until all distros move to a more recent
145 # version of bash. Fortunately we don't need to eval unset and export.
147 if [ -z "$_newvalue" ]; then
153 $_append && _newvalue="$_oldvalue $_newvalue"
154 export $_var="$_newvalue"
155 echo setmodopts: ${_var}=${_newvalue}
157 [ -n "$_savevar" ] && eval $_savevar=\""$_oldvalue"\"
160 echoerr () { echo "$@" 1>&2 ; }
163 echoerr "$(date +'%F %H:%M:%S'): client load was signaled to terminate"
165 local PGID=$(ps -eo "%c %p %r" | awk "/ $PPID / {print \$3}")
172 local mpirun="$MPIRUN $MPIRUN_OPTIONS --oversubscribe"
173 local command="$mpirun $@"
174 local mpilog=$TMP/mpi.log
177 if [ -n "$MPI_USER" -a "$MPI_USER" != root -a -n "$mpirun" ]; then
178 echo "+ chmod 0777 $MOUNT"
180 command="su $MPI_USER sh -c \"$command \""
185 eval $command 2>&1 | tee $mpilog || true
188 if [ $rc -eq 0 ] && grep -q "p4_error:" $mpilog ; then
197 for i in ${1//,/ }; do
198 if [ "$list" = "" ]; then
201 list="$list$escape $i@$NETTYPE"
207 # FIXME: all setup/cleanup can be done without rpc.sh
210 [ x$1 = x--verbose ] && verbose=true
212 export LST_SESSION=`$LST show_session 2>/dev/null | awk -F " " '{print $5}'`
213 [ "$LST_SESSION" == "" ] && return
222 lst_session_cleanup_all () {
223 local list=$(comma_list $(nodes_list))
224 do_rpc_nodes $list lst_end_session
228 lsmod | grep -q lnet_selftest && \
229 rmmod lnet_selftest > /dev/null 2>&1 || true
233 local list=$(comma_list $(nodes_list))
235 # lst end_session needs to be executed only locally
236 # i.e. on node where lst new_session was called
237 lst_end_session --verbose
238 do_rpc_nodes $list lst_cleanup
242 load_module lnet_selftest
246 local list=$(comma_list $(nodes_list))
247 do_rpc_nodes $list lst_setup
253 # Passed a single argument, strips everything off following
254 # and includes the first period.
255 # client-20.lab.whamcloud.com becomes client-20
257 echo $(sed 's/\..*//' <<< $1)
263 # Find remote nodename, stripped of any domain, etc.
264 # 'hostname -s' is easy, but not implemented on all systems
266 local rname=$(do_node $1 "uname -n" || echo -1)
267 if [[ "$rname" = "-1" ]]; then
270 echo $(short_hostname $rname)
280 echo "${var}=${!var}"
282 [ -e $MACHINEFILE ] && cat $MACHINEFILE
287 local cbench_DIR=${cbench_DIR:-""}
288 local cbench_IDIRS=${cbench_IDIRS:-2}
289 local cbench_RUNS=${cbench_RUNS:-2}
291 print_opts cbench_DIR cbench_IDIRS cbench_RUNS
293 [ x$cbench_DIR = x ] &&
294 skip_env "compilebench not found"
296 [ -e $cbench_DIR/compilebench ] ||
297 skip_env "No compilebench build"
300 # compile dir kernel-0 ~1GB
301 # required space ~1GB * cbench_IDIRS
302 local space=$(df -P $dir | tail -n 1 | awk '{ print $4 }')
303 if [[ $space -le $((1024 * 1024 * cbench_IDIRS)) ]]; then
304 cbench_IDIRS=$((space / 1024 / 1024))
305 [[ $cbench_IDIRS -eq 0 ]] &&
306 skip_env "Need free space at least 1GB, have $space"
308 echo "reducing initial dirs to $cbench_IDIRS"
310 echo "free space = $space KB"
313 # t-f _base needs to be modifyed to set properly tdir
314 # for new "test_foo" functions names
315 # local testdir=$DIR/$tdir
316 local testdir=$dir/d0.compilebench.$$
321 local cmd="./compilebench -D $testdir -i $cbench_IDIRS \
322 -r $cbench_RUNS --makej"
331 [ $rc = 0 ] || error "compilebench failed: $rc"
337 local mntpt=${2:-$MOUNT}
338 METABENCH=${METABENCH:-$(which metabench 2> /dev/null || true)}
339 mbench_NFILES=${mbench_NFILES:-30400}
341 mbench_THREADS=${mbench_THREADS:-4}
342 mbench_OPTIONS=${mbench_OPTIONS:-}
343 mbench_CLEANUP=${mbench_CLEANUP:-true}
345 [ x$METABENCH = x ] && skip_env "metabench not found"
347 print_opts METABENCH clients mbench_NFILES mbench_THREADS
349 local testdir=$dir/d0.metabench
351 # mpi_run uses mpiuser
354 # -C Run the file creation tests. Creates zero byte files.
355 # -S Run the file stat tests.
356 # -c nfile Number of files to be used in each test.
357 # -k Cleanup files when finished.
358 local cmd="$METABENCH -w $testdir -c $mbench_NFILES -C -S $mbench_OPTIONS"
361 # find out if we need to use srun by checking $SRUN_PARTITION
362 if [ "$SRUN_PARTITION" ]; then
363 $SRUN $SRUN_OPTIONS -D $testdir -w $clients -N $num_clients \
364 -n $((num_clients * mbench_THREADS)) \
365 -p $SRUN_PARTITION -- $cmd
367 mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \
368 -np $((num_clients * $mbench_THREADS)) $cmd
372 if [ $rc != 0 ] ; then
373 error "metabench failed! $rc"
376 if $mbench_CLEANUP; then
379 mv $dir/d0.metabench $mntpt/_xxx.$(date +%s).d0.metabench
384 SIMUL=${SIMUL:=$(which simul 2> /dev/null || true)}
385 [ x$SIMUL = x ] && skip_env "simul not found"
386 [ "$NFSCLIENT" ] && skip "skipped for NFSCLIENT mode"
389 simul_THREADS=${simul_THREADS:-2}
390 simul_REP=${simul_REP:-20}
393 # Need space estimation here.
395 print_opts SIMUL clients simul_REP simul_THREADS
397 local testdir=$DIR/d0.simul
399 # mpi_run uses mpiuser
402 # -n # : repeat each test # times
403 # -N # : repeat the entire set of tests # times
405 local cmd="$SIMUL -d $testdir -n $simul_REP -N $simul_REP"
408 # find out if we need to use srun by checking $SRUN_PARTITION
409 if [ "$SRUN_PARTITION" ]; then
410 $SRUN $SRUN_OPTIONS -D $testdir -w $clients -N $num_clients \
411 -n $((num_clients * simul_THREADS)) -p $SRUN_PARTITION \
414 mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \
415 -np $((num_clients * simul_THREADS)) $cmd
419 if [ $rc != 0 ] ; then
420 error "simul failed! $rc"
426 MDTEST=${MDTEST:=$(which mdtest 2> /dev/null || true)}
427 [ x$MDTEST = x ] && skip_env "mdtest not found"
428 [ "$NFSCLIENT" ] && skip "skipped for NFSCLIENT mode"
431 mdtest_THREADS=${mdtest_THREADS:-2}
432 mdtest_nFiles=${mdtest_nFiles:-"100000"}
433 # We devide the files by number of core
434 mdtest_nFiles=$((mdtest_nFiles/mdtest_THREADS/num_clients))
435 mdtest_iteration=${mdtest_iteration:-1}
436 local mdtest_custom_params=${mdtest_custom_params:-""}
437 local type=${1:-"ssf"}
440 # Need space estimation here.
442 print_opts MDTEST mdtest_iteration mdtest_THREADS mdtest_nFiles
444 local testdir=$DIR/d0.mdtest
446 # mpi_run uses mpiuser
449 # -i # : repeat each test # times
451 # -n # : number of file/dir to create/stat/remove
452 # -u : each process create/stat/remove individually
454 local cmd="$MDTEST -d $testdir -i $mdtest_iteration \
455 -n $mdtest_nFiles $mdtest_custom_params"
457 [ $type = "fpp" ] && cmd="$cmd -u"
460 # find out if we need to use srun by checking $SRUN_PARTITION
461 if [ "$SRUN_PARTITION" ]; then
462 $SRUN $SRUN_OPTIONS -D $testdir -w $clients -N $num_clients \
463 -n $((num_clients * mdtest_THREADS)) \
464 -p $SRUN_PARTITION -- $cmd
466 mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \
467 -np $((num_clients * mdtest_THREADS)) $cmd
471 if [ $rc != 0 ] ; then
472 error "mdtest failed! $rc"
479 cnt_DIR=${cnt_DIR:-""}
480 cnt_NRUN=${cnt_NRUN:-10}
482 print_opts cnt_DIR cnt_NRUN
484 [ x$cnt_DIR = x ] && skip_env "connectathon dir not found"
485 [ -e $cnt_DIR/runtests ] || skip_env "No connectathon runtests found"
488 # "special" tests create a 30 MB file + misc. small files
489 # required space ~40 MB
490 local space=$(df -P $dir | tail -n 1 | awk '{ print $4 }')
491 if [[ $space -le $((1024 * 40)) ]]; then
492 skip_env "Need free space at least 40MB, have $space KB"
494 echo "free space = $space KB"
496 local testdir=$dir/d0.connectathon
503 # To run connectathon:
504 # runtests [-a|-b|-g|-s|-l] [-f|-n|-t] [-N numpasses] [test-directory]
506 # One of the following test types
511 # -a all of the above
513 # -f a quick functional test
514 # -n suppress directory operations (mkdir and rmdir)
515 # -t run with time statistics (default for basic tests)
517 # -N numpasses - specifies the number of times to run
518 # the tests. Optional.
521 # Include lock tests unless we're running on nfsv4
522 local fstype=$(df -TP $testdir | awk 'NR==2 {print $2}')
523 echo "$testdir: $fstype"
524 if [[ $fstype != "nfs4" ]]; then
528 for test in $tests; do
529 local cmd="sh ./runtests -N $cnt_NRUN $test -f $testdir"
535 [ $rc = 0 ] || error "connectathon failed: $rc"
543 local type=${1:="ssf"}
545 local testdir=$dir/d0.ior.$type
546 local nfs_srvmntpt=$3
548 if [ "$NFSCLIENT" ]; then
549 [[ -n $nfs_srvmntpt ]] ||
550 { error "NFSCLIENT mode, but nfs exported dir"\
551 "is not set!" && return 1; }
554 IOR=${IOR:-$(which IOR 2> /dev/null || true)}
555 [ x$IOR = x ] && skip_env "IOR not found"
558 ior_THREADS=${ior_THREADS:-2}
559 ior_iteration=${ior_iteration:-1}
560 ior_blockSize=${ior_blockSize:-6}
561 ior_blockUnit=${ior_blockUnit:-M} # K, M, G
562 ior_xferSize=${ior_xferSize:-1M}
563 ior_type=${ior_type:-POSIX}
564 ior_DURATION=${ior_DURATION:-30} # minutes
566 case ${ior_blockUnit} in
568 multiplier=$((1024 * 1024 * 1024))
571 multiplier=$((1024 * 1024))
576 *) error "Incorrect block unit should be one of [KMG]"
580 # calculate the space in bytes
581 local space=$(df -B 1 -P $dir | tail -n 1 | awk '{ print $4 }')
582 local total_threads=$((num_clients * ior_THREADS))
583 echo "+ $ior_blockSize * $multiplier * $total_threads "
584 if [ $((space / 2)) -le \
585 $((ior_blockSize * multiplier * total_threads)) ]; then
586 ior_blockSize=$((space / 2 / multiplier / total_threads))
587 [ $ior_blockSize -eq 0 ] &&
588 skip_env "Need free space more than $((2 * total_threads)) \
589 ${ior_blockUnit}: have $((space / multiplier))"
591 echo "(reduced blockSize to $ior_blockSize \
592 ${ior_blockUnit} bytes)"
595 print_opts IOR ior_THREADS ior_DURATION MACHINEFILE
598 # mpi_run uses mpiuser
600 if [ -z "$NFSCLIENT" ]; then
601 $LFS setstripe $testdir -c -1 ||
602 { error "setstripe failed" && return 2; }
607 # contiguous bytes to write per task (e.g.: 8, 4K, 2M, 1G)"
609 # -t N transferSize -- size of transfer in bytes (e.g.: 8, 4K, 2M, 1G)"
610 # -w writeFile -- write file"
611 # -r readFile -- read existing file"
612 # -W checkWrite -- check read after write"
613 # -C reorderTasks -- changes task ordering to n+1 ordering for readback
614 # -T maxTimeDuration -- max time in minutes to run tests"
615 # -k keepFile -- keep testFile(s) on program exit
617 local cmd="$IOR -a $ior_type -b ${ior_blockSize}${ior_blockUnit} \
618 -o $testdir/iorData -t $ior_xferSize -v -C -w -r -W \
619 -i $ior_iteration -T $ior_DURATION -k"
621 [ $type = "fpp" ] && cmd="$cmd -F"
624 # find out if we need to use srun by checking $SRUN_PARTITION
625 if [ "$SRUN_PARTITION" ]; then
626 $SRUN $SRUN_OPTIONS -D $testdir -w $clients -N $num_clients \
627 -n $((num_clients * ior_THREADS)) -p $SRUN_PARTITION \
630 mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \
631 -np $((num_clients * $ior_THREADS)) $cmd
635 if [ $rc != 0 ] ; then
636 error "ior failed! $rc"
642 MIB=${MIB:=$(which mib 2> /dev/null || true)}
643 [ "$NFSCLIENT" ] && skip "skipped for NFSCLIENT mode"
644 [ x$MIB = x ] && skip_env "MIB not found"
647 mib_THREADS=${mib_THREADS:-2}
648 mib_xferSize=${mib_xferSize:-1m}
649 mib_xferLimit=${mib_xferLimit:-5000}
650 mib_timeLimit=${mib_timeLimit:-300}
652 print_opts MIB mib_THREADS mib_xferSize mib_xferLimit mib_timeLimit \
655 local testdir=$DIR/d0.mib
657 # mpi_run uses mpiuser
659 $LFS setstripe $testdir -c -1 ||
660 error "setstripe failed"
662 # -I Show intermediate values in output
663 # -H Show headers in output
664 # -L Do not issue new system calls after this many seconds
665 # -s Use system calls of this size
667 # -l Issue no more than this many system calls
668 local cmd="$MIB -t $testdir -s $mib_xferSize -l $mib_xferLimit \
669 -L $mib_timeLimit -HI -p mib.$(date +%Y%m%d%H%M%S)"
672 # find out if we need to use srun by checking $SRUN_PARTITION
673 if [ "$SRUN_PARTITION" ]; then
674 $SRUN $SRUN_OPTIONS -D $testdir -w $clients -N $num_clients \
675 -n $((num_clients * mib_THREADS)) -p $SRUN_PARTITION \
678 mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \
679 -np $((num_clients * mib_THREADS)) $cmd
683 if [ $rc != 0 ] ; then
684 error "mib failed! $rc"
690 CASC_RW=${CASC_RW:-$(which cascading_rw 2> /dev/null || true)}
691 [ x$CASC_RW = x ] && skip_env "cascading_rw not found"
692 [ "$NFSCLIENT" ] && skip "skipped for NFSCLIENT mode"
695 casc_THREADS=${casc_THREADS:-2}
696 casc_REP=${casc_REP:-300}
699 # Need space estimation here.
701 print_opts CASC_RW clients casc_THREADS casc_REP MACHINEFILE
703 local testdir=$DIR/d0.cascading_rw
705 # mpi_run uses mpiuser
709 # -n: repeat test # times
711 local cmd="$CASC_RW -g -d $testdir -n $casc_REP"
714 mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \
715 -np $((num_clients * $casc_THREADS)) $cmd
718 if [ $rc != 0 ] ; then
719 error "cascading_rw failed! $rc"
724 run_write_append_truncate() {
725 [ "$NFSCLIENT" ] && skip "skipped for NFSCLIENT mode"
726 # location is lustre/tests dir
727 if ! which write_append_truncate > /dev/null 2>&1 ; then
728 skip_env "write_append_truncate not found"
732 write_THREADS=${write_THREADS:-8}
733 write_REP=${write_REP:-10000}
736 # Need space estimation here.
738 local testdir=$DIR/d0.write_append_truncate
739 local file=$testdir/f0.wat
741 print_opts clients write_REP write_THREADS MACHINEFILE
744 # mpi_run uses mpiuser
747 local cmd="write_append_truncate -n $write_REP $file"
750 mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \
751 -np $((num_clients * $write_THREADS)) $cmd
754 if [ $rc != 0 ] ; then
755 error "write_append_truncate failed! $rc"
761 run_write_disjoint() {
762 WRITE_DISJOINT=${WRITE_DISJOINT:-$(which write_disjoint 2> /dev/null ||
764 [ x$WRITE_DISJOINT = x ] && skip_env "write_disjoint not found"
765 [ "$NFSCLIENT" ] && skip "skipped for NFSCLIENT mode"
768 wdisjoint_THREADS=${wdisjoint_THREADS:-4}
769 wdisjoint_REP=${wdisjoint_REP:-10000}
773 # Need space estimation here.
775 print_opts WRITE_DISJOINT clients wdisjoint_THREADS wdisjoint_REP \
777 local testdir=$DIR/d0.write_disjoint
779 # mpi_run uses mpiuser
782 local cmd="$WRITE_DISJOINT -f $testdir/file -n $wdisjoint_REP -m \
786 mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \
787 -np $((num_clients * $wdisjoint_THREADS)) $cmd
790 if [ $rc != 0 ] ; then
791 error "write_disjoint failed! $rc"
796 run_parallel_grouplock() {
797 PARALLEL_GROUPLOCK=${PARALLEL_GROUPLOCK:-$(which parallel_grouplock \
798 2> /dev/null || true)}
800 [ x$PARALLEL_GROUPLOCK = x ] && skip "PARALLEL_GROUPLOCK not found"
801 [ "$NFSCLIENT" ] && skip "skipped for NFSCLIENT mode"
803 parallel_grouplock_MINTASKS=${parallel_grouplock_MINTASKS:-5}
805 print_opts clients parallel_grouplock_MINTASKS MACHINEFILE
807 local testdir=$DIR/d0.parallel_grouplock
809 # mpi_run uses mpiuser
815 for i in $(seq 12); do
817 local cmd="$PARALLEL_GROUPLOCK -g -v -d $testdir $subtest"
820 mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \
821 -np $parallel_grouplock_MINTASKS $cmd
823 if [ $rc != 0 ] ; then
824 error_noexit "parallel_grouplock subtests $subtest " \
827 echo "parallel_grouplock subtests $subtest PASS"
829 let status=$((status + rc))
830 # clear debug to collect one log per one test
831 do_nodes $(comma_list $(nodes_list)) lctl clear
833 [ $status -eq 0 ] || error "parallel_grouplock status: $status"
837 cleanup_statahead () {
844 for i in $(seq 0 $num_mntpts);do
845 zconf_umount_clients $clients ${mntpt_root}$i ||
846 error_exit "Failed to umount lustre on ${mntpt_root}$i"
851 if [[ -n $NFSCLIENT ]]; then
852 skip "Statahead testing is not supported on NFS clients."
854 [ x$MDSRATE = x ] && skip_env "mdsrate not found"
856 statahead_NUMMNTPTS=${statahead_NUMMNTPTS:-5}
857 statahead_NUMFILES=${statahead_NUMFILES:-500000}
859 print_opts MDSRATE clients statahead_NUMMNTPTS statahead_NUMFILES
863 # do not use default "d[0-9]*" dir name
864 # to avoid of rm $statahead_NUMFILES (500k) files in t-f cleanup
866 local testdir=$DIR/$dir
868 # cleanup only if dir exists
869 # cleanup only $statahead_NUMFILES number of files
870 # ignore the other files created by someone else
872 mdsrate_cleanup $((num_clients * 32)) $MACHINEFILE \
873 $statahead_NUMFILES $testdir 'f%%d' --ignore
876 # mpi_run uses mpiuser
879 local num_files=$statahead_NUMFILES
881 local IFree=$(inodes_available)
882 if [ $IFree -lt $num_files ]; then
888 local cmd1="${MDSRATE} ${MDSRATE_DEBUG} --mknod --dir $testdir"
889 local cmd2="--nfiles $num_files --filefmt 'f%%d'"
890 local cmd="$cmd1 $cmd2"
893 mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \
894 -np $((num_clients * 32)) $cmd
897 if [ $rc != 0 ] ; then
898 error "mdsrate failed to create $rc"
902 local num_mntpts=$statahead_NUMMNTPTS
903 local mntpt_root=$TMP/mntpt/lustre
904 local mntopts=$MNTOPTSTATAHEAD
906 echo "Mounting $num_mntpts lustre clients starts on $clients"
907 trap "cleanup_statahead $clients $mntpt_root $num_mntpts" EXIT ERR
908 for i in $(seq 0 $num_mntpts); do
909 zconf_mount_clients $clients ${mntpt_root}$i "$mntopts" ||
910 error_exit "Failed to mount lustre on ${mntpt_root}$i on $clients"
913 do_rpc_nodes $clients cancel_lru_locks mdc
915 do_rpc_nodes $clients do_ls $mntpt_root $num_mntpts $dir
917 mdsrate_cleanup $((num_clients * 32)) $MACHINEFILE \
918 $num_files $testdir 'f%%d' --ignore
920 # use rm instead of rmdir because of
921 # testdir could contain the files created by someone else,
922 # or by previous run where is num_files prev > num_files current
924 cleanup_statahead $clients $mntpt_root $num_mntpts
927 cleanup_rr_alloc () {
930 local mntpt_root="$2"
931 local rr_alloc_MNTPTS="$3"
932 local mntpt_dir=$(dirname ${mntpt_root})
934 for i in $(seq 0 $((rr_alloc_MNTPTS - 1))); do
935 zconf_umount_clients $clients ${mntpt_root}$i ||
936 error_exit "Failed to umount lustre on ${mntpt_root}$i"
938 do_nodes $clients "rm -rf $mntpt_dir"
942 remote_mds_nodsh && skip "remote MDS with nodsh"
943 echo "===Test gives more reproduction percentage if number of "\
944 "client and ost are more. Test with 44 or more clients "\
945 "and 73 or more OSTs gives 100% reproduction rate=="
947 RR_ALLOC=${RR_ALLOC:-$(which rr_alloc 2> /dev/null || true)}
948 [ x$RR_ALLOC = x ] && skip_env "rr_alloc not found"
949 declare -a diff_max_min_arr
950 # foeo = file on each ost. calc = calculated.
953 local qos_prec_objs="${TMP}/qos_and_precreated_objects"
954 local rr_alloc_NFILES=${rr_alloc_NFILES:-555}
955 local rr_alloc_MNTPTS=${rr_alloc_MNTPTS:-11}
956 local total_MNTPTS=$((rr_alloc_MNTPTS * num_clients))
957 local mntpt_root="${TMP}/rr_alloc_mntpt/lustre"
958 if [ $MDSCOUNT -lt 2 ]; then
959 [ -e $DIR/$tdir ] || mkdir -p $DIR/$tdir
961 [ -e $DIR/$tdir ] || $LFS mkdir -i 0 $DIR/$tdir
963 chmod 0777 $DIR/$tdir
964 $SETSTRIPE -c 1 /$DIR/$tdir
966 trap "cleanup_rr_alloc $clients $mntpt_root $rr_alloc_MNTPTS" EXIT ERR
967 for i in $(seq 0 $((rr_alloc_MNTPTS - 1))); do
968 zconf_mount_clients $clients ${mntpt_root}$i $MOUNT_OPTS ||
969 error_exit "Failed to mount lustre on ${mntpt_root}$i $clients"
972 local cmd="$RR_ALLOC $mntpt_root/$tdir/ash $rr_alloc_NFILES \
975 # Save mdt values, set threshold to 100% i.e always Round Robin,
976 # restore the saved values again after creating files...
977 save_lustre_params mds1 \
978 "lov.$FSNAME-MDT0000*.qos_threshold_rr" > $qos_prec_objs
979 save_lustre_params mds1 \
980 "osp.$FSNAME-OST*-osc-MDT0000.create_count" >> $qos_prec_objs
982 local old_create_count=$(grep -e "create_count" $qos_prec_objs |
983 cut -d'=' -f 2 | sort -nr | head -n1)
985 # Make sure that every osp has enough precreated objects for the file
988 # create_count is always set to the power of 2 only, so if the files
989 # per OST are not multiple of that then it will be set to nearest
990 # lower power of 2. So set 'create_count' to the upper power of 2.
992 foeo_calc=$((rr_alloc_NFILES * total_MNTPTS / OSTCOUNT))
993 local create_count=$((2 * foeo_calc))
994 do_facet mds1 "$LCTL set_param -n \
995 lov.$FSNAME-MDT0000*.qos_threshold_rr 100 \
996 osp.$FSNAME-OST*-osc-MDT0000.create_count $create_count" ||
997 error "failed while setting qos_threshold_rr & creat_count"
999 # Create few temporary files in order to increase the precreated objects
1000 # to a desired value, before starting 'rr_alloc' app. Due to default
1001 # value 32 of precreation count (OST_MIN_PRECREATE=32), precreated
1002 # objects available are 32 initially, these gets exhausted very soon,
1003 # which causes skip of some osps when very large number of files
1004 # is created per OSTs.
1005 createmany -o $DIR/$tdir/foo- $(((old_create_count + 1) * OSTCOUNT)) \
1007 rm -f /$DIR/$tdir/foo*
1009 # Check for enough precreated objects... We should not
1010 # fail here because code(osp_precreate.c) also takes care of it.
1011 # So we have good chances of passing test even if this check fails.
1013 for ost_idx in $(seq 0 $((OSTCOUNT - 1))); do
1014 [[ $(precreated_ost_obj_count $mdt_idx $ost_idx) -ge \
1015 $foeo_calc ]] || echo "Warning: test may fail because" \
1016 "of lack of precreated objects on OST${ost_idx}"
1019 if [[ $total_MNTPTS -ne 0 ]]; then
1020 # Now start the actual file creation app.
1021 mpi_run "-np $total_MNTPTS" $cmd || return
1023 error "No mount point"
1026 restore_lustre_params < $qos_prec_objs
1027 rm -f $qos_prec_objs
1029 diff_max_min_arr=($($GETSTRIPE -r $DIR/$tdir/ |
1030 grep "lmm_stripe_offset:" | awk '{print $2}' | sort -n |
1031 uniq -c | awk 'NR==1 {min=max=$1} \
1032 { $1<min ? min=$1 : min; $1>max ? max=$1 : max} \
1033 END {print max-min, max, min}'))
1037 # In-case of fairly large number of file creation using RR (round-robin)
1038 # there can be two cases in which deviation will occur than the regular
1039 # RR algo behaviour-
1040 # 1- When rr_alloc does not start right with 'lqr_start_count' reseeded,
1041 # 2- When rr_alloc does not finish with 'lqr_start_count == 0'.
1042 # So the difference of files b/w any 2 OST should not be more than 2.
1043 [[ ${diff_max_min_arr[0]} -le 2 ]] ||
1044 error "Uneven distribution detected: difference between" \
1045 "maximum files per OST (${diff_max_min_arr[1]}) and" \
1046 "minimum files per OST (${diff_max_min_arr[2]}) must not be" \
1051 # fs_test.x is the default name for exe
1052 FS_TEST=${FS_TEST:=$(which fs_test.x 2> /dev/null || true)}
1054 local clients=${CLIENTS:-$(hostname)}
1055 local testdir=$DIR/d0.fs_test
1056 local file=${testdir}/fs_test
1057 fs_test_threads=${fs_test_threads:-2}
1058 fs_test_type=${fs_test_type:-1}
1059 fs_test_nobj=${fs_test_nobj:-10}
1060 fs_test_check=${fs_test_check:-3}
1061 fs_test_strided=${fs_test_strided:-1}
1062 fs_test_touch=${fs_test_touch:-3}
1063 fs_test_supersize=${fs_test_supersize:-1}
1064 fs_test_op=${fs_test_op:-write}
1065 fs_test_barriers=${fs_test_barriers:-bopen,bwrite,bclose}
1066 fs_test_io=${fs_test_io:-mpi}
1067 fs_test_objsize=${fs_test_objsize:-100}
1068 fs_test_objunit=${fs_test_objunit:-1048576} # 1 mb
1069 fs_test_ndirs=${fs_test_ndirs:-80000}
1071 [ x$FS_TEST = x ] && skip "FS_TEST not found"
1073 # Space estimation in bytes
1074 local space=$(df -B 1 -P $dir | tail -n 1 | awk '{ print $4 }')
1075 local total_threads=$((num_clients * fs_test_threads))
1076 echo "+ $fs_test_objsize * $fs_test_objunit * $total_threads "
1077 if [ $((space / 2)) -le \
1078 $((fs_test_objsize * fs_test_objunit * total_threads)) ]; then
1079 fs_test_objsize=$((space / 2 / fs_test_objunit / \
1081 [ $fs_test_objsize -eq 0 ] &&
1082 skip_env "Need free space more than \
1083 $((2 * total_threads * fs_test_objunit)) \
1084 : have $((space / fs_test_objunit))"
1086 echo "(reduced objsize to \
1087 $((fs_test_objsize * fs_test_objunit)) bytes)"
1090 print_opts FS_TEST clients fs_test_threads fs_test_objsize MACHINEFILE
1093 # mpi_run uses mpiuser
1096 # --nodb Turn off the database code at runtime
1097 # -g --target The path to the data file
1098 # -t --type Whether to do N-N (1) or N-1 (2)
1099 # -n --nobj The number of objects written/read by each proc
1100 # -z --size The size of each object
1101 # -d ---num_nn_dirs Number of subdirectories for files
1102 # -C --check Check every byte using argument 3.
1103 # --collective Whether to use collective I/O (for N-1, mpi-io only)
1104 # -s --strided Whether to use a strided pattern (for N-1 only)
1105 # -T --touch Touch every byte using argument 3
1106 # -o --op Whether to read only (read) or write only (write)
1107 # -b --barriers When to barrier.
1108 # -i --io Use POSIX, MPI, or PLFS IO routines (mpi|posix|plfs)
1109 # -S --supersize Specify how many objects per superblock
1111 local cmd="$FS_TEST -nodb -g $file -t $fs_test_type -n $fs_test_nobj \
1112 -z $((fs_test_objsize * fs_test_objunit)) -d $fs_test_ndirs \
1113 -C $fs_test_check -collective -s $fs_test_strided \
1114 -T $fs_test_touch -o $fs_test_op -b $fs_test_barriers \
1115 -i $fs_test_io -S $fs_test_supersize"
1118 mpi_run "-np $((num_clients * fs_test_threads))" $cmd
1121 if [ $rc != 0 ] ; then
1122 error "fs_test failed! $rc"
1128 ior_mdtest_parallel() {
1136 run_mdtest $type || rc2=$?
1137 [[ $rc2 -ne 0 ]] && echo "mdtest failed with error $rc2"
1139 wait $pids || rc1=$?
1140 [[ $rc1 -ne 0 ]] && echo "ior failed with error $rc1"
1142 [[ $rc1 -ne 0 || $rc2 -ne 0 ]] && return 1
1147 FIO=${FIO:=$(which fio 2> /dev/null || true)}
1149 local clients=${CLIENTS:-$(hostname)}
1150 local fio_jobNum=${fio_jobNum:-4}
1151 local fio_jobFile=${fio_jobFile:-$TMP/fiojobfile.$(date +%s)}
1152 local fio_bs=${fio_bs:-1}
1153 local testdir=$DIR/d0.fio
1154 local file=${testdir}/fio
1156 local propagate=false
1158 [ "$SLOW" = "no" ] || runtime=600
1160 [ x$FIO = x ] && skip_env "FIO not found"
1164 # use fio job file if exists,
1165 # create a simple one if missing
1166 if ! [ -f $fio_jobFile ]; then
1167 cat >> $fio_jobFile <<EOF
1173 filename=${file}_\$(hostname)
1175 # bs size increased by $i for each job
1176 for ((i=1; i<=fio_jobNum; i++)); do
1177 cat >> $fio_jobFile <<EOF
1180 bs=$(( fio_bs * i ))m
1183 # job file is created, should be propagated to all clients
1188 # propagate the job file if not all clients have it yet or
1189 # if the job file was created during the test run
1190 if ! do_nodesv $clients " [ -f $fio_jobFile ] " ||
1192 local cfg=$(cat $fio_jobFile)
1193 do_nodes $clients "echo \\\"$cfg\\\" > ${fio_jobFile}" ||
1194 error "job file $fio_jobFile is not propagated"
1195 do_nodesv $clients "cat ${fio_jobFile}"
1198 cmd="$FIO $fio_jobFile"
1201 log "clients: $clients $cmd"
1204 do_nodesv $clients "$cmd "
1207 [ $rc = 0 ] || error "fio failed: $rc"
1212 XDD=${XDD:=$(which xdd 2> /dev/null || true)}
1214 local clients=${CLIENTS:-$(hostname)}
1215 local testdir=$DIR/d0.xdd
1216 xdd_queuedepth=${xdd_queuedepth:-4}
1217 xdd_blocksize=${xdd_blocksize:-512}
1218 xdd_reqsize=${xdd_reqsize:-128}
1219 xdd_mbytes=${xdd_mbytes:-100}
1220 xdd_passes=${xdd_passes:-40}
1221 xdd_rwratio=${xdd_rwratio:-0}
1222 xdd_ntargets=${xdd_ntargets:-6}
1223 local xdd_custom_params=${xdd_custom_params:-"-dio -stoponerror \
1224 -maxpri -minall -noproclock -nomemlock"}
1226 [ x$XDD = x ] && skip "XDD not found"
1228 print_opts XDD clients xdd_queuedepth xdd_blocksize xdd_reqsize \
1229 xdd_mbytes xdd_passes xdd_rwratio
1234 # Target files creates based on the given number of targets
1235 for (( i=0; i < $xdd_ntargets; i++ ))
1237 files+="${testdir}/xdd"$i" "
1240 # -targets specifies the devices or files to perform operation
1241 # -reqsize number of 'blocks' per operation
1242 # -mbytes number of 1024*1024-byte blocks to transfer
1243 # -blocksize size of a single 'block'
1244 # -passes number of times to read mbytes
1245 # -queuedepth number of commands to queue on the target
1246 # -rwratio percentage of read to write operations
1247 # -verbose will print out statistics on each pass
1249 local cmd="$XDD -targets $xdd_ntargets $files -reqsize $xdd_reqsize \
1250 -mbytes $xdd_mbytes -blocksize $xdd_blocksize \
1251 -passes $xdd_passes -queuedepth $xdd_queuedepth \
1252 -rwratio $xdd_rwratio -verbose $xdd_custom_params"
1256 do_nodesv $clients "$cmd "
1259 [ $rc = 0 ] || error "xdd failed: $rc"