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=${ior_blockSize:-6} # GB
554 ior_xferSize=${ior_xferSize:-2m}
555 ior_type=${ior_type:-POSIX}
556 ior_DURATION=${ior_DURATION:-30} # minutes
559 { skip_env "IOR not found" && return; }
561 local space=$(df -P $DIR | tail -n 1 | awk '{ print $4 }')
562 local total_threads=$(( num_clients * ior_THREADS ))
563 echo "+ $ior_blockSize * 1024 * 1024 * $total_threads "
564 if [ $((space / 2)) -le \
565 $(( ior_blockSize * 1024 * 1024 * total_threads)) ]; then
566 echo "+ $space * 9/10 / 1024 / 1024 / $num_clients / $ior_THREADS"
567 ior_blockSize=$(( space /2 /1024 /1024 / num_clients / ior_THREADS ))
568 [ $ior_blockSize = 0 ] && \
569 skip_env "Need free space more than $((2 * total_threads))GB: \
570 $((total_threads *1024 *1024*2)), have $space" && return
572 local reduced_size="$num_clients x $ior_THREADS x $ior_blockSize"
573 echo "free space=$space, Need: $reduced_size GB"
574 echo "(blockSize reduced to $ior_blockSize Gb)"
577 print_opts IOR ior_THREADS ior_DURATION MACHINEFILE
579 local testdir=$DIR/d0.ior.$type
581 # mpi_run uses mpiuser
583 if [ "$NFSCLIENT" ]; then
584 setstripe_nfsserver $testdir -c -1 ||
585 { error "setstripe on nfsserver failed" && return 1; }
587 $LFS setstripe $testdir -c -1 ||
588 { error "setstripe failed" && return 2; }
592 # contiguous bytes to write per task (e.g.: 8, 4k, 2m, 1g)"
594 # -t N transferSize -- size of transfer in bytes (e.g.: 8, 4k, 2m, 1g)"
595 # -w writeFile -- write file"
596 # -r readFile -- read existing file"
597 # -W checkWrite -- check read after write"
598 # -C reorderTasks -- changes task ordering to n+1 ordering for readback
599 # -T maxTimeDuration -- max time in minutes to run tests"
600 # -k keepFile -- keep testFile(s) on program exit
602 local cmd="$IOR -a $ior_type -b ${ior_blockSize}g -o $testdir/iorData \
603 -t $ior_xferSize -v -C -w -r -W -i $ior_iteration -T $ior_DURATION -k"
604 [ $type = "fpp" ] && cmd="$cmd -F"
607 # find out if we need to use srun by checking $SRUN_PARTITION
608 if [ "$SRUN_PARTITION" ]; then
609 $SRUN $SRUN_OPTIONS -D $testdir -w $clients -N $num_clients \
610 -n $((num_clients * ior_THREADS)) -p $SRUN_PARTITION \
613 mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \
614 -np $((num_clients * $ior_THREADS)) $cmd
618 if [ $rc != 0 ] ; then
619 error "ior failed! $rc"
626 MIB=${MIB:=$(which mib 2> /dev/null || true)}
628 mib_THREADS=${mib_THREADS:-2}
629 mib_xferSize=${mib_xferSize:-1m}
630 mib_xferLimit=${mib_xferLimit:-5000}
631 mib_timeLimit=${mib_timeLimit:-300}
633 if [ "$NFSCLIENT" ]; then
634 skip "skipped for NFSCLIENT mode"
639 { skip_env "MIB not found" && return; }
641 print_opts MIB mib_THREADS mib_xferSize mib_xferLimit mib_timeLimit \
644 local testdir=$DIR/d0.mib
646 # mpi_run uses mpiuser
648 $LFS setstripe $testdir -c -1 ||
649 { error "setstripe failed" && return 2; }
651 # -I Show intermediate values in output
652 # -H Show headers in output
653 # -L Do not issue new system calls after this many seconds
654 # -s Use system calls of this size
656 # -l Issue no more than this many system calls
657 local cmd="$MIB -t $testdir -s $mib_xferSize -l $mib_xferLimit \
658 -L $mib_timeLimit -HI -p mib.$(date +%Y%m%d%H%M%S)"
661 # find out if we need to use srun by checking $SRUN_PARTITION
662 if [ "$SRUN_PARTITION" ]; then
663 $SRUN $SRUN_OPTIONS -D $testdir -w $clients -N $num_clients \
664 -n $((num_clients * mib_THREADS)) -p $SRUN_PARTITION \
667 mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \
668 -np $((num_clients * mib_THREADS)) $cmd
672 if [ $rc != 0 ] ; then
673 error "mib failed! $rc"
680 CASC_RW=${CASC_RW:-$(which cascading_rw 2> /dev/null || true)}
682 casc_THREADS=${casc_THREADS:-2}
683 casc_REP=${casc_REP:-300}
685 if [ "$NFSCLIENT" ]; then
686 skip "skipped for NFSCLIENT mode"
691 { skip_env "cascading_rw not found" && return; }
694 # Need space estimation here.
696 print_opts CASC_RW clients casc_THREADS casc_REP MACHINEFILE
698 local testdir=$DIR/d0.cascading_rw
700 # mpi_run uses mpiuser
704 # -n: repeat test # times
706 local cmd="$CASC_RW -g -d $testdir -n $casc_REP"
709 mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \
710 -np $((num_clients * $casc_THREADS)) $cmd
713 if [ $rc != 0 ] ; then
714 error "cascading_rw failed! $rc"
719 run_write_append_truncate() {
722 write_THREADS=${write_THREADS:-8}
723 write_REP=${write_REP:-10000}
725 if [ "$NFSCLIENT" ]; then
726 skip "skipped for NFSCLIENT mode"
730 # location is lustre/tests dir
731 if ! which write_append_truncate > /dev/null 2>&1 ; then
732 skip_env "write_append_truncate not found"
737 # Need space estimation here.
739 local testdir=$DIR/d0.write_append_truncate
740 local file=$testdir/f0.wat
742 print_opts clients write_REP write_THREADS MACHINEFILE
745 # mpi_run uses mpiuser
748 local cmd="write_append_truncate -n $write_REP $file"
751 mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \
752 -np $((num_clients * $write_THREADS)) $cmd
755 if [ $rc != 0 ] ; then
756 error "write_append_truncate failed! $rc"
762 run_write_disjoint() {
764 WRITE_DISJOINT=${WRITE_DISJOINT:-$(which write_disjoint \
765 2> /dev/null || true)}
767 wdisjoint_THREADS=${wdisjoint_THREADS:-4}
768 wdisjoint_REP=${wdisjoint_REP:-10000}
770 if [ "$NFSCLIENT" ]; then
771 skip "skipped for NFSCLIENT mode"
775 [ x$WRITE_DISJOINT = x ] &&
776 { skip_env "write_disjoint not found" && return; }
779 # Need space estimation here.
781 print_opts WRITE_DISJOINT clients wdisjoint_THREADS wdisjoint_REP \
783 local testdir=$DIR/d0.write_disjoint
785 # mpi_run uses mpiuser
788 local cmd="$WRITE_DISJOINT -f $testdir/file -n $wdisjoint_REP"
791 mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \
792 -np $((num_clients * $wdisjoint_THREADS)) $cmd
795 if [ $rc != 0 ] ; then
796 error "write_disjoint failed! $rc"
801 run_parallel_grouplock() {
803 PARALLEL_GROUPLOCK=${PARALLEL_GROUPLOCK:-$(which parallel_grouplock \
804 2> /dev/null || true)}
805 parallel_grouplock_MINTASKS=${parallel_grouplock_MINTASKS:-5}
807 if [ "$NFSCLIENT" ]; then
808 skip "skipped for NFSCLIENT mode"
812 [ x$PARALLEL_GROUPLOCK = x ] &&
813 { skip "PARALLEL_GROUPLOCK not found" && return; }
815 print_opts clients parallel_grouplock_MINTASKS MACHINEFILE
817 local testdir=$DIR/d0.parallel_grouplock
819 # mpi_run uses mpiuser
825 for i in $(seq 12); do
827 local cmd="$PARALLEL_GROUPLOCK -g -v -d $testdir $subtest"
830 mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \
831 -np $parallel_grouplock_MINTASKS $cmd
833 if [ $rc != 0 ] ; then
834 error_noexit "parallel_grouplock subtests $subtest " \
837 echo "parallel_grouplock subtests $subtest PASS"
839 let status=$((status + rc))
840 # clear debug to collect one log per one test
841 do_nodes $(comma_list $(nodes_list)) lctl clear
843 [ $status -eq 0 ] || error "parallel_grouplock status: $status"
847 cleanup_statahead () {
854 for i in $(seq 0 $num_mntpts);do
855 zconf_umount_clients $clients ${mntpt_root}$i ||
856 error_exit "Failed to umount lustre on ${mntpt_root}$i"
862 statahead_NUMMNTPTS=${statahead_NUMMNTPTS:-5}
863 statahead_NUMFILES=${statahead_NUMFILES:-500000}
865 if [[ -n $NFSCLIENT ]]; then
866 skip "Statahead testing is not supported on NFS clients."
871 { skip_env "mdsrate not found" && return; }
873 print_opts MDSRATE clients statahead_NUMMNTPTS statahead_NUMFILES
877 # do not use default "d[0-9]*" dir name
878 # to avoid of rm $statahead_NUMFILES (500k) files in t-f cleanup
880 local testdir=$DIR/$dir
882 # cleanup only if dir exists
883 # cleanup only $statahead_NUMFILES number of files
884 # ignore the other files created by someone else
886 mdsrate_cleanup $((num_clients * 32)) $MACHINEFILE \
887 $statahead_NUMFILES $testdir 'f%%d' --ignore
890 # mpi_run uses mpiuser
893 local num_files=$statahead_NUMFILES
895 local IFree=$(inodes_available)
896 if [ $IFree -lt $num_files ]; then
902 local cmd1="${MDSRATE} ${MDSRATE_DEBUG} --mknod --dir $testdir"
903 local cmd2="--nfiles $num_files --filefmt 'f%%d'"
904 local cmd="$cmd1 $cmd2"
907 mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \
908 -np $((num_clients * 32)) $cmd
911 if [ $rc != 0 ] ; then
912 error "mdsrate failed to create $rc"
916 local num_mntpts=$statahead_NUMMNTPTS
917 local mntpt_root=$TMP/mntpt/lustre
918 local mntopts=$MNTOPTSTATAHEAD
920 echo "Mounting $num_mntpts lustre clients starts on $clients"
921 trap "cleanup_statahead $clients $mntpt_root $num_mntpts" EXIT ERR
922 for i in $(seq 0 $num_mntpts); do
923 zconf_mount_clients $clients ${mntpt_root}$i "$mntopts" ||
924 error_exit "Failed to mount lustre on ${mntpt_root}$i on $clients"
927 do_rpc_nodes $clients cancel_lru_locks mdc
929 do_rpc_nodes $clients do_ls $mntpt_root $num_mntpts $dir
931 mdsrate_cleanup $((num_clients * 32)) $MACHINEFILE \
932 $num_files $testdir 'f%%d' --ignore
934 # use rm instead of rmdir because of
935 # testdir could contain the files created by someone else,
936 # or by previous run where is num_files prev > num_files current
938 cleanup_statahead $clients $mntpt_root $num_mntpts