2 # -*- mode: Bash; tab-width: 4; indent-tabs-mode: t; -*-
3 # vim:shiftwidth=4:softtabstop=4:tabstop=4:
5 # Simple function used by run_*.sh scripts
10 if [ -z "${!name}" ]; then
11 echo "$0: $name must be set"
15 [ $failed ] && exit 1 || true
18 # lrepl - Lustre test Read-Eval-Print Loop.
20 # This function implements a REPL for the Lustre test framework. It
21 # doesn't exec an actual shell because the user may want to inspect
22 # variables and use functions from the test framework.
29 This is an interactive read-eval-print loop interactive shell
30 simulation that you can use to debug failing tests. You can
31 enter most bash command lines (see notes below).
33 Use this REPL to inspect variables, set them, call test
34 framework shell functions, etcetera.
36 'exit' or EOF to exit this shell.
38 set \$retcode to 0 to cause the assertion failure that
39 triggered this REPL to be ignored.
42 do_facet ost1 lctl get_param ost.*.ost.threads_*
43 do_rpc_nodes \$OSTNODES unload_modules
46 All but the last line of multi-line statements or blocks
47 must end in a backslash.
49 "Here documents" are not supported.
51 History is not supported, but command-line editing is.
55 # Prompt escapes don't work in read -p, sadly.
56 prompt=":test_${testnum:-UNKNOWN}:$(uname -n):$(basename $PWD)% "
58 # We use read -r to get close to a shell experience
59 while read -e -r -p "$prompt" rawline; do
62 # Don't want to exit-exit, just exit the REPL
64 # We need to handle continuations, and read -r doesn't do
65 # that for us. Yet we need read -r.
67 # We also use case/esac to compare lines read to "*\\"
68 # because [ "$line" = *\\ ] and variants of that don't work.
70 while read -e -r -p '> ' rawline
72 line="$line"$'\n'"$rawline"
74 # We could check for here documents by matching
75 # against *<<*, but who cares.
88 # Finally! Time to eval.
92 echo $'\n\tExiting interactive shell...\n'
96 # lassert - Lustre test framework assert
98 # Arguments: failure code, failure message, expression/statement
100 # lassert evaluates the expression given, and, if false, calls
101 # error() to trigger test failure. If REPL_ON_LASSERT is true then
102 # lassert will call lrepl() to give the user an interactive shell.
103 # If the REPL sets retcode=0 then the assertion failure will be
110 echo "checking $* ($(eval echo \""$*"\"))..."
111 eval "$@" && return 0;
113 if ${REPL_ON_LASSERT:-false}; then
114 echo "Assertion $retcode failed: $* (expanded: $(eval echo \""$*"\"))
119 error "Assertion $retcode failed: $* (expanded: $(eval echo \""$*"\"))
124 # setmodopts- set module options for subsequent calls to load_modules
126 # Usage: setmodopts module_name new_value [var_in_which_to_save_old_value]
127 # setmodopts -a module_name new_value [var_in_which_to_save_old_value]
129 # In the second usage the new value is appended to the old.
133 if [ "$1" = -a ]; then
138 local _var=MODOPTS_$1
143 # Dynamic naming of variables is a pain in bash. In ksh93 we could
144 # write "nameref opts_var=${modname}_MODOPTS" then assign directly
145 # to opts_var. Associative arrays would also help, alternatively.
146 # Alas, we're stuck with eval until all distros move to a more recent
147 # version of bash. Fortunately we don't need to eval unset and export.
149 if [ -z "$_newvalue" ]; then
155 $_append && _newvalue="$_oldvalue $_newvalue"
156 export $_var="$_newvalue"
157 echo setmodopts: ${_var}=${_newvalue}
159 [ -n "$_savevar" ] && eval $_savevar=\""$_oldvalue"\"
162 echoerr () { echo "$@" 1>&2 ; }
165 echoerr "$(date +'%F %H:%M:%S'): client load was signaled to terminate"
167 local PGID=$(ps -eo "%c %p %r" | awk "/ $PPID / {print \$3}")
174 local mpirun="$MPIRUN $MPIRUN_OPTIONS"
175 local command="$mpirun $@"
176 local mpilog=$TMP/mpi.log
179 if [ -n "$MPI_USER" -a "$MPI_USER" != root -a -n "$mpirun" ]; then
180 echo "+ chmod 0777 $MOUNT"
182 command="su $MPI_USER sh -c \"$command \""
187 eval $command 2>&1 | tee $mpilog || true
190 if [ $rc -eq 0 ] && grep -q "p4_error:" $mpilog ; then
199 for i in ${1//,/ }; do
200 if [ "$list" = "" ]; then
203 list="$list$escape $i@$NETTYPE"
209 # FIXME: all setup/cleanup can be done without rpc.sh
212 [ x$1 = x--verbose ] && verbose=true
214 export LST_SESSION=`$LST show_session 2>/dev/null | awk -F " " '{print $5}'`
215 [ "$LST_SESSION" == "" ] && return
224 lst_session_cleanup_all () {
225 local list=$(comma_list $(nodes_list))
226 do_rpc_nodes $list lst_end_session
230 lsmod | grep -q lnet_selftest && \
231 rmmod lnet_selftest > /dev/null 2>&1 || true
235 local list=$(comma_list $(nodes_list))
237 # lst end_session needs to be executed only locally
238 # i.e. on node where lst new_session was called
239 lst_end_session --verbose
240 do_rpc_nodes $list lst_cleanup
244 load_module lnet_selftest
248 local list=$(comma_list $(nodes_list))
249 do_rpc_nodes $list lst_setup
255 # Passed a single argument, strips everything off following
256 # and includes the first period.
257 # client-20.lab.whamcloud.com becomes client-20
259 echo $(sed 's/\..*//' <<< $1)
265 # Find remote nodename, stripped of any domain, etc.
266 # 'hostname -s' is easy, but not implemented on all systems
268 local rname=$(do_node $1 "uname -n" || echo -1)
269 if [[ "$rname" = "-1" ]]; then
272 echo $(short_hostname $rname)
282 echo "${var}=${!var}"
284 [ -e $MACHINEFILE ] && cat $MACHINEFILE
289 # compile dir kernel-0 ~1GB
290 # required space ~1GB * cbench_IDIRS
294 cbench_DIR=${cbench_DIR:-""}
295 cbench_IDIRS=${cbench_IDIRS:-2}
296 cbench_RUNS=${cbench_RUNS:-2}
298 print_opts cbench_DIR cbench_IDIRS cbench_RUNS
300 [ x$cbench_DIR = x ] &&
301 { skip_env "compilebench not found" && return; }
303 [ -e $cbench_DIR/compilebench ] || \
304 { skip_env "No compilebench build" && return; }
306 local space=$(df -P $dir | tail -n 1 | awk '{ print $4 }')
307 if [[ $space -le $((1024 * 1024 * cbench_IDIRS)) ]]; then
308 cbench_IDIRS=$((space / 1024 / 1024))
309 [[ $cbench_IDIRS -eq 0 ]] &&
310 skip_env "Need free space at least 1GB, have $space" &&
313 echo "free space=$space, reducing initial dirs to $cbench_IDIRS"
317 # t-f _base needs to be modifyed to set properly tdir
318 # for new "test_foo" functions names
319 # local testdir=$DIR/$tdir
320 local testdir=$dir/d0.compilebench
325 local cmd="./compilebench -D $testdir -i $cbench_IDIRS \
326 -r $cbench_RUNS --makej"
335 [ $rc = 0 ] || error "compilebench failed: $rc"
341 METABENCH=${METABENCH:-$(which metabench 2> /dev/null || true)}
342 mbench_NFILES=${mbench_NFILES:-30400}
344 mbench_THREADS=${mbench_THREADS:-4}
345 mbench_OPTIONS=${mbench_OPTIONS:-}
347 [ x$METABENCH = x ] &&
348 { skip_env "metabench not found" && return; }
351 # Need space estimation here.
353 print_opts METABENCH clients mbench_NFILES mbench_THREADS
355 local testdir=$DIR/d0.metabench
357 # mpi_run uses mpiuser
360 # -C Run the file creation tests.
361 # -S Run the file stat tests.
362 # -c nfile Number of files to be used in each test.
363 # -k Cleanup. Remove the test directories.
364 local cmd="$METABENCH -w $testdir -c $mbench_NFILES -C -S -k $mbench_OPTIONS"
367 # find out if we need to use srun by checking $SRUN_PARTITION
368 if [ "$SRUN_PARTITION" ]; then
369 $SRUN $SRUN_OPTIONS -D $testdir -w $clients -N $num_clients \
370 -n $((num_clients * mbench_THREADS)) \
371 -p $SRUN_PARTITION -- $cmd
373 mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \
374 -np $((num_clients * $mbench_THREADS)) $cmd
378 if [ $rc != 0 ] ; then
379 error "metabench failed! $rc"
386 SIMUL=${SIMUL:=$(which simul 2> /dev/null || true)}
388 simul_THREADS=${simul_THREADS:-2}
389 simul_REP=${simul_REP:-20}
391 if [ "$NFSCLIENT" ]; then
392 skip "skipped for NFSCLIENT mode"
397 { skip_env "simul not found" && return; }
400 # Need space estimation here.
402 print_opts SIMUL clients simul_REP simul_THREADS
404 local testdir=$DIR/d0.simul
406 # mpi_run uses mpiuser
409 # -n # : repeat each test # times
410 # -N # : repeat the entire set of tests # times
412 local cmd="$SIMUL -d $testdir -n $simul_REP -N $simul_REP"
415 # find out if we need to use srun by checking $SRUN_PARTITION
416 if [ "$SRUN_PARTITION" ]; then
417 $SRUN $SRUN_OPTIONS -D $testdir -w $clients -N $num_clients \
418 -n $((num_clients * simul_THREADS)) -p $SRUN_PARTITION \
421 mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \
422 -np $((num_clients * simul_THREADS)) $cmd
426 if [ $rc != 0 ] ; then
427 error "simul failed! $rc"
434 MDTEST=${MDTEST:=$(which mdtest 2> /dev/null || true)}
436 mdtest_THREADS=${mdtest_THREADS:-2}
437 mdtest_nFiles=${mdtest_nFiles:-"100000"}
438 # We devide the files by number of core
439 mdtest_nFiles=$((mdtest_nFiles/mdtest_THREADS/num_clients))
440 mdtest_iteration=${mdtest_iteration:-1}
442 local type=${1:-"ssf"}
444 if [ "$NFSCLIENT" ]; then
445 skip "skipped for NFSCLIENT mode"
450 { skip_env "mdtest not found" && return; }
453 # Need space estimation here.
455 print_opts MDTEST mdtest_iteration mdtest_THREADS mdtest_nFiles
457 local testdir=$DIR/d0.mdtest
459 # mpi_run uses mpiuser
462 # -i # : repeat each test # times
464 # -n # : number of file/dir to create/stat/remove
465 # -u : each process create/stat/remove individually
467 local cmd="$MDTEST -d $testdir -i $mdtest_iteration -n $mdtest_nFiles"
468 [ $type = "fpp" ] && cmd="$cmd -u"
471 # find out if we need to use srun by checking $SRUN_PARTITION
472 if [ "$SRUN_PARTITION" ]; then
473 $SRUN $SRUN_OPTIONS -D $testdir -w $clients -N $num_clients \
474 -n $((num_clients * mdtest_THREADS)) \
475 -p $SRUN_PARTITION -- $cmd
477 mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \
478 -np $((num_clients * mdtest_THREADS)) $cmd
482 if [ $rc != 0 ] ; then
483 error "mdtest failed! $rc"
490 cnt_DIR=${cnt_DIR:-""}
491 cnt_NRUN=${cnt_NRUN:-10}
493 print_opts cnt_DIR cnt_NRUN
496 { skip_env "connectathon dir not found" && return; }
498 [ -e $cnt_DIR/runtests ] || \
499 { skip_env "No connectathon runtests found" && return; }
501 local testdir=$DIR/d0.connectathon
508 # cthon options (must be in this order)
510 # -N numpasses - will be passed to the runtests script. This argument
511 # is optional. It specifies the number of times to run
514 # One of these test types
519 # -a all of the above
521 # -f a quick functionality test
525 # Include lock tests unless we're running on nfsv4
526 local fstype=$(df -TP $testdir | awk 'NR==2 {print $2}')
527 echo "$testdir: $fstype"
528 if [[ $fstype != "nfs4" ]]; then
532 for test in $tests; do
533 local cmd="./runtests -N $cnt_NRUN $test -f $testdir"
539 [ $rc = 0 ] || error "connectathon failed: $rc"
547 local type=${1:="ssf"}
549 IOR=${IOR:-$(which IOR 2> /dev/null || true)}
551 ior_THREADS=${ior_THREADS:-2}
552 ior_iteration=${ior_iteration:-1}
553 ior_blockSize=${ior_blockSize:-6}
554 ior_blockUnit=${ior_blockUnit:-M} # K, M, G
555 ior_xferSize=${ior_xferSize:-1M}
556 ior_type=${ior_type:-POSIX}
557 ior_DURATION=${ior_DURATION:-30} # minutes
559 case ${ior_blockUnit} in
561 multiplier=$((1024 * 1024 * 1024))
564 multiplier=$((1024 * 1024))
569 *) error "Incorrect block unit should be one of [KMG]"
574 { skip_env "IOR not found" && return; }
576 # calculate the space in bytes
577 local space=$(df -B 1 -P $DIR | tail -n 1 | awk '{ print $4 }')
578 local total_threads=$((num_clients * ior_THREADS))
579 echo "+ $ior_blockSize * $multiplier * $total_threads "
580 if [ $((space / 2)) -le \
581 $((ior_blockSize * multiplier * total_threads)) ]; then
582 ior_blockSize=$((space / 2 / multiplier / total_threads))
583 [ $ior_blockSize -eq 0 ] && \
584 skip_env "Need free space more than $((2 * total_threads)) \
585 ${ior_blockUnit}: have $((space / multiplier))" &&
588 echo "(reduced blockSize to $ior_blockSize \
589 ${ior_blockUnit} bytes)"
592 print_opts IOR ior_THREADS ior_DURATION MACHINEFILE
594 local testdir=$DIR/d0.ior.$type
596 # mpi_run uses mpiuser
598 if [ "$NFSCLIENT" ]; then
599 setstripe_nfsserver $testdir -c -1 ||
600 { error "setstripe on nfsserver failed" && return 1; }
602 $LFS setstripe $testdir -c -1 ||
603 { error "setstripe failed" && return 2; }
607 # contiguous bytes to write per task (e.g.: 8, 4K, 2M, 1G)"
609 # -t N transferSize -- size of transfer in bytes (e.g.: 8, 4K, 2M, 1G)"
610 # -w writeFile -- write file"
611 # -r readFile -- read existing file"
612 # -W checkWrite -- check read after write"
613 # -C reorderTasks -- changes task ordering to n+1 ordering for readback
614 # -T maxTimeDuration -- max time in minutes to run tests"
615 # -k keepFile -- keep testFile(s) on program exit
617 local cmd="$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! $rc"
643 MIB=${MIB:=$(which mib 2> /dev/null || true)}
645 mib_THREADS=${mib_THREADS:-2}
646 mib_xferSize=${mib_xferSize:-1m}
647 mib_xferLimit=${mib_xferLimit:-5000}
648 mib_timeLimit=${mib_timeLimit:-300}
650 if [ "$NFSCLIENT" ]; then
651 skip "skipped for NFSCLIENT mode"
656 { skip_env "MIB not found" && return; }
658 print_opts MIB mib_THREADS mib_xferSize mib_xferLimit mib_timeLimit \
661 local testdir=$DIR/d0.mib
663 # mpi_run uses mpiuser
665 $LFS setstripe $testdir -c -1 ||
666 { error "setstripe failed" && return 2; }
668 # -I Show intermediate values in output
669 # -H Show headers in output
670 # -L Do not issue new system calls after this many seconds
671 # -s Use system calls of this size
673 # -l Issue no more than this many system calls
674 local cmd="$MIB -t $testdir -s $mib_xferSize -l $mib_xferLimit \
675 -L $mib_timeLimit -HI -p mib.$(date +%Y%m%d%H%M%S)"
678 # find out if we need to use srun by checking $SRUN_PARTITION
679 if [ "$SRUN_PARTITION" ]; then
680 $SRUN $SRUN_OPTIONS -D $testdir -w $clients -N $num_clients \
681 -n $((num_clients * mib_THREADS)) -p $SRUN_PARTITION \
684 mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \
685 -np $((num_clients * mib_THREADS)) $cmd
689 if [ $rc != 0 ] ; then
690 error "mib failed! $rc"
697 CASC_RW=${CASC_RW:-$(which cascading_rw 2> /dev/null || true)}
699 casc_THREADS=${casc_THREADS:-2}
700 casc_REP=${casc_REP:-300}
702 if [ "$NFSCLIENT" ]; then
703 skip "skipped for NFSCLIENT mode"
708 { skip_env "cascading_rw not found" && return; }
711 # Need space estimation here.
713 print_opts CASC_RW clients casc_THREADS casc_REP MACHINEFILE
715 local testdir=$DIR/d0.cascading_rw
717 # mpi_run uses mpiuser
721 # -n: repeat test # times
723 local cmd="$CASC_RW -g -d $testdir -n $casc_REP"
726 mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \
727 -np $((num_clients * $casc_THREADS)) $cmd
730 if [ $rc != 0 ] ; then
731 error "cascading_rw failed! $rc"
736 run_write_append_truncate() {
739 write_THREADS=${write_THREADS:-8}
740 write_REP=${write_REP:-10000}
742 if [ "$NFSCLIENT" ]; then
743 skip "skipped for NFSCLIENT mode"
747 # location is lustre/tests dir
748 if ! which write_append_truncate > /dev/null 2>&1 ; then
749 skip_env "write_append_truncate not found"
754 # Need space estimation here.
756 local testdir=$DIR/d0.write_append_truncate
757 local file=$testdir/f0.wat
759 print_opts clients write_REP write_THREADS MACHINEFILE
762 # mpi_run uses mpiuser
765 local cmd="write_append_truncate -n $write_REP $file"
768 mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \
769 -np $((num_clients * $write_THREADS)) $cmd
772 if [ $rc != 0 ] ; then
773 error "write_append_truncate failed! $rc"
779 run_write_disjoint() {
781 WRITE_DISJOINT=${WRITE_DISJOINT:-$(which write_disjoint \
782 2> /dev/null || true)}
784 wdisjoint_THREADS=${wdisjoint_THREADS:-4}
785 wdisjoint_REP=${wdisjoint_REP:-10000}
787 if [ "$NFSCLIENT" ]; then
788 skip "skipped for NFSCLIENT mode"
792 [ x$WRITE_DISJOINT = x ] &&
793 { skip_env "write_disjoint not found" && return; }
796 # Need space estimation here.
798 print_opts WRITE_DISJOINT clients wdisjoint_THREADS wdisjoint_REP \
800 local testdir=$DIR/d0.write_disjoint
802 # mpi_run uses mpiuser
805 local cmd="$WRITE_DISJOINT -f $testdir/file -n $wdisjoint_REP"
808 mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \
809 -np $((num_clients * $wdisjoint_THREADS)) $cmd
812 if [ $rc != 0 ] ; then
813 error "write_disjoint failed! $rc"
818 run_parallel_grouplock() {
820 PARALLEL_GROUPLOCK=${PARALLEL_GROUPLOCK:-$(which parallel_grouplock \
821 2> /dev/null || true)}
822 parallel_grouplock_MINTASKS=${parallel_grouplock_MINTASKS:-5}
824 if [ "$NFSCLIENT" ]; then
825 skip "skipped for NFSCLIENT mode"
829 [ x$PARALLEL_GROUPLOCK = x ] &&
830 { skip "PARALLEL_GROUPLOCK not found" && return; }
832 print_opts clients parallel_grouplock_MINTASKS MACHINEFILE
834 local testdir=$DIR/d0.parallel_grouplock
836 # mpi_run uses mpiuser
842 for i in $(seq 12); do
844 local cmd="$PARALLEL_GROUPLOCK -g -v -d $testdir $subtest"
847 mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \
848 -np $parallel_grouplock_MINTASKS $cmd
850 if [ $rc != 0 ] ; then
851 error_noexit "parallel_grouplock subtests $subtest " \
854 echo "parallel_grouplock subtests $subtest PASS"
856 let status=$((status + rc))
857 # clear debug to collect one log per one test
858 do_nodes $(comma_list $(nodes_list)) lctl clear
860 [ $status -eq 0 ] || error "parallel_grouplock status: $status"
864 cleanup_statahead () {
871 for i in $(seq 0 $num_mntpts);do
872 zconf_umount_clients $clients ${mntpt_root}$i ||
873 error_exit "Failed to umount lustre on ${mntpt_root}$i"
879 statahead_NUMMNTPTS=${statahead_NUMMNTPTS:-5}
880 statahead_NUMFILES=${statahead_NUMFILES:-500000}
882 if [[ -n $NFSCLIENT ]]; then
883 skip "Statahead testing is not supported on NFS clients."
888 { skip_env "mdsrate not found" && return; }
890 print_opts MDSRATE clients statahead_NUMMNTPTS statahead_NUMFILES
894 # do not use default "d[0-9]*" dir name
895 # to avoid of rm $statahead_NUMFILES (500k) files in t-f cleanup
897 local testdir=$DIR/$dir
899 # cleanup only if dir exists
900 # cleanup only $statahead_NUMFILES number of files
901 # ignore the other files created by someone else
903 mdsrate_cleanup $((num_clients * 32)) $MACHINEFILE \
904 $statahead_NUMFILES $testdir 'f%%d' --ignore
907 # mpi_run uses mpiuser
910 local num_files=$statahead_NUMFILES
912 local IFree=$(inodes_available)
913 if [ $IFree -lt $num_files ]; then
919 local cmd1="${MDSRATE} ${MDSRATE_DEBUG} --mknod --dir $testdir"
920 local cmd2="--nfiles $num_files --filefmt 'f%%d'"
921 local cmd="$cmd1 $cmd2"
924 mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \
925 -np $((num_clients * 32)) $cmd
928 if [ $rc != 0 ] ; then
929 error "mdsrate failed to create $rc"
933 local num_mntpts=$statahead_NUMMNTPTS
934 local mntpt_root=$TMP/mntpt/lustre
935 local mntopts=$MNTOPTSTATAHEAD
937 echo "Mounting $num_mntpts lustre clients starts on $clients"
938 trap "cleanup_statahead $clients $mntpt_root $num_mntpts" EXIT ERR
939 for i in $(seq 0 $num_mntpts); do
940 zconf_mount_clients $clients ${mntpt_root}$i "$mntopts" ||
941 error_exit "Failed to mount lustre on ${mntpt_root}$i on $clients"
944 do_rpc_nodes $clients cancel_lru_locks mdc
946 do_rpc_nodes $clients do_ls $mntpt_root $num_mntpts $dir
948 mdsrate_cleanup $((num_clients * 32)) $MACHINEFILE \
949 $num_files $testdir 'f%%d' --ignore
951 # use rm instead of rmdir because of
952 # testdir could contain the files created by someone else,
953 # or by previous run where is num_files prev > num_files current
955 cleanup_statahead $clients $mntpt_root $num_mntpts