3 # Simple function used by run_*.sh scripts
8 if [ -z "${!name}" ]; then
9 echo "$0: $name must be set"
13 [ $failed ] && exit 1 || true
16 # lrepl - Lustre test Read-Eval-Print Loop.
18 # This function implements a REPL for the Lustre test framework. It
19 # doesn't exec an actual shell because the user may want to inspect
20 # variables and use functions from the test framework.
27 This is an interactive read-eval-print loop interactive shell
28 simulation that you can use to debug failing tests. You can
29 enter most bash command lines (see notes below).
31 Use this REPL to inspect variables, set them, call test
32 framework shell functions, etcetera.
34 'exit' or EOF to exit this shell.
36 set \$retcode to 0 to cause the assertion failure that
37 triggered this REPL to be ignored.
40 do_facet ost1 lctl get_param ost.*.ost.threads_*
41 do_rpc_nodes \$OSTNODES unload_modules
44 All but the last line of multi-line statements or blocks
45 must end in a backslash.
47 "Here documents" are not supported.
49 History is not supported, but command-line editing is.
53 # Prompt escapes don't work in read -p, sadly.
54 prompt=":${TESTNAME:-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 $MACHINEFILE
286 [ "$(stat -f -c %T $1)" = "lustre" ]
289 setstripe_getstripe () {
294 is_lustre $file || return 0
296 if [ -n "$params" ]; then
297 $LFS setstripe $params $file ||
298 error "setstripe $params failed"
300 $LFS getstripe $file ||
301 error "getstripe $file failed"
306 local cbench_DIR=${cbench_DIR:-""}
307 local cbench_IDIRS=${cbench_IDIRS:-2}
308 local cbench_RUNS=${cbench_RUNS:-2}
310 print_opts cbench_DIR cbench_IDIRS cbench_RUNS
312 [ x$cbench_DIR = x ] &&
313 skip_env "compilebench not found"
315 [ -e $cbench_DIR/compilebench ] ||
316 skip_env "No compilebench build"
319 # compile dir kernel-0 ~1GB
320 # required space ~1GB * cbench_IDIRS
321 local space=$(df -P $dir | tail -n 1 | awk '{ print $4 }')
322 if [[ $space -le $((1024 * 1024 * cbench_IDIRS)) ]]; then
323 cbench_IDIRS=$((space / 1024 / 1024))
324 [[ $cbench_IDIRS -eq 0 ]] &&
325 skip_env "Need free space at least 1GB, have $space"
327 echo "reducing initial dirs to $cbench_IDIRS"
329 echo "free space = $space KB"
332 # t-f _base needs to be modifyed to set properly tdir
333 # for new "test_foo" functions names
334 # local testdir=$DIR/$tdir
335 local testdir=$dir/d0.compilebench.$$
337 setstripe_getstripe $testdir $cbench_STRIPEPARAMS
341 local cmd="./compilebench -D $testdir -i $cbench_IDIRS \
342 -r $cbench_RUNS --makej"
351 [ $rc = 0 ] || error "compilebench failed: $rc"
357 local mntpt=${2:-$MOUNT}
358 METABENCH=${METABENCH:-$(which metabench 2> /dev/null || true)}
359 mbench_NFILES=${mbench_NFILES:-30400}
361 mbench_THREADS=${mbench_THREADS:-4}
362 mbench_OPTIONS=${mbench_OPTIONS:-}
363 mbench_CLEANUP=${mbench_CLEANUP:-true}
365 [ x$METABENCH = x ] && skip_env "metabench not found"
367 print_opts METABENCH clients mbench_NFILES mbench_THREADS
369 local testdir=$dir/d0.metabench
371 setstripe_getstripe $testdir $mbench_STRIPEPARAMS
373 # mpi_run uses mpiuser
376 # -C Run the file creation tests. Creates zero byte files.
377 # -S Run the file stat tests.
378 # -c nfile Number of files to be used in each test.
379 # -k Cleanup files when finished.
380 local cmd="$METABENCH -w $testdir -c $mbench_NFILES -C -S $mbench_OPTIONS"
383 # find out if we need to use srun by checking $SRUN_PARTITION
384 if [ "$SRUN_PARTITION" ]; then
385 $SRUN $SRUN_OPTIONS -D $testdir -w $clients -N $num_clients \
386 -n $((num_clients * mbench_THREADS)) \
387 -p $SRUN_PARTITION -- $cmd
389 mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \
390 -np $((num_clients * $mbench_THREADS)) $cmd
394 if [ $rc != 0 ] ; then
395 error "metabench failed! $rc"
398 if $mbench_CLEANUP; then
401 mv $dir/d0.metabench $mntpt/_xxx.$(date +%s).d0.metabench
406 SIMUL=${SIMUL:=$(which simul 2> /dev/null || true)}
407 [ x$SIMUL = x ] && skip_env "simul not found"
408 [ "$NFSCLIENT" ] && skip "skipped for NFSCLIENT mode"
411 simul_THREADS=${simul_THREADS:-2}
412 simul_REP=${simul_REP:-20}
415 # Need space estimation here.
417 print_opts SIMUL clients simul_REP simul_THREADS
419 local testdir=$DIR/d0.simul
421 setstripe_getstripe $testdir $simul_STRIPEPARAMS
423 # mpi_run uses mpiuser
426 # -n # : repeat each test # times
427 # -N # : repeat the entire set of tests # times
429 local cmd="$SIMUL -d $testdir -n $simul_REP -N $simul_REP"
432 # find out if we need to use srun by checking $SRUN_PARTITION
433 if [ "$SRUN_PARTITION" ]; then
434 $SRUN $SRUN_OPTIONS -D $testdir -w $clients -N $num_clients \
435 -n $((num_clients * simul_THREADS)) -p $SRUN_PARTITION \
438 mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \
439 -np $((num_clients * simul_THREADS)) $cmd
443 if [ $rc != 0 ] ; then
444 error "simul failed! $rc"
450 MDTEST=${MDTEST:=$(which mdtest 2> /dev/null || true)}
451 [ x$MDTEST = x ] && skip_env "mdtest not found"
452 [ "$NFSCLIENT" ] && skip "skipped for NFSCLIENT mode"
455 mdtest_THREADS=${mdtest_THREADS:-2}
456 mdtest_nFiles=${mdtest_nFiles:-"100000"}
457 # We devide the files by number of core
458 mdtest_nFiles=$((mdtest_nFiles/mdtest_THREADS/num_clients))
459 mdtest_iteration=${mdtest_iteration:-1}
460 local mdtest_custom_params=${mdtest_custom_params:-""}
461 local type=${1:-"ssf"}
464 # Need space estimation here.
466 print_opts MDTEST mdtest_iteration mdtest_THREADS mdtest_nFiles
468 local testdir=$DIR/d0.mdtest
470 setstripe_getstripe $testdir $mdtest_STRIPEPARAMS
472 # mpi_run uses mpiuser
475 # -i # : repeat each test # times
477 # -n # : number of file/dir to create/stat/remove
478 # -u : each process create/stat/remove individually
480 local cmd="$MDTEST -d $testdir -i $mdtest_iteration \
481 -n $mdtest_nFiles $mdtest_custom_params"
483 [ $type = "fpp" ] && cmd="$cmd -u"
486 # find out if we need to use srun by checking $SRUN_PARTITION
487 if [ "$SRUN_PARTITION" ]; then
488 $SRUN $SRUN_OPTIONS -D $testdir -w $clients -N $num_clients \
489 -n $((num_clients * mdtest_THREADS)) \
490 -p $SRUN_PARTITION -- $cmd
492 mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \
493 -np $((num_clients * mdtest_THREADS)) $cmd
497 if [ $rc != 0 ] ; then
498 error "mdtest failed! $rc"
505 cnt_DIR=${cnt_DIR:-""}
506 cnt_NRUN=${cnt_NRUN:-10}
508 print_opts cnt_DIR cnt_NRUN
510 [ x$cnt_DIR = x ] && skip_env "connectathon dir not found"
511 [ -e $cnt_DIR/runtests ] || skip_env "No connectathon runtests found"
514 # "special" tests create a 30 MB file + misc. small files
515 # required space ~40 MB
516 local space=$(df -P $dir | tail -n 1 | awk '{ print $4 }')
517 if [[ $space -le $((1024 * 40)) ]]; then
518 skip_env "Need free space at least 40MB, have $space KB"
520 echo "free space = $space KB"
522 local testdir=$dir/d0.connectathon
524 setstripe_getstripe $testdir $cnt_STRIPEPARAMS
530 # To run connectathon:
531 # runtests [-a|-b|-g|-s|-l] [-f|-n|-t] [-N numpasses] [test-directory]
533 # One of the following test types
538 # -a all of the above
540 # -f a quick functional test
541 # -n suppress directory operations (mkdir and rmdir)
542 # -t run with time statistics (default for basic tests)
544 # -N numpasses - specifies the number of times to run
545 # the tests. Optional.
548 # Include lock tests unless we're running on nfsv4
549 local fstype=$(df -TP $testdir | awk 'NR==2 {print $2}')
550 echo "$testdir: $fstype"
551 if [[ $fstype != "nfs4" ]]; then
555 for test in $tests; do
556 local cmd="sh ./runtests -N $cnt_NRUN $test -f $testdir"
562 [ $rc = 0 ] || error "connectathon failed: $rc"
570 local type=${1:="ssf"}
572 local testdir=$dir/d0.ior.$type
573 local nfs_srvmntpt=$3
575 if [ "$NFSCLIENT" ]; then
576 [[ -n $nfs_srvmntpt ]] ||
577 { error "NFSCLIENT mode, but nfs exported dir"\
578 "is not set!" && return 1; }
581 IOR=${IOR:-$(which IOR 2> /dev/null || true)}
582 [ x$IOR = x ] && skip_env "IOR not found"
585 ior_THREADS=${ior_THREADS:-2}
586 ior_iteration=${ior_iteration:-1}
587 ior_blockSize=${ior_blockSize:-6}
588 ior_blockUnit=${ior_blockUnit:-M} # K, M, G
589 ior_xferSize=${ior_xferSize:-1M}
590 ior_type=${ior_type:-POSIX}
591 ior_DURATION=${ior_DURATION:-30} # minutes
593 case ${ior_blockUnit} in
595 multiplier=$((1024 * 1024 * 1024))
598 multiplier=$((1024 * 1024))
603 *) error "Incorrect block unit should be one of [KMG]"
607 # calculate the space in bytes
608 local space=$(df -B 1 -P $dir | tail -n 1 | awk '{ print $4 }')
609 local total_threads=$((num_clients * ior_THREADS))
610 echo "+ $ior_blockSize * $multiplier * $total_threads "
611 if [ $((space / 2)) -le \
612 $((ior_blockSize * multiplier * total_threads)) ]; then
613 ior_blockSize=$((space / 2 / multiplier / total_threads))
614 [ $ior_blockSize -eq 0 ] &&
615 skip_env "Need free space more than $((2 * total_threads)) \
616 ${ior_blockUnit}: have $((space / multiplier))"
618 echo "(reduced blockSize to $ior_blockSize \
619 ${ior_blockUnit} bytes)"
622 print_opts IOR ior_THREADS ior_DURATION MACHINEFILE
626 # mpi_run uses mpiuser
628 [[ "$ior_stripe_params" && -z "$ior_STRIPEPARAMS" ]] &&
629 ior_STRIPEPARAMS="$ior_stripe_params" &&
630 echo "got deprecated ior_stripe_params,"\
631 "use ior_STRIPEPARAMS instead"
632 setstripe_getstripe $testdir $ior_STRIPEPARAMS
636 # contiguous bytes to write per task (e.g.: 8, 4K, 2M, 1G)"
638 # -t N transferSize -- size of transfer in bytes (e.g.: 8, 4K, 2M, 1G)"
639 # -w writeFile -- write file"
640 # -r readFile -- read existing file"
641 # -W checkWrite -- check read after write"
642 # -C reorderTasks -- changes task ordering to n+1 ordering for readback
643 # -T maxTimeDuration -- max time in minutes to run tests"
644 # -k keepFile -- keep testFile(s) on program exit
647 if [ -n "$ior_custom_params" ]; then
648 cmd="$IOR -o $testdir/iorData $ior_custom_params"
650 cmd="$IOR -a $ior_type -b ${ior_blockSize}${ior_blockUnit} \
651 -o $testdir/iorData -t $ior_xferSize -v -C -w -r -W \
652 -i $ior_iteration -T $ior_DURATION -k"
655 [ $type = "fpp" ] && cmd="$cmd -F"
658 # find out if we need to use srun by checking $SRUN_PARTITION
659 if [ "$SRUN_PARTITION" ]; then
660 $SRUN $SRUN_OPTIONS -D $testdir -w $clients -N $num_clients \
661 -n $((num_clients * ior_THREADS)) -p $SRUN_PARTITION \
664 mpi_ior_custom_threads=${mpi_ior_custom_threads:-"$((num_clients * ior_THREADS))"}
665 mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \
666 -np $mpi_ior_custom_threads $cmd
670 if [ $rc != 0 ] ; then
671 error "ior failed! $rc"
677 MIB=${MIB:=$(which mib 2> /dev/null || true)}
678 [ "$NFSCLIENT" ] && skip "skipped for NFSCLIENT mode"
679 [ x$MIB = x ] && skip_env "MIB not found"
682 mib_THREADS=${mib_THREADS:-2}
683 mib_xferSize=${mib_xferSize:-1m}
684 mib_xferLimit=${mib_xferLimit:-5000}
685 mib_timeLimit=${mib_timeLimit:-300}
686 mib_STRIPEPARAMS=${mib_STRIPEPARAMS:-"-c -1"}
688 print_opts MIB mib_THREADS mib_xferSize mib_xferLimit mib_timeLimit \
691 local testdir=$DIR/d0.mib
693 setstripe_getstripe $testdir $mib_STRIPEPARAMS
695 # mpi_run uses mpiuser
699 # -I Show intermediate values in output
700 # -H Show headers in output
701 # -L Do not issue new system calls after this many seconds
702 # -s Use system calls of this size
704 # -l Issue no more than this many system calls
705 local cmd="$MIB -t $testdir -s $mib_xferSize -l $mib_xferLimit \
706 -L $mib_timeLimit -HI -p mib.$(date +%Y%m%d%H%M%S)"
709 # find out if we need to use srun by checking $SRUN_PARTITION
710 if [ "$SRUN_PARTITION" ]; then
711 $SRUN $SRUN_OPTIONS -D $testdir -w $clients -N $num_clients \
712 -n $((num_clients * mib_THREADS)) -p $SRUN_PARTITION \
715 mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \
716 -np $((num_clients * mib_THREADS)) $cmd
720 if [ $rc != 0 ] ; then
721 error "mib failed! $rc"
727 CASC_RW=${CASC_RW:-$(which cascading_rw 2> /dev/null || true)}
728 [ x$CASC_RW = x ] && skip_env "cascading_rw not found"
729 [ "$NFSCLIENT" ] && skip "skipped for NFSCLIENT mode"
732 casc_THREADS=${casc_THREADS:-2}
733 casc_REP=${casc_REP:-300}
736 # Need space estimation here.
738 print_opts CASC_RW clients casc_THREADS casc_REP MACHINEFILE
740 local testdir=$DIR/d0.cascading_rw
742 setstripe_getstripe $testdir $casc_STRIPEPARAMS
744 # mpi_run uses mpiuser
748 # -n: repeat test # times
750 local cmd="$CASC_RW -g -d $testdir -n $casc_REP"
753 mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \
754 -np $((num_clients * $casc_THREADS)) $cmd
757 if [ $rc != 0 ] ; then
758 error "cascading_rw failed! $rc"
763 run_write_append_truncate() {
764 [ "$NFSCLIENT" ] && skip "skipped for NFSCLIENT mode"
765 # location is lustre/tests dir
766 if ! which write_append_truncate > /dev/null 2>&1 ; then
767 skip_env "write_append_truncate not found"
771 write_THREADS=${write_THREADS:-8}
772 write_REP=${write_REP:-10000}
775 # Need space estimation here.
777 local testdir=$DIR/d0.write_append_truncate
778 local file=$testdir/f0.wat
780 print_opts clients write_REP write_THREADS MACHINEFILE
783 # mpi_run uses mpiuser
784 setstripe_getstripe $testdir $write_STRIPEPARAMS
788 local cmd="write_append_truncate -n $write_REP $file"
791 mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \
792 -np $((num_clients * $write_THREADS)) $cmd
795 if [ $rc != 0 ] ; then
796 error "write_append_truncate failed! $rc"
802 run_write_disjoint() {
803 WRITE_DISJOINT=${WRITE_DISJOINT:-$(which write_disjoint 2> /dev/null ||
805 [ x$WRITE_DISJOINT = x ] && skip_env "write_disjoint not found"
806 [ "$NFSCLIENT" ] && skip "skipped for NFSCLIENT mode"
809 wdisjoint_THREADS=${wdisjoint_THREADS:-4}
810 wdisjoint_REP=${wdisjoint_REP:-10000}
814 # Need space estimation here.
816 print_opts WRITE_DISJOINT clients wdisjoint_THREADS wdisjoint_REP \
818 local testdir=$DIR/d0.write_disjoint
820 setstripe_getstripe $testdir $wdisjoint_STRIPEPARAMS
822 # mpi_run uses mpiuser
825 local cmd="$WRITE_DISJOINT -f $testdir/file -n $wdisjoint_REP -m \
829 mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \
830 -np $((num_clients * $wdisjoint_THREADS)) $cmd
833 if [ $rc != 0 ] ; then
834 error "write_disjoint failed! $rc"
839 run_parallel_grouplock() {
840 PARALLEL_GROUPLOCK=${PARALLEL_GROUPLOCK:-$(which parallel_grouplock \
841 2> /dev/null || true)}
843 [ x$PARALLEL_GROUPLOCK = x ] && skip "PARALLEL_GROUPLOCK not found"
844 [ "$NFSCLIENT" ] && skip "skipped for NFSCLIENT mode"
846 parallel_grouplock_MINTASKS=${parallel_grouplock_MINTASKS:-5}
848 print_opts clients parallel_grouplock_MINTASKS MACHINEFILE
850 local testdir=$DIR/d0.parallel_grouplock
852 setstripe_getstripe $testdir $parallel_grouplock_STRIPEPARAMS
854 # mpi_run uses mpiuser
860 for i in $(seq 12); do
862 local cmd="$PARALLEL_GROUPLOCK -g -v -d $testdir $subtest"
865 mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \
866 -np $parallel_grouplock_MINTASKS $cmd
868 if [ $rc != 0 ] ; then
869 error_noexit "parallel_grouplock subtests $subtest " \
872 echo "parallel_grouplock subtests $subtest PASS"
874 let status=$((status + rc))
875 # clear debug to collect one log per one test
876 do_nodes $(comma_list $(nodes_list)) lctl clear
878 [ $status -eq 0 ] || error "parallel_grouplock status: $status"
882 cleanup_statahead () {
889 for i in $(seq 0 $num_mntpts);do
890 zconf_umount_clients $clients ${mntpt_root}$i ||
891 error_exit "Failed to umount lustre on ${mntpt_root}$i"
896 if [[ -n $NFSCLIENT ]]; then
897 skip "Statahead testing is not supported on NFS clients."
899 [ x$MDSRATE = x ] && skip_env "mdsrate not found"
901 statahead_NUMMNTPTS=${statahead_NUMMNTPTS:-5}
902 statahead_NUMFILES=${statahead_NUMFILES:-500000}
904 print_opts MDSRATE clients statahead_NUMMNTPTS statahead_NUMFILES
908 # do not use default "d[0-9]*" dir name
909 # to avoid of rm $statahead_NUMFILES (500k) files in t-f cleanup
911 local testdir=$DIR/$dir
913 # cleanup only if dir exists
914 # cleanup only $statahead_NUMFILES number of files
915 # ignore the other files created by someone else
917 mdsrate_cleanup $((num_clients * 32)) $MACHINEFILE \
918 $statahead_NUMFILES $testdir 'f%%d' --ignore
921 setstripe_getstripe $testdir $statahead_STRIPEPARAMS
923 # mpi_run uses mpiuser
926 local num_files=$statahead_NUMFILES
928 local IFree=$(inodes_available)
929 if [ $IFree -lt $num_files ]; then
935 local cmd1="${MDSRATE} ${MDSRATE_DEBUG} --mknod --dir $testdir"
936 local cmd2="--nfiles $num_files --filefmt 'f%%d'"
937 local cmd="$cmd1 $cmd2"
940 mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \
941 -np $((num_clients * 32)) $cmd
944 if [ $rc != 0 ] ; then
945 error "mdsrate failed to create $rc"
949 local num_mntpts=$statahead_NUMMNTPTS
950 local mntpt_root=$TMP/mntpt/lustre
951 local mntopts=$MNTOPTSTATAHEAD
953 echo "Mounting $num_mntpts lustre clients starts on $clients"
954 trap "cleanup_statahead $clients $mntpt_root $num_mntpts" EXIT ERR
955 for i in $(seq 0 $num_mntpts); do
956 zconf_mount_clients $clients ${mntpt_root}$i "$mntopts" ||
957 error_exit "Failed to mount lustre on ${mntpt_root}$i on $clients"
960 do_rpc_nodes $clients cancel_lru_locks mdc
962 do_rpc_nodes $clients do_ls $mntpt_root $num_mntpts $dir
964 mdsrate_cleanup $((num_clients * 32)) $MACHINEFILE \
965 $num_files $testdir 'f%%d' --ignore
967 # use rm instead of rmdir because of
968 # testdir could contain the files created by someone else,
969 # or by previous run where is num_files prev > num_files current
971 cleanup_statahead $clients $mntpt_root $num_mntpts
974 cleanup_rr_alloc () {
977 local mntpt_root="$2"
978 local rr_alloc_MNTPTS="$3"
979 local mntpt_dir=$(dirname ${mntpt_root})
981 for i in $(seq 0 $((rr_alloc_MNTPTS - 1))); do
982 zconf_umount_clients $clients ${mntpt_root}$i ||
983 error_exit "Failed to umount lustre on ${mntpt_root}$i"
985 do_nodes $clients "rm -rf $mntpt_dir"
989 remote_mds_nodsh && skip "remote MDS with nodsh"
990 echo "===Test gives more reproduction percentage if number of "\
991 "client and ost are more. Test with 44 or more clients "\
992 "and 73 or more OSTs gives 100% reproduction rate=="
994 RR_ALLOC=${RR_ALLOC:-$(which rr_alloc 2> /dev/null || true)}
995 [ x$RR_ALLOC = x ] && skip_env "rr_alloc not found"
996 declare -a diff_max_min_arr
997 # foeo = file on each ost. calc = calculated.
1000 local qos_prec_objs="${TMP}/qos_and_precreated_objects"
1001 local rr_alloc_NFILES=${rr_alloc_NFILES:-555}
1002 local rr_alloc_MNTPTS=${rr_alloc_MNTPTS:-11}
1003 local total_MNTPTS=$((rr_alloc_MNTPTS * num_clients))
1004 local mntpt_root="${TMP}/rr_alloc_mntpt/lustre"
1005 if [ $MDSCOUNT -lt 2 ]; then
1006 [ -e $DIR/$tdir ] || mkdir -p $DIR/$tdir
1008 [ -e $DIR/$tdir ] || $LFS mkdir -i 0 $DIR/$tdir
1010 setstripe_getstripe $DIR/$tdir $rr_alloc_STRIPEPARAMS
1012 chmod 0777 $DIR/$tdir
1014 trap "cleanup_rr_alloc $clients $mntpt_root $rr_alloc_MNTPTS" EXIT ERR
1015 for i in $(seq 0 $((rr_alloc_MNTPTS - 1))); do
1016 zconf_mount_clients $clients ${mntpt_root}$i $MOUNT_OPTS ||
1017 error_exit "Failed to mount lustre on ${mntpt_root}$i $clients"
1020 local cmd="$RR_ALLOC $mntpt_root/$tdir/ash $rr_alloc_NFILES \
1023 # Save mdt values, set threshold to 100% i.e always Round Robin,
1024 # restore the saved values again after creating files...
1025 save_lustre_params mds1 \
1026 "lov.$FSNAME-MDT0000*.qos_threshold_rr" > $qos_prec_objs
1027 save_lustre_params mds1 \
1028 "osp.$FSNAME-OST*-osc-MDT0000.create_count" >> $qos_prec_objs
1030 local old_create_count=$(grep -e "create_count" $qos_prec_objs |
1031 cut -d'=' -f 2 | sort -nr | head -n1)
1033 # Make sure that every osp has enough precreated objects for the file
1036 # create_count is always set to the power of 2 only, so if the files
1037 # per OST are not multiple of that then it will be set to nearest
1038 # lower power of 2. So set 'create_count' to the upper power of 2.
1040 foeo_calc=$((rr_alloc_NFILES * total_MNTPTS / OSTCOUNT))
1041 local create_count=$((2 * foeo_calc))
1042 do_facet mds1 "$LCTL set_param -n \
1043 lov.$FSNAME-MDT0000*.qos_threshold_rr 100 \
1044 osp.$FSNAME-OST*-osc-MDT0000.create_count $create_count" ||
1045 error "failed while setting qos_threshold_rr & creat_count"
1047 # Create few temporary files in order to increase the precreated objects
1048 # to a desired value, before starting 'rr_alloc' app. Due to default
1049 # value 32 of precreation count (OST_MIN_PRECREATE=32), precreated
1050 # objects available are 32 initially, these gets exhausted very soon,
1051 # which causes skip of some osps when very large number of files
1052 # is created per OSTs.
1053 createmany -o $DIR/$tdir/foo- $(((old_create_count + 1) * OSTCOUNT)) \
1055 rm -f /$DIR/$tdir/foo*
1057 # Check for enough precreated objects... We should not
1058 # fail here because code(osp_precreate.c) also takes care of it.
1059 # So we have good chances of passing test even if this check fails.
1061 for ost_idx in $(seq 0 $((OSTCOUNT - 1))); do
1062 [[ $(precreated_ost_obj_count $mdt_idx $ost_idx) -ge \
1063 $foeo_calc ]] || echo "Warning: test may fail because" \
1064 "of lack of precreated objects on OST${ost_idx}"
1067 if [[ $total_MNTPTS -ne 0 ]]; then
1068 # Now start the actual file creation app.
1069 mpi_run "-np $total_MNTPTS" $cmd || return
1071 error "No mount point"
1074 restore_lustre_params < $qos_prec_objs
1075 rm -f $qos_prec_objs
1077 diff_max_min_arr=($($LFS getstripe -r $DIR/$tdir/ |
1078 grep "lmm_stripe_offset:" | awk '{print $2}' | sort -n |
1079 uniq -c | awk 'NR==1 {min=max=$1} \
1080 { $1<min ? min=$1 : min; $1>max ? max=$1 : max} \
1081 END {print max-min, max, min}'))
1085 # In-case of fairly large number of file creation using RR (round-robin)
1086 # there can be two cases in which deviation will occur than the regular
1087 # RR algo behaviour-
1088 # 1- When rr_alloc does not start right with 'lqr_start_count' reseeded,
1089 # 2- When rr_alloc does not finish with 'lqr_start_count == 0'.
1090 # So the difference of files b/w any 2 OST should not be more than 2.
1091 [[ ${diff_max_min_arr[0]} -le 2 ]] ||
1092 error "Uneven distribution detected: difference between" \
1093 "maximum files per OST (${diff_max_min_arr[1]}) and" \
1094 "minimum files per OST (${diff_max_min_arr[2]}) must not be" \
1099 # fs_test.x is the default name for exe
1100 FS_TEST=${FS_TEST:=$(which fs_test.x 2> /dev/null || true)}
1102 local clients=${CLIENTS:-$(hostname)}
1103 local testdir=$DIR/d0.fs_test
1104 local file=${testdir}/fs_test
1105 fs_test_threads=${fs_test_threads:-2}
1106 fs_test_type=${fs_test_type:-1}
1107 fs_test_nobj=${fs_test_nobj:-10}
1108 fs_test_check=${fs_test_check:-3}
1109 fs_test_strided=${fs_test_strided:-1}
1110 fs_test_touch=${fs_test_touch:-3}
1111 fs_test_supersize=${fs_test_supersize:-1}
1112 fs_test_op=${fs_test_op:-write}
1113 fs_test_barriers=${fs_test_barriers:-bopen,bwrite,bclose}
1114 fs_test_io=${fs_test_io:-mpi}
1115 fs_test_objsize=${fs_test_objsize:-100}
1116 fs_test_objunit=${fs_test_objunit:-1048576} # 1 mb
1117 fs_test_ndirs=${fs_test_ndirs:-80000}
1119 [ x$FS_TEST = x ] && skip "FS_TEST not found"
1121 # Space estimation in bytes
1122 local space=$(df -B 1 -P $dir | tail -n 1 | awk '{ print $4 }')
1123 local total_threads=$((num_clients * fs_test_threads))
1124 echo "+ $fs_test_objsize * $fs_test_objunit * $total_threads "
1125 if [ $((space / 2)) -le \
1126 $((fs_test_objsize * fs_test_objunit * total_threads)) ]; then
1127 fs_test_objsize=$((space / 2 / fs_test_objunit / \
1129 [ $fs_test_objsize -eq 0 ] &&
1130 skip_env "Need free space more than \
1131 $((2 * total_threads * fs_test_objunit)) \
1132 : have $((space / fs_test_objunit))"
1134 echo "(reduced objsize to \
1135 $((fs_test_objsize * fs_test_objunit)) bytes)"
1138 print_opts FS_TEST clients fs_test_threads fs_test_objsize MACHINEFILE
1141 setstripe_getstripe $testdir $fs_test_STRIPEPARAMS
1143 # mpi_run uses mpiuser
1146 # --nodb Turn off the database code at runtime
1147 # -g --target The path to the data file
1148 # -t --type Whether to do N-N (1) or N-1 (2)
1149 # -n --nobj The number of objects written/read by each proc
1150 # -z --size The size of each object
1151 # -d ---num_nn_dirs Number of subdirectories for files
1152 # -C --check Check every byte using argument 3.
1153 # --collective Whether to use collective I/O (for N-1, mpi-io only)
1154 # -s --strided Whether to use a strided pattern (for N-1 only)
1155 # -T --touch Touch every byte using argument 3
1156 # -o --op Whether to read only (read) or write only (write)
1157 # -b --barriers When to barrier.
1158 # -i --io Use POSIX, MPI, or PLFS IO routines (mpi|posix|plfs)
1159 # -S --supersize Specify how many objects per superblock
1161 local cmd="$FS_TEST -nodb -g $file -t $fs_test_type -n $fs_test_nobj \
1162 -z $((fs_test_objsize * fs_test_objunit)) -d $fs_test_ndirs \
1163 -C $fs_test_check -collective -s $fs_test_strided \
1164 -T $fs_test_touch -o $fs_test_op -b $fs_test_barriers \
1165 -i $fs_test_io -S $fs_test_supersize"
1168 mpi_run "-np $((num_clients * fs_test_threads))" $cmd
1171 if [ $rc != 0 ] ; then
1172 error "fs_test failed! $rc"
1178 ior_mdtest_parallel() {
1186 run_mdtest $type || rc2=$?
1187 [[ $rc2 -ne 0 ]] && echo "mdtest failed with error $rc2"
1189 wait $pids || rc1=$?
1190 [[ $rc1 -ne 0 ]] && echo "ior failed with error $rc1"
1192 [[ $rc1 -ne 0 || $rc2 -ne 0 ]] && return 1
1197 FIO=${FIO:=$(which fio 2> /dev/null || true)}
1199 local clients=${CLIENTS:-$(hostname)}
1200 local fio_jobNum=${fio_jobNum:-4}
1201 local fio_jobFile=${fio_jobFile:-$TMP/fiojobfile.$(date +%s)}
1202 local fio_bs=${fio_bs:-1}
1203 local testdir=$DIR/d0.fio
1204 local file=${testdir}/fio
1206 local propagate=false
1208 [ "$SLOW" = "no" ] || runtime=600
1210 [ x$FIO = x ] && skip_env "FIO not found"
1213 setstripe_getstripe $testdir $fio_STRIPEPARAMS
1215 # use fio job file if exists,
1216 # create a simple one if missing
1217 if ! [ -f $fio_jobFile ]; then
1218 cat >> $fio_jobFile <<EOF
1224 filename=${file}_\$(hostname)
1226 # bs size increased by $i for each job
1227 for ((i=1; i<=fio_jobNum; i++)); do
1228 cat >> $fio_jobFile <<EOF
1231 bs=$(( fio_bs * i ))m
1234 # job file is created, should be propagated to all clients
1239 # propagate the job file if not all clients have it yet or
1240 # if the job file was created during the test run
1241 if ! do_nodesv $clients " [ -f $fio_jobFile ] " ||
1243 local cfg=$(cat $fio_jobFile)
1244 do_nodes $clients "echo \\\"$cfg\\\" > ${fio_jobFile}" ||
1245 error "job file $fio_jobFile is not propagated"
1246 do_nodesv $clients "cat ${fio_jobFile}"
1249 cmd="$FIO $fio_jobFile"
1252 log "clients: $clients $cmd"
1255 do_nodesv $clients "$cmd "
1258 [ $rc = 0 ] || error "fio failed: $rc"
1263 XDD=${XDD:=$(which xdd 2> /dev/null || true)}
1265 local clients=${CLIENTS:-$(hostname)}
1266 local testdir=$DIR/d0.xdd
1267 xdd_queuedepth=${xdd_queuedepth:-4}
1268 xdd_blocksize=${xdd_blocksize:-512}
1269 xdd_reqsize=${xdd_reqsize:-128}
1270 xdd_mbytes=${xdd_mbytes:-100}
1271 xdd_passes=${xdd_passes:-40}
1272 xdd_rwratio=${xdd_rwratio:-0}
1273 xdd_ntargets=${xdd_ntargets:-6}
1274 local xdd_custom_params=${xdd_custom_params:-"-dio -stoponerror \
1275 -maxpri -minall -noproclock -nomemlock"}
1277 [ x$XDD = x ] && skip "XDD not found"
1279 print_opts XDD clients xdd_queuedepth xdd_blocksize xdd_reqsize \
1280 xdd_mbytes xdd_passes xdd_rwratio
1283 setstripe_getstripe $testdir $xdd_STRIPEPARAMS
1286 # Target files creates based on the given number of targets
1287 for (( i=0; i < $xdd_ntargets; i++ ))
1289 files+="${testdir}/xdd"$i" "
1292 # -targets specifies the devices or files to perform operation
1293 # -reqsize number of 'blocks' per operation
1294 # -mbytes number of 1024*1024-byte blocks to transfer
1295 # -blocksize size of a single 'block'
1296 # -passes number of times to read mbytes
1297 # -queuedepth number of commands to queue on the target
1298 # -rwratio percentage of read to write operations
1299 # -verbose will print out statistics on each pass
1301 local cmd="$XDD -targets $xdd_ntargets $files -reqsize $xdd_reqsize \
1302 -mbytes $xdd_mbytes -blocksize $xdd_blocksize \
1303 -passes $xdd_passes -queuedepth $xdd_queuedepth \
1304 -rwratio $xdd_rwratio -verbose $xdd_custom_params"
1308 do_nodesv $clients "$cmd "
1311 [ $rc = 0 ] || error "xdd failed: $rc"