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
198 for i in ${1//,/ }; do
199 list="$list $i@$NETTYPE"
204 # FIXME: all setup/cleanup can be done without rpc.sh
207 [ x$1 = x--verbose ] && verbose=true
209 export LST_SESSION=`$LST show_session 2>/dev/null | awk -F " " '{print $5}'`
210 [ "$LST_SESSION" == "" ] && return
219 lst_session_cleanup_all () {
220 local list=$(comma_list $(nodes_list))
221 do_rpc_nodes $list lst_end_session
225 lsmod | grep -q lnet_selftest && \
226 rmmod lnet_selftest > /dev/null 2>&1 || true
230 local list=$(comma_list $(nodes_list))
232 # lst end_session needs to be executed only locally
233 # i.e. on node where lst new_session was called
234 lst_end_session --verbose
235 do_rpc_nodes $list lst_cleanup
239 load_module lnet_selftest
243 local list=$(comma_list $(nodes_list))
244 do_rpc_nodes $list lst_setup
250 # Passed a single argument, strips everything off following
251 # and includes the first period.
252 # client-20.lab.whamcloud.com becomes client-20
254 echo $(sed 's/\..*//' <<< $1)
260 # Find remote nodename, stripped of any domain, etc.
261 # 'hostname -s' is easy, but not implemented on all systems
263 local rname=$(do_node $1 "uname -n" || echo -1)
264 if [[ "$rname" = "-1" ]]; then
267 echo $(short_hostname $rname)
277 echo "${var}=${!var}"
279 [ -e $MACHINEFILE ] && cat $MACHINEFILE
284 # compile dir kernel-0 ~1GB
285 # required space ~1GB * cbench_IDIRS
287 cbench_DIR=${cbench_DIR:-""}
288 cbench_IDIRS=${cbench_IDIRS:-2}
289 cbench_RUNS=${cbench_RUNS:-2}
291 print_opts cbench_DIR cbench_IDIRS cbench_RUNS
293 [ x$cbench_DIR = x ] &&
294 { skip_env "compilebench not found" && return; }
296 [ -e $cbench_DIR/compilebench ] || \
297 { skip_env "No compilebench build" && return; }
299 local space=$(df -P $DIR | tail -n 1 | awk '{ print $4 }')
300 if [[ $space -le $((1024 * 1024 * cbench_IDIRS)) ]]; then
301 cbench_IDIRS=$((space / 1024 / 1024))
302 [[ $cbench_IDIRS -eq 0 ]] &&
303 skip_env "Need free space at least 1GB, have $space" &&
306 echo "free space=$space, reducing initial dirs to $cbench_IDIRS"
310 # t-f _base needs to be modifyed to set properly tdir
311 # for new "test_foo" functions names
312 # local testdir=$DIR/$tdir
313 local testdir=$DIR/d0.compilebench
318 local cmd="./compilebench -D $testdir -i $cbench_IDIRS \
319 -r $cbench_RUNS --makej"
328 [ $rc = 0 ] || error "compilebench failed: $rc"
334 METABENCH=${METABENCH:-$(which metabench 2> /dev/null || true)}
335 mbench_NFILES=${mbench_NFILES:-30400}
337 mbench_THREADS=${mbench_THREADS:-4}
338 mbench_OPTIONS=${mbench_OPTIONS:-}
340 [ x$METABENCH = x ] &&
341 { skip_env "metabench not found" && return; }
344 # Need space estimation here.
346 print_opts METABENCH clients mbench_NFILES mbench_THREADS
348 local testdir=$DIR/d0.metabench
350 # mpi_run uses mpiuser
353 # -C Run the file creation tests.
354 # -S Run the file stat tests.
355 # -c nfile Number of files to be used in each test.
356 # -k Cleanup. Remove the test directories.
357 local cmd="$METABENCH -w $testdir -c $mbench_NFILES -C -S -k $mbench_OPTIONS"
360 # find out if we need to use srun by checking $SRUN_PARTITION
361 if [ "$SRUN_PARTITION" ]; then
362 $SRUN $SRUN_OPTIONS -D $testdir -w $clients -N $num_clients \
363 -n $((num_clients * mbench_THREADS)) \
364 -p $SRUN_PARTITION -- $cmd
366 mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \
367 -np $((num_clients * $mbench_THREADS)) $cmd
371 if [ $rc != 0 ] ; then
372 error "metabench failed! $rc"
379 SIMUL=${SIMUL:=$(which simul 2> /dev/null || true)}
381 simul_THREADS=${simul_THREADS:-2}
382 simul_REP=${simul_REP:-20}
384 if [ "$NFSCLIENT" ]; then
385 skip "skipped for NFSCLIENT mode"
390 { skip_env "simul not found" && return; }
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! $rc"
427 MDTEST=${MDTEST:=$(which mdtest 2> /dev/null || true)}
429 mdtest_THREADS=${mdtest_THREADS:-2}
430 mdtest_nFiles=${mdtest_nFiles:-"100000"}
431 # We devide the files by number of core
432 mdtest_nFiles=$((mdtest_nFiles/mdtest_THREADS/num_clients))
433 mdtest_iteration=${mdtest_iteration:-1}
435 local type=${1:-"ssf"}
437 if [ "$NFSCLIENT" ]; then
438 skip "skipped for NFSCLIENT mode"
443 { skip_env "mdtest not found" && return; }
446 # Need space estimation here.
448 print_opts MDTEST mdtest_iteration mdtest_THREADS mdtest_nFiles
450 local testdir=$DIR/d0.mdtest
452 # mpi_run uses mpiuser
455 # -i # : repeat each test # times
457 # -n # : number of file/dir to create/stat/remove
458 # -u : each process create/stat/remove individually
460 local cmd="$MDTEST -d $testdir -i $mdtest_iteration -n $mdtest_nFiles"
461 [ $type = "fpp" ] && cmd="$cmd -u"
464 # find out if we need to use srun by checking $SRUN_PARTITION
465 if [ "$SRUN_PARTITION" ]; then
466 $SRUN $SRUN_OPTIONS -D $testdir -w $clients -N $num_clients \
467 -n $((num_clients * mdtest_THREADS)) \
468 -p $SRUN_PARTITION -- $cmd
470 mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \
471 -np $((num_clients * mdtest_THREADS)) $cmd
475 if [ $rc != 0 ] ; then
476 error "mdtest failed! $rc"
483 cnt_DIR=${cnt_DIR:-""}
484 cnt_NRUN=${cnt_NRUN:-10}
486 print_opts cnt_DIR cnt_NRUN
489 { skip_env "connectathon dir not found" && return; }
491 [ -e $cnt_DIR/runtests ] || \
492 { skip_env "No connectathon runtests found" && return; }
494 local testdir=$DIR/d0.connectathon
501 # cthon options (must be in this order)
503 # -N numpasses - will be passed to the runtests script. This argument
504 # is optional. It specifies the number of times to run
507 # One of these test types
512 # -a all of the above
514 # -f a quick functionality test
518 # Include lock tests unless we're running on nfsv4
519 local fstype=$(df -TP $testdir | awk 'NR==2 {print $2}')
520 echo "$testdir: $fstype"
521 if [[ $fstype != "nfs4" ]]; then
525 for test in $tests; do
526 local cmd="./runtests -N $cnt_NRUN $test -f $testdir"
532 [ $rc = 0 ] || error "connectathon failed: $rc"
540 local type=${1:="ssf"}
542 IOR=${IOR:-$(which IOR 2> /dev/null || true)}
544 ior_THREADS=${ior_THREADS:-2}
545 ior_iteration=${ior_iteration:-1}
546 ior_blockSize=${ior_blockSize:-6} # GB
547 ior_xferSize=${ior_xferSize:-2m}
548 ior_type=${ior_type:-POSIX}
549 ior_DURATION=${ior_DURATION:-30} # minutes
552 { skip_env "IOR not found" && return; }
554 local space=$(df -P $DIR | tail -n 1 | awk '{ print $4 }')
555 local total_threads=$(( num_clients * ior_THREADS ))
556 echo "+ $ior_blockSize * 1024 * 1024 * $total_threads "
557 if [ $((space / 2)) -le \
558 $(( ior_blockSize * 1024 * 1024 * total_threads)) ]; then
559 echo "+ $space * 9/10 / 1024 / 1024 / $num_clients / $ior_THREADS"
560 ior_blockSize=$(( space /2 /1024 /1024 / num_clients / ior_THREADS ))
561 [ $ior_blockSize = 0 ] && \
562 skip_env "Need free space more than $((2 * total_threads))GB: \
563 $((total_threads *1024 *1024*2)), have $space" && return
565 local reduced_size="$num_clients x $ior_THREADS x $ior_blockSize"
566 echo "free space=$space, Need: $reduced_size GB"
567 echo "(blockSize reduced to $ior_blockSize Gb)"
570 print_opts IOR ior_THREADS ior_DURATION MACHINEFILE
572 local testdir=$DIR/d0.ior.$type
574 # mpi_run uses mpiuser
576 if [ "$NFSCLIENT" ]; then
577 setstripe_nfsserver $testdir -c -1 ||
578 { error "setstripe on nfsserver failed" && return 1; }
580 $LFS setstripe $testdir -c -1 ||
581 { error "setstripe failed" && return 2; }
585 # contiguous bytes to write per task (e.g.: 8, 4k, 2m, 1g)"
587 # -t N transferSize -- size of transfer in bytes (e.g.: 8, 4k, 2m, 1g)"
588 # -w writeFile -- write file"
589 # -r readFile -- read existing file"
590 # -W checkWrite -- check read after write"
591 # -C reorderTasks -- changes task ordering to n+1 ordering for readback
592 # -T maxTimeDuration -- max time in minutes to run tests"
593 # -k keepFile -- keep testFile(s) on program exit
595 local cmd="$IOR -a $ior_type -b ${ior_blockSize}g -o $testdir/iorData \
596 -t $ior_xferSize -v -C -w -r -W -i $ior_iteration -T $ior_DURATION -k"
597 [ $type = "fpp" ] && cmd="$cmd -F"
600 # find out if we need to use srun by checking $SRUN_PARTITION
601 if [ "$SRUN_PARTITION" ]; then
602 $SRUN $SRUN_OPTIONS -D $testdir -w $clients -N $num_clients \
603 -n $((num_clients * ior_THREADS)) -p $SRUN_PARTITION \
606 mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \
607 -np $((num_clients * $ior_THREADS)) $cmd
611 if [ $rc != 0 ] ; then
612 error "ior failed! $rc"
619 MIB=${MIB:=$(which mib 2> /dev/null || true)}
621 mib_THREADS=${mib_THREADS:-2}
622 mib_xferSize=${mib_xferSize:-1m}
623 mib_xferLimit=${mib_xferLimit:-5000}
624 mib_timeLimit=${mib_timeLimit:-300}
626 if [ "$NFSCLIENT" ]; then
627 skip "skipped for NFSCLIENT mode"
632 { skip_env "MIB not found" && return; }
634 print_opts MIB mib_THREADS mib_xferSize mib_xferLimit mib_timeLimit \
637 local testdir=$DIR/d0.mib
639 # mpi_run uses mpiuser
641 $LFS setstripe $testdir -c -1 ||
642 { error "setstripe failed" && return 2; }
644 # -I Show intermediate values in output
645 # -H Show headers in output
646 # -L Do not issue new system calls after this many seconds
647 # -s Use system calls of this size
649 # -l Issue no more than this many system calls
650 local cmd="$MIB -t $testdir -s $mib_xferSize -l $mib_xferLimit \
651 -L $mib_timeLimit -HI -p mib.$(date +%Y%m%d%H%M%S)"
654 # find out if we need to use srun by checking $SRUN_PARTITION
655 if [ "$SRUN_PARTITION" ]; then
656 $SRUN $SRUN_OPTIONS -D $testdir -w $clients -N $num_clients \
657 -n $((num_clients * mib_THREADS)) -p $SRUN_PARTITION \
660 mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \
661 -np $((num_clients * mib_THREADS)) $cmd
665 if [ $rc != 0 ] ; then
666 error "mib failed! $rc"
673 CASC_RW=${CASC_RW:-$(which cascading_rw 2> /dev/null || true)}
675 casc_THREADS=${casc_THREADS:-2}
676 casc_REP=${casc_REP:-300}
678 if [ "$NFSCLIENT" ]; then
679 skip "skipped for NFSCLIENT mode"
684 { skip_env "cascading_rw not found" && return; }
687 # Need space estimation here.
689 print_opts CASC_RW clients casc_THREADS casc_REP MACHINEFILE
691 local testdir=$DIR/d0.cascading_rw
693 # mpi_run uses mpiuser
697 # -n: repeat test # times
699 local cmd="$CASC_RW -g -d $testdir -n $casc_REP"
702 mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \
703 -np $((num_clients * $casc_THREADS)) $cmd
706 if [ $rc != 0 ] ; then
707 error "cascading_rw failed! $rc"
712 run_write_append_truncate() {
715 write_THREADS=${write_THREADS:-8}
716 write_REP=${write_REP:-10000}
718 if [ "$NFSCLIENT" ]; then
719 skip "skipped for NFSCLIENT mode"
723 # location is lustre/tests dir
724 if ! which write_append_truncate > /dev/null 2>&1 ; then
725 skip_env "write_append_truncate not found"
730 # Need space estimation here.
732 local testdir=$DIR/d0.write_append_truncate
733 local file=$testdir/f0.wat
735 print_opts clients write_REP write_THREADS MACHINEFILE
738 # mpi_run uses mpiuser
741 local cmd="write_append_truncate -n $write_REP $file"
744 mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \
745 -np $((num_clients * $write_THREADS)) $cmd
748 if [ $rc != 0 ] ; then
749 error "write_append_truncate failed! $rc"
755 run_write_disjoint() {
757 WRITE_DISJOINT=${WRITE_DISJOINT:-$(which write_disjoint \
758 2> /dev/null || true)}
760 wdisjoint_THREADS=${wdisjoint_THREADS:-4}
761 wdisjoint_REP=${wdisjoint_REP:-10000}
763 if [ "$NFSCLIENT" ]; then
764 skip "skipped for NFSCLIENT mode"
768 [ x$WRITE_DISJOINT = x ] &&
769 { skip_env "write_disjoint not found" && return; }
772 # Need space estimation here.
774 print_opts WRITE_DISJOINT clients wdisjoint_THREADS wdisjoint_REP \
776 local testdir=$DIR/d0.write_disjoint
778 # mpi_run uses mpiuser
781 local cmd="$WRITE_DISJOINT -f $testdir/file -n $wdisjoint_REP"
784 mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \
785 -np $((num_clients * $wdisjoint_THREADS)) $cmd
788 if [ $rc != 0 ] ; then
789 error "write_disjoint failed! $rc"
794 run_parallel_grouplock() {
796 PARALLEL_GROUPLOCK=${PARALLEL_GROUPLOCK:-$(which parallel_grouplock \
797 2> /dev/null || true)}
798 parallel_grouplock_MINTASKS=${parallel_grouplock_MINTASKS:-5}
800 if [ "$NFSCLIENT" ]; then
801 skip "skipped for NFSCLIENT mode"
805 [ x$PARALLEL_GROUPLOCK = x ] &&
806 { skip "PARALLEL_GROUPLOCK not found" && return; }
808 print_opts clients parallel_grouplock_MINTASKS MACHINEFILE
810 local testdir=$DIR/d0.parallel_grouplock
812 # mpi_run uses mpiuser
818 for i in $(seq 12); do
820 local cmd="$PARALLEL_GROUPLOCK -g -v -d $testdir $subtest"
823 mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \
824 -np $parallel_grouplock_MINTASKS $cmd
826 if [ $rc != 0 ] ; then
827 error_noexit "parallel_grouplock subtests $subtest " \
830 echo "parallel_grouplock subtests $subtest PASS"
832 let status=$((status + rc))
833 # clear debug to collect one log per one test
834 do_nodes $(comma_list $(nodes_list)) lctl clear
836 [ $status -eq 0 ] || error "parallel_grouplock status: $status"
840 cleanup_statahead () {
847 for i in $(seq 0 $num_mntpts);do
848 zconf_umount_clients $clients ${mntpt_root}$i ||
849 error_exit "Failed to umount lustre on ${mntpt_root}$i"
855 statahead_NUMMNTPTS=${statahead_NUMMNTPTS:-5}
856 statahead_NUMFILES=${statahead_NUMFILES:-500000}
858 if [[ -n $NFSCLIENT ]]; then
859 skip "Statahead testing is not supported on NFS clients."
864 { skip_env "mdsrate not found" && return; }
866 print_opts MDSRATE clients statahead_NUMMNTPTS statahead_NUMFILES
870 # do not use default "d[0-9]*" dir name
871 # to avoid of rm $statahead_NUMFILES (500k) files in t-f cleanup
873 local testdir=$DIR/$dir
875 # cleanup only if dir exists
876 # cleanup only $statahead_NUMFILES number of files
877 # ignore the other files created by someone else
879 mdsrate_cleanup $((num_clients * 32)) $MACHINEFILE \
880 $statahead_NUMFILES $testdir 'f%%d' --ignore
883 # mpi_run uses mpiuser
886 local num_files=$statahead_NUMFILES
888 local IFree=$(inodes_available)
889 if [ $IFree -lt $num_files ]; then
895 local cmd1="${MDSRATE} ${MDSRATE_DEBUG} --mknod --dir $testdir"
896 local cmd2="--nfiles $num_files --filefmt 'f%%d'"
897 local cmd="$cmd1 $cmd2"
900 mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \
901 -np $((num_clients * 32)) $cmd
904 if [ $rc != 0 ] ; then
905 error "mdsrate failed to create $rc"
909 local num_mntpts=$statahead_NUMMNTPTS
910 local mntpt_root=$TMP/mntpt/lustre
911 local mntopts=$MNTOPTSTATAHEAD
913 echo "Mounting $num_mntpts lustre clients starts on $clients"
914 trap "cleanup_statahead $clients $mntpt_root $num_mntpts" EXIT ERR
915 for i in $(seq 0 $num_mntpts); do
916 zconf_mount_clients $clients ${mntpt_root}$i "$mntopts" ||
917 error_exit "Failed to mount lustre on ${mntpt_root}$i on $clients"
920 do_rpc_nodes $clients cancel_lru_locks mdc
922 do_rpc_nodes $clients do_ls $mntpt_root $num_mntpts $dir
924 mdsrate_cleanup $((num_clients * 32)) $MACHINEFILE \
925 $num_files $testdir 'f%%d' --ignore
927 # use rm instead of rmdir because of
928 # testdir could contain the files created by someone else,
929 # or by previous run where is num_files prev > num_files current
931 cleanup_statahead $clients $mntpt_root $num_mntpts