X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Ftests%2Ffunctions.sh;h=1bd63841c81cf92de8726f0e63be3c3501e992fa;hb=c8790ae52393d96fc71aa4edea0e5051c54431e9;hp=cada840c457e750680c472f2592d12ee0a88e1d8;hpb=ad0a085d95f80fa16a0c96fddd03577d499ab903;p=fs%2Flustre-release.git diff --git a/lustre/tests/functions.sh b/lustre/tests/functions.sh index cada840..1bd6384 100644 --- a/lustre/tests/functions.sh +++ b/lustre/tests/functions.sh @@ -1,6 +1,4 @@ #!/bin/bash -# -*- mode: Bash; tab-width: 4; indent-tabs-mode: t; -*- -# vim:shiftwidth=4:softtabstop=4:tabstop=4: # Simple function used by run_*.sh scripts @@ -171,7 +169,7 @@ signaled() { } mpi_run () { - local mpirun="$MPIRUN $MPIRUN_OPTIONS" + local mpirun="$MPIRUN $MPIRUN_OPTIONS --oversubscribe" local command="$mpirun $@" local mpilog=$TMP/mpi.log local rc @@ -285,40 +283,38 @@ print_opts () { } run_compilebench() { - # Space estimation: - # compile dir kernel-0 ~1GB - # required space ~1GB * cbench_IDIRS - local dir=${1:-$DIR} + local cbench_DIR=${cbench_DIR:-""} + local cbench_IDIRS=${cbench_IDIRS:-2} + local cbench_RUNS=${cbench_RUNS:-2} - cbench_DIR=${cbench_DIR:-""} - cbench_IDIRS=${cbench_IDIRS:-2} - cbench_RUNS=${cbench_RUNS:-2} + print_opts cbench_DIR cbench_IDIRS cbench_RUNS - print_opts cbench_DIR cbench_IDIRS cbench_RUNS + [ x$cbench_DIR = x ] && + skip_env "compilebench not found" - [ x$cbench_DIR = x ] && - { skip_env "compilebench not found" && return; } - - [ -e $cbench_DIR/compilebench ] || \ - { skip_env "No compilebench build" && return; } + [ -e $cbench_DIR/compilebench ] || + skip_env "No compilebench build" + # Space estimation: + # compile dir kernel-0 ~1GB + # required space ~1GB * cbench_IDIRS local space=$(df -P $dir | tail -n 1 | awk '{ print $4 }') if [[ $space -le $((1024 * 1024 * cbench_IDIRS)) ]]; then cbench_IDIRS=$((space / 1024 / 1024)) [[ $cbench_IDIRS -eq 0 ]] && - skip_env "Need free space at least 1GB, have $space" && - return + skip_env "Need free space at least 1GB, have $space" - echo "free space=$space, reducing initial dirs to $cbench_IDIRS" + echo "reducing initial dirs to $cbench_IDIRS" fi + echo "free space = $space KB" - # FIXME: - # t-f _base needs to be modifyed to set properly tdir - # for new "test_foo" functions names - # local testdir=$DIR/$tdir - local testdir=$dir/d0.compilebench - mkdir -p $testdir + # FIXME: + # t-f _base needs to be modifyed to set properly tdir + # for new "test_foo" functions names + # local testdir=$DIR/$tdir + local testdir=$dir/d0.compilebench.$$ + mkdir -p $testdir local savePWD=$PWD cd $cbench_DIR @@ -337,32 +333,30 @@ run_compilebench() { } run_metabench() { - - METABENCH=${METABENCH:-$(which metabench 2> /dev/null || true)} - mbench_NFILES=${mbench_NFILES:-30400} - # threads per client - mbench_THREADS=${mbench_THREADS:-4} + local dir=${1:-$DIR} + local mntpt=${2:-$MOUNT} + METABENCH=${METABENCH:-$(which metabench 2> /dev/null || true)} + mbench_NFILES=${mbench_NFILES:-30400} + # threads per client + mbench_THREADS=${mbench_THREADS:-4} mbench_OPTIONS=${mbench_OPTIONS:-} + mbench_CLEANUP=${mbench_CLEANUP:-true} - [ x$METABENCH = x ] && - { skip_env "metabench not found" && return; } + [ x$METABENCH = x ] && skip_env "metabench not found" - # FIXME - # Need space estimation here. - - print_opts METABENCH clients mbench_NFILES mbench_THREADS + print_opts METABENCH clients mbench_NFILES mbench_THREADS - local testdir=$DIR/d0.metabench - mkdir -p $testdir - # mpi_run uses mpiuser - chmod 0777 $testdir + local testdir=$dir/d0.metabench + mkdir -p $testdir + # mpi_run uses mpiuser + chmod 0777 $testdir - # -C Run the file creation tests. - # -S Run the file stat tests. - # -c nfile Number of files to be used in each test. - # -k Cleanup. Remove the test directories. - local cmd="$METABENCH -w $testdir -c $mbench_NFILES -C -S -k $mbench_OPTIONS" - echo "+ $cmd" + # -C Run the file creation tests. Creates zero byte files. + # -S Run the file stat tests. + # -c nfile Number of files to be used in each test. + # -k Cleanup files when finished. + local cmd="$METABENCH -w $testdir -c $mbench_NFILES -C -S $mbench_OPTIONS" + echo "+ $cmd" # find out if we need to use srun by checking $SRUN_PARTITION if [ "$SRUN_PARTITION" ]; then @@ -374,42 +368,41 @@ run_metabench() { -np $((num_clients * $mbench_THREADS)) $cmd fi - local rc=$? - if [ $rc != 0 ] ; then - error "metabench failed! $rc" - fi - rm -rf $testdir + local rc=$? + if [ $rc != 0 ] ; then + error "metabench failed! $rc" + fi + + if $mbench_CLEANUP; then + rm -rf $testdir + else + mv $dir/d0.metabench $mntpt/_xxx.$(date +%s).d0.metabench + fi } run_simul() { + SIMUL=${SIMUL:=$(which simul 2> /dev/null || true)} + [ x$SIMUL = x ] && skip_env "simul not found" + [ "$NFSCLIENT" ] && skip "skipped for NFSCLIENT mode" - SIMUL=${SIMUL:=$(which simul 2> /dev/null || true)} - # threads per client - simul_THREADS=${simul_THREADS:-2} - simul_REP=${simul_REP:-20} + # threads per client + simul_THREADS=${simul_THREADS:-2} + simul_REP=${simul_REP:-20} - if [ "$NFSCLIENT" ]; then - skip "skipped for NFSCLIENT mode" - return - fi + # FIXME + # Need space estimation here. - [ x$SIMUL = x ] && - { skip_env "simul not found" && return; } + print_opts SIMUL clients simul_REP simul_THREADS - # FIXME - # Need space estimation here. - - print_opts SIMUL clients simul_REP simul_THREADS - - local testdir=$DIR/d0.simul - mkdir -p $testdir - # mpi_run uses mpiuser - chmod 0777 $testdir + local testdir=$DIR/d0.simul + mkdir -p $testdir + # mpi_run uses mpiuser + chmod 0777 $testdir - # -n # : repeat each test # times - # -N # : repeat the entire set of tests # times + # -n # : repeat each test # times + # -N # : repeat the entire set of tests # times - local cmd="$SIMUL -d $testdir -n $simul_REP -N $simul_REP" + local cmd="$SIMUL -d $testdir -n $simul_REP -N $simul_REP" echo "+ $cmd" # find out if we need to use srun by checking $SRUN_PARTITION @@ -430,40 +423,33 @@ run_simul() { } run_mdtest() { + MDTEST=${MDTEST:=$(which mdtest 2> /dev/null || true)} + [ x$MDTEST = x ] && skip_env "mdtest not found" + [ "$NFSCLIENT" ] && skip "skipped for NFSCLIENT mode" - MDTEST=${MDTEST:=$(which mdtest 2> /dev/null || true)} - # threads per client - mdtest_THREADS=${mdtest_THREADS:-2} - mdtest_nFiles=${mdtest_nFiles:-"100000"} - # We devide the files by number of core - mdtest_nFiles=$((mdtest_nFiles/mdtest_THREADS/num_clients)) - mdtest_iteration=${mdtest_iteration:-1} + # threads per client + mdtest_THREADS=${mdtest_THREADS:-2} + mdtest_nFiles=${mdtest_nFiles:-"100000"} + # We devide the files by number of core + mdtest_nFiles=$((mdtest_nFiles/mdtest_THREADS/num_clients)) + mdtest_iteration=${mdtest_iteration:-1} local mdtest_custom_params=${mdtest_custom_params:-""} + local type=${1:-"ssf"} - local type=${1:-"ssf"} + # FIXME + # Need space estimation here. - if [ "$NFSCLIENT" ]; then - skip "skipped for NFSCLIENT mode" - return - fi - - [ x$MDTEST = x ] && - { skip_env "mdtest not found" && return; } + print_opts MDTEST mdtest_iteration mdtest_THREADS mdtest_nFiles - # FIXME - # Need space estimation here. - - print_opts MDTEST mdtest_iteration mdtest_THREADS mdtest_nFiles - - local testdir=$DIR/d0.mdtest - mkdir -p $testdir - # mpi_run uses mpiuser - chmod 0777 $testdir + local testdir=$DIR/d0.mdtest + mkdir -p $testdir + # mpi_run uses mpiuser + chmod 0777 $testdir - # -i # : repeat each test # times - # -d : test dir - # -n # : number of file/dir to create/stat/remove - # -u : each process create/stat/remove individually + # -i # : repeat each test # times + # -d : test dir + # -n # : number of file/dir to create/stat/remove + # -u : each process create/stat/remove individually local cmd="$MDTEST -d $testdir -i $mdtest_iteration \ -n $mdtest_nFiles $mdtest_custom_params" @@ -489,67 +475,85 @@ run_mdtest() { } run_connectathon() { + local dir=${1:-$DIR} + cnt_DIR=${cnt_DIR:-""} + cnt_NRUN=${cnt_NRUN:-10} - cnt_DIR=${cnt_DIR:-""} - cnt_NRUN=${cnt_NRUN:-10} + print_opts cnt_DIR cnt_NRUN - print_opts cnt_DIR cnt_NRUN + [ x$cnt_DIR = x ] && skip_env "connectathon dir not found" + [ -e $cnt_DIR/runtests ] || skip_env "No connectathon runtests found" - [ x$cnt_DIR = x ] && - { skip_env "connectathon dir not found" && return; } + # Space estimation: + # "special" tests create a 30 MB file + misc. small files + # required space ~40 MB + local space=$(df -P $dir | tail -n 1 | awk '{ print $4 }') + if [[ $space -le $((1024 * 40)) ]]; then + skip_env "Need free space at least 40MB, have $space KB" + fi + echo "free space = $space KB" - [ -e $cnt_DIR/runtests ] || \ - { skip_env "No connectathon runtests found" && return; } + local testdir=$dir/d0.connectathon + mkdir -p $testdir - local testdir=$DIR/d0.connectathon - mkdir -p $testdir + local savePWD=$PWD + cd $cnt_DIR - local savePWD=$PWD - cd $cnt_DIR - - # - # cthon options (must be in this order) - # - # -N numpasses - will be passed to the runtests script. This argument - # is optional. It specifies the number of times to run - # through the tests. - # - # One of these test types - # -b basic - # -g general - # -s special - # -l lock - # -a all of the above - # - # -f a quick functionality test - # - - tests="-b -g -s" - # Include lock tests unless we're running on nfsv4 - local fstype=$(df -TP $testdir | awk 'NR==2 {print $2}') - echo "$testdir: $fstype" - if [[ $fstype != "nfs4" ]]; then - tests="$tests -l" - fi - echo "tests: $tests" - for test in $tests; do - local cmd="./runtests -N $cnt_NRUN $test -f $testdir" - local rc=0 - - log "$cmd" - eval $cmd - rc=$? - [ $rc = 0 ] || error "connectathon failed: $rc" - done + # + # To run connectathon: + # runtests [-a|-b|-g|-s|-l] [-f|-n|-t] [-N numpasses] [test-directory] + # + # One of the following test types + # -b basic + # -g general + # -s special + # -l lock + # -a all of the above + # + # -f a quick functional test + # -n suppress directory operations (mkdir and rmdir) + # -t run with time statistics (default for basic tests) + # + # -N numpasses - specifies the number of times to run + # the tests. Optional. + + tests="-b -g -s" + # Include lock tests unless we're running on nfsv4 + local fstype=$(df -TP $testdir | awk 'NR==2 {print $2}') + echo "$testdir: $fstype" + if [[ $fstype != "nfs4" ]]; then + tests="$tests -l" + fi + echo "tests: $tests" + for test in $tests; do + local cmd="sh ./runtests -N $cnt_NRUN $test -f $testdir" + local rc=0 + + log "$cmd" + eval $cmd + rc=$? + [ $rc = 0 ] || error "connectathon failed: $rc" + done - cd $savePWD - rm -rf $testdir + cd $savePWD + rm -rf $testdir } run_ior() { local type=${1:="ssf"} + local dir=${2:-$DIR} + local testdir=$dir/d0.ior.$type + local nfs_srvmntpt=$3 + + if [ "$NFSCLIENT" ]; then + [[ -n $nfs_srvmntpt ]] || + { error "NFSCLIENT mode, but nfs exported dir"\ + "is not set!" && return 1; } + fi IOR=${IOR:-$(which IOR 2> /dev/null || true)} + [ x$IOR = x ] && skip_env "IOR not found" + # threads per client ior_THREADS=${ior_THREADS:-2} ior_iteration=${ior_iteration:-1} @@ -573,38 +577,32 @@ run_ior() { ;; esac - [ x$IOR = x ] && - { skip_env "IOR not found" && return; } - # calculate the space in bytes - local space=$(df -B 1 -P $DIR | tail -n 1 | awk '{ print $4 }') + local space=$(df -B 1 -P $dir | tail -n 1 | awk '{ print $4 }') local total_threads=$((num_clients * ior_THREADS)) echo "+ $ior_blockSize * $multiplier * $total_threads " if [ $((space / 2)) -le \ $((ior_blockSize * multiplier * total_threads)) ]; then ior_blockSize=$((space / 2 / multiplier / total_threads)) - [ $ior_blockSize -eq 0 ] && \ + [ $ior_blockSize -eq 0 ] && skip_env "Need free space more than $((2 * total_threads)) \ - ${ior_blockUnit}: have $((space / multiplier))" && - return + ${ior_blockUnit}: have $((space / multiplier))" echo "(reduced blockSize to $ior_blockSize \ ${ior_blockUnit} bytes)" fi - print_opts IOR ior_THREADS ior_DURATION MACHINEFILE + print_opts IOR ior_THREADS ior_DURATION MACHINEFILE + + mkdir -p $testdir + # mpi_run uses mpiuser + chmod 0777 $testdir + if [ -z "$NFSCLIENT" ]; then + ior_stripe_params=${ior_stripe_params:-"-c -1"} + $LFS setstripe $testdir $ior_stripe_params || + { error "setstripe failed" && return 2; } + fi - local testdir=$DIR/d0.ior.$type - mkdir -p $testdir - # mpi_run uses mpiuser - chmod 0777 $testdir - if [ "$NFSCLIENT" ]; then - setstripe_nfsserver $testdir -c -1 || - { error "setstripe on nfsserver failed" && return 1; } - else - $LFS setstripe $testdir -c -1 || - { error "setstripe failed" && return 2; } - fi # # -b N blockSize -- # contiguous bytes to write per task (e.g.: 8, 4K, 2M, 1G)" @@ -617,9 +615,14 @@ run_ior() { # -T maxTimeDuration -- max time in minutes to run tests" # -k keepFile -- keep testFile(s) on program exit - local cmd="$IOR -a $ior_type -b ${ior_blockSize}${ior_blockUnit} \ + local cmd + if [ -n "$ior_custom_params" ]; then + cmd="$IOR $ior_custom_params -o $testdir/iorData" + else + cmd="$IOR -a $ior_type -b ${ior_blockSize}${ior_blockUnit} \ -o $testdir/iorData -t $ior_xferSize -v -C -w -r -W \ -i $ior_iteration -T $ior_DURATION -k" + fi [ $type = "fpp" ] && cmd="$cmd -F" @@ -630,8 +633,9 @@ run_ior() { -n $((num_clients * ior_THREADS)) -p $SRUN_PARTITION \ -- $cmd else + mpi_ior_custom_threads=${mpi_ior_custom_threads:-"$((num_clients * ior_THREADS))"} mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \ - -np $((num_clients * $ior_THREADS)) $cmd + -np $mpi_ior_custom_threads $cmd fi local rc=$? @@ -642,40 +646,34 @@ run_ior() { } run_mib() { + MIB=${MIB:=$(which mib 2> /dev/null || true)} + [ "$NFSCLIENT" ] && skip "skipped for NFSCLIENT mode" + [ x$MIB = x ] && skip_env "MIB not found" - MIB=${MIB:=$(which mib 2> /dev/null || true)} - # threads per client - mib_THREADS=${mib_THREADS:-2} - mib_xferSize=${mib_xferSize:-1m} - mib_xferLimit=${mib_xferLimit:-5000} - mib_timeLimit=${mib_timeLimit:-300} - - if [ "$NFSCLIENT" ]; then - skip "skipped for NFSCLIENT mode" - return - fi - - [ x$MIB = x ] && - { skip_env "MIB not found" && return; } + # threads per client + mib_THREADS=${mib_THREADS:-2} + mib_xferSize=${mib_xferSize:-1m} + mib_xferLimit=${mib_xferLimit:-5000} + mib_timeLimit=${mib_timeLimit:-300} - print_opts MIB mib_THREADS mib_xferSize mib_xferLimit mib_timeLimit \ - MACHINEFILE + print_opts MIB mib_THREADS mib_xferSize mib_xferLimit mib_timeLimit \ + MACHINEFILE - local testdir=$DIR/d0.mib - mkdir -p $testdir - # mpi_run uses mpiuser - chmod 0777 $testdir - $LFS setstripe $testdir -c -1 || - { error "setstripe failed" && return 2; } - # - # -I Show intermediate values in output - # -H Show headers in output - # -L Do not issue new system calls after this many seconds - # -s Use system calls of this size - # -t test dir - # -l Issue no more than this many system calls - local cmd="$MIB -t $testdir -s $mib_xferSize -l $mib_xferLimit \ - -L $mib_timeLimit -HI -p mib.$(date +%Y%m%d%H%M%S)" + local testdir=$DIR/d0.mib + mkdir -p $testdir + # mpi_run uses mpiuser + chmod 0777 $testdir + $LFS setstripe $testdir -c -1 || + error "setstripe failed" + # + # -I Show intermediate values in output + # -H Show headers in output + # -L Do not issue new system calls after this many seconds + # -s Use system calls of this size + # -t test dir + # -l Issue no more than this many system calls + local cmd="$MIB -t $testdir -s $mib_xferSize -l $mib_xferLimit \ + -L $mib_timeLimit -HI -p mib.$(date +%Y%m%d%H%M%S)" echo "+ $cmd" # find out if we need to use srun by checking $SRUN_PARTITION @@ -696,34 +694,28 @@ run_mib() { } run_cascading_rw() { + CASC_RW=${CASC_RW:-$(which cascading_rw 2> /dev/null || true)} + [ x$CASC_RW = x ] && skip_env "cascading_rw not found" + [ "$NFSCLIENT" ] && skip "skipped for NFSCLIENT mode" - CASC_RW=${CASC_RW:-$(which cascading_rw 2> /dev/null || true)} - # threads per client - casc_THREADS=${casc_THREADS:-2} - casc_REP=${casc_REP:-300} - - if [ "$NFSCLIENT" ]; then - skip "skipped for NFSCLIENT mode" - return - fi - - [ x$CASC_RW = x ] && - { skip_env "cascading_rw not found" && return; } + # threads per client + casc_THREADS=${casc_THREADS:-2} + casc_REP=${casc_REP:-300} - # FIXME - # Need space estimation here. + # FIXME + # Need space estimation here. - print_opts CASC_RW clients casc_THREADS casc_REP MACHINEFILE + print_opts CASC_RW clients casc_THREADS casc_REP MACHINEFILE - local testdir=$DIR/d0.cascading_rw - mkdir -p $testdir - # mpi_run uses mpiuser - chmod 0777 $testdir + local testdir=$DIR/d0.cascading_rw + mkdir -p $testdir + # mpi_run uses mpiuser + chmod 0777 $testdir - # -g: debug mode - # -n: repeat test # times + # -g: debug mode + # -n: repeat test # times - local cmd="$CASC_RW -g -d $testdir -n $casc_REP" + local cmd="$CASC_RW -g -d $testdir -n $casc_REP" echo "+ $cmd" mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \ @@ -737,35 +729,29 @@ run_cascading_rw() { } run_write_append_truncate() { + [ "$NFSCLIENT" ] && skip "skipped for NFSCLIENT mode" + # location is lustre/tests dir + if ! which write_append_truncate > /dev/null 2>&1 ; then + skip_env "write_append_truncate not found" + fi - # threads per client - write_THREADS=${write_THREADS:-8} - write_REP=${write_REP:-10000} - - if [ "$NFSCLIENT" ]; then - skip "skipped for NFSCLIENT mode" - return - fi - - # location is lustre/tests dir - if ! which write_append_truncate > /dev/null 2>&1 ; then - skip_env "write_append_truncate not found" - return - fi + # threads per client + write_THREADS=${write_THREADS:-8} + write_REP=${write_REP:-10000} - # FIXME - # Need space estimation here. + # FIXME + # Need space estimation here. - local testdir=$DIR/d0.write_append_truncate - local file=$testdir/f0.wat + local testdir=$DIR/d0.write_append_truncate + local file=$testdir/f0.wat - print_opts clients write_REP write_THREADS MACHINEFILE + print_opts clients write_REP write_THREADS MACHINEFILE - mkdir -p $testdir - # mpi_run uses mpiuser - chmod 0777 $testdir + mkdir -p $testdir + # mpi_run uses mpiuser + chmod 0777 $testdir - local cmd="write_append_truncate -n $write_REP $file" + local cmd="write_append_truncate -n $write_REP $file" echo "+ $cmd" mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \ @@ -780,20 +766,15 @@ run_write_append_truncate() { } run_write_disjoint() { + WRITE_DISJOINT=${WRITE_DISJOINT:-$(which write_disjoint 2> /dev/null || + true)} + [ x$WRITE_DISJOINT = x ] && skip_env "write_disjoint not found" + [ "$NFSCLIENT" ] && skip "skipped for NFSCLIENT mode" - WRITE_DISJOINT=${WRITE_DISJOINT:-$(which write_disjoint \ - 2> /dev/null || true)} - # threads per client - wdisjoint_THREADS=${wdisjoint_THREADS:-4} - wdisjoint_REP=${wdisjoint_REP:-10000} - - if [ "$NFSCLIENT" ]; then - skip "skipped for NFSCLIENT mode" - return - fi - - [ x$WRITE_DISJOINT = x ] && - { skip_env "write_disjoint not found" && return; } + # threads per client + wdisjoint_THREADS=${wdisjoint_THREADS:-4} + wdisjoint_REP=${wdisjoint_REP:-10000} + chunk_size_limit=$1 # FIXME # Need space estimation here. @@ -805,7 +786,8 @@ run_write_disjoint() { # mpi_run uses mpiuser chmod 0777 $testdir - local cmd="$WRITE_DISJOINT -f $testdir/file -n $wdisjoint_REP" + local cmd="$WRITE_DISJOINT -f $testdir/file -n $wdisjoint_REP -m \ + $chunk_size_limit" echo "+ $cmd" mpi_run ${MACHINEFILE_OPTION} ${MACHINEFILE} \ @@ -819,29 +801,24 @@ run_write_disjoint() { } run_parallel_grouplock() { + PARALLEL_GROUPLOCK=${PARALLEL_GROUPLOCK:-$(which parallel_grouplock \ + 2> /dev/null || true)} - PARALLEL_GROUPLOCK=${PARALLEL_GROUPLOCK:-$(which parallel_grouplock \ - 2> /dev/null || true)} - parallel_grouplock_MINTASKS=${parallel_grouplock_MINTASKS:-5} + [ x$PARALLEL_GROUPLOCK = x ] && skip "PARALLEL_GROUPLOCK not found" + [ "$NFSCLIENT" ] && skip "skipped for NFSCLIENT mode" - if [ "$NFSCLIENT" ]; then - skip "skipped for NFSCLIENT mode" - return - fi - - [ x$PARALLEL_GROUPLOCK = x ] && - { skip "PARALLEL_GROUPLOCK not found" && return; } + parallel_grouplock_MINTASKS=${parallel_grouplock_MINTASKS:-5} - print_opts clients parallel_grouplock_MINTASKS MACHINEFILE + print_opts clients parallel_grouplock_MINTASKS MACHINEFILE - local testdir=$DIR/d0.parallel_grouplock - mkdir -p $testdir - # mpi_run uses mpiuser - chmod 0777 $testdir + local testdir=$DIR/d0.parallel_grouplock + mkdir -p $testdir + # mpi_run uses mpiuser + chmod 0777 $testdir - local cmd - local status=0 - local subtest + local cmd + local status=0 + local subtest for i in $(seq 12); do subtest="-t $i" local cmd="$PARALLEL_GROUPLOCK -g -v -d $testdir $subtest" @@ -878,21 +855,17 @@ cleanup_statahead () { } run_statahead () { + if [[ -n $NFSCLIENT ]]; then + skip "Statahead testing is not supported on NFS clients." + fi + [ x$MDSRATE = x ] && skip_env "mdsrate not found" - statahead_NUMMNTPTS=${statahead_NUMMNTPTS:-5} - statahead_NUMFILES=${statahead_NUMFILES:-500000} - - if [[ -n $NFSCLIENT ]]; then - skip "Statahead testing is not supported on NFS clients." - return 0 - fi - - [ x$MDSRATE = x ] && - { skip_env "mdsrate not found" && return; } + statahead_NUMMNTPTS=${statahead_NUMMNTPTS:-5} + statahead_NUMFILES=${statahead_NUMFILES:-500000} - print_opts MDSRATE clients statahead_NUMMNTPTS statahead_NUMFILES + print_opts MDSRATE clients statahead_NUMMNTPTS statahead_NUMFILES - # create large dir + # create large dir # do not use default "d[0-9]*" dir name # to avoid of rm $statahead_NUMFILES (500k) files in t-f cleanup @@ -958,6 +931,129 @@ run_statahead () { cleanup_statahead $clients $mntpt_root $num_mntpts } +cleanup_rr_alloc () { + trap 0 + local clients="$1" + local mntpt_root="$2" + local rr_alloc_MNTPTS="$3" + local mntpt_dir=$(dirname ${mntpt_root}) + + for i in $(seq 0 $((rr_alloc_MNTPTS - 1))); do + zconf_umount_clients $clients ${mntpt_root}$i || + error_exit "Failed to umount lustre on ${mntpt_root}$i" + done + do_nodes $clients "rm -rf $mntpt_dir" +} + +run_rr_alloc() { + remote_mds_nodsh && skip "remote MDS with nodsh" + echo "===Test gives more reproduction percentage if number of "\ + "client and ost are more. Test with 44 or more clients "\ + "and 73 or more OSTs gives 100% reproduction rate==" + + RR_ALLOC=${RR_ALLOC:-$(which rr_alloc 2> /dev/null || true)} + [ x$RR_ALLOC = x ] && skip_env "rr_alloc not found" + declare -a diff_max_min_arr + # foeo = file on each ost. calc = calculated. + local ost_idx + local foeo_calc + local qos_prec_objs="${TMP}/qos_and_precreated_objects" + local rr_alloc_NFILES=${rr_alloc_NFILES:-555} + local rr_alloc_MNTPTS=${rr_alloc_MNTPTS:-11} + local total_MNTPTS=$((rr_alloc_MNTPTS * num_clients)) + local mntpt_root="${TMP}/rr_alloc_mntpt/lustre" + if [ $MDSCOUNT -lt 2 ]; then + [ -e $DIR/$tdir ] || mkdir -p $DIR/$tdir + else + [ -e $DIR/$tdir ] || $LFS mkdir -i 0 $DIR/$tdir + fi + chmod 0777 $DIR/$tdir + $LFS setstripe -c 1 /$DIR/$tdir + + trap "cleanup_rr_alloc $clients $mntpt_root $rr_alloc_MNTPTS" EXIT ERR + for i in $(seq 0 $((rr_alloc_MNTPTS - 1))); do + zconf_mount_clients $clients ${mntpt_root}$i $MOUNT_OPTS || + error_exit "Failed to mount lustre on ${mntpt_root}$i $clients" + done + + local cmd="$RR_ALLOC $mntpt_root/$tdir/ash $rr_alloc_NFILES \ + $num_clients" + + # Save mdt values, set threshold to 100% i.e always Round Robin, + # restore the saved values again after creating files... + save_lustre_params mds1 \ + "lov.$FSNAME-MDT0000*.qos_threshold_rr" > $qos_prec_objs + save_lustre_params mds1 \ + "osp.$FSNAME-OST*-osc-MDT0000.create_count" >> $qos_prec_objs + + local old_create_count=$(grep -e "create_count" $qos_prec_objs | + cut -d'=' -f 2 | sort -nr | head -n1) + + # Make sure that every osp has enough precreated objects for the file + # creation app + + # create_count is always set to the power of 2 only, so if the files + # per OST are not multiple of that then it will be set to nearest + # lower power of 2. So set 'create_count' to the upper power of 2. + + foeo_calc=$((rr_alloc_NFILES * total_MNTPTS / OSTCOUNT)) + local create_count=$((2 * foeo_calc)) + do_facet mds1 "$LCTL set_param -n \ + lov.$FSNAME-MDT0000*.qos_threshold_rr 100 \ + osp.$FSNAME-OST*-osc-MDT0000.create_count $create_count" || + error "failed while setting qos_threshold_rr & creat_count" + + # Create few temporary files in order to increase the precreated objects + # to a desired value, before starting 'rr_alloc' app. Due to default + # value 32 of precreation count (OST_MIN_PRECREATE=32), precreated + # objects available are 32 initially, these gets exhausted very soon, + # which causes skip of some osps when very large number of files + # is created per OSTs. + createmany -o $DIR/$tdir/foo- $(((old_create_count + 1) * OSTCOUNT)) \ + > /dev/null + rm -f /$DIR/$tdir/foo* + + # Check for enough precreated objects... We should not + # fail here because code(osp_precreate.c) also takes care of it. + # So we have good chances of passing test even if this check fails. + local mdt_idx=0 + for ost_idx in $(seq 0 $((OSTCOUNT - 1))); do + [[ $(precreated_ost_obj_count $mdt_idx $ost_idx) -ge \ + $foeo_calc ]] || echo "Warning: test may fail because" \ + "of lack of precreated objects on OST${ost_idx}" + done + + if [[ $total_MNTPTS -ne 0 ]]; then + # Now start the actual file creation app. + mpi_run "-np $total_MNTPTS" $cmd || return + else + error "No mount point" + fi + + restore_lustre_params < $qos_prec_objs + rm -f $qos_prec_objs + + diff_max_min_arr=($($LFS getstripe -r $DIR/$tdir/ | + grep "lmm_stripe_offset:" | awk '{print $2}' | sort -n | + uniq -c | awk 'NR==1 {min=max=$1} \ + { $1max ? max=$1 : max} \ + END {print max-min, max, min}')) + + rm -rf $DIR/$tdir + + # In-case of fairly large number of file creation using RR (round-robin) + # there can be two cases in which deviation will occur than the regular + # RR algo behaviour- + # 1- When rr_alloc does not start right with 'lqr_start_count' reseeded, + # 2- When rr_alloc does not finish with 'lqr_start_count == 0'. + # So the difference of files b/w any 2 OST should not be more than 2. + [[ ${diff_max_min_arr[0]} -le 2 ]] || + error "Uneven distribution detected: difference between" \ + "maximum files per OST (${diff_max_min_arr[1]}) and" \ + "minimum files per OST (${diff_max_min_arr[2]}) must not be" \ + "greater than 2" +} + run_fs_test() { # fs_test.x is the default name for exe FS_TEST=${FS_TEST:=$(which fs_test.x 2> /dev/null || true)} @@ -979,8 +1075,7 @@ run_fs_test() { fs_test_objunit=${fs_test_objunit:-1048576} # 1 mb fs_test_ndirs=${fs_test_ndirs:-80000} - [ x$FS_TEST = x ] && - { skip "FS_TEST not found" && return; } + [ x$FS_TEST = x ] && skip "FS_TEST not found" # Space estimation in bytes local space=$(df -B 1 -P $dir | tail -n 1 | awk '{ print $4 }') @@ -990,11 +1085,10 @@ run_fs_test() { $((fs_test_objsize * fs_test_objunit * total_threads)) ]; then fs_test_objsize=$((space / 2 / fs_test_objunit / \ total_threads)) - [ $fs_test_objsize -eq 0 ] && \ + [ $fs_test_objsize -eq 0 ] && skip_env "Need free space more than \ $((2 * total_threads * fs_test_objunit)) \ - : have $((space / fs_test_objunit))" && - return + : have $((space / fs_test_objunit))" echo "(reduced objsize to \ $((fs_test_objsize * fs_test_objunit)) bytes)" @@ -1055,3 +1149,121 @@ ior_mdtest_parallel() { [[ $rc1 -ne 0 || $rc2 -ne 0 ]] && return 1 return 0 } + +run_fio() { + FIO=${FIO:=$(which fio 2> /dev/null || true)} + + local clients=${CLIENTS:-$(hostname)} + local fio_jobNum=${fio_jobNum:-4} + local fio_jobFile=${fio_jobFile:-$TMP/fiojobfile.$(date +%s)} + local fio_bs=${fio_bs:-1} + local testdir=$DIR/d0.fio + local file=${testdir}/fio + local runtime=60 + local propagate=false + + [ "$SLOW" = "no" ] || runtime=600 + + [ x$FIO = x ] && skip_env "FIO not found" + + mkdir -p $testdir + + # use fio job file if exists, + # create a simple one if missing + if ! [ -f $fio_jobFile ]; then + cat >> $fio_jobFile <> $fio_jobFile < ${fio_jobFile}" || + error "job file $fio_jobFile is not propagated" + do_nodesv $clients "cat ${fio_jobFile}" + fi + + cmd="$FIO $fio_jobFile" + echo "+ $cmd" + + log "clients: $clients $cmd" + + local rc=0 + do_nodesv $clients "$cmd " + rc=$? + + [ $rc = 0 ] || error "fio failed: $rc" + rm -rf $testdir +} + +run_xdd() { + XDD=${XDD:=$(which xdd 2> /dev/null || true)} + + local clients=${CLIENTS:-$(hostname)} + local testdir=$DIR/d0.xdd + xdd_queuedepth=${xdd_queuedepth:-4} + xdd_blocksize=${xdd_blocksize:-512} + xdd_reqsize=${xdd_reqsize:-128} + xdd_mbytes=${xdd_mbytes:-100} + xdd_passes=${xdd_passes:-40} + xdd_rwratio=${xdd_rwratio:-0} + xdd_ntargets=${xdd_ntargets:-6} + local xdd_custom_params=${xdd_custom_params:-"-dio -stoponerror \ + -maxpri -minall -noproclock -nomemlock"} + + [ x$XDD = x ] && skip "XDD not found" + + print_opts XDD clients xdd_queuedepth xdd_blocksize xdd_reqsize \ + xdd_mbytes xdd_passes xdd_rwratio + + mkdir -p $testdir + + local files="" + # Target files creates based on the given number of targets + for (( i=0; i < $xdd_ntargets; i++ )) + do + files+="${testdir}/xdd"$i" " + done + + # -targets specifies the devices or files to perform operation + # -reqsize number of 'blocks' per operation + # -mbytes number of 1024*1024-byte blocks to transfer + # -blocksize size of a single 'block' + # -passes number of times to read mbytes + # -queuedepth number of commands to queue on the target + # -rwratio percentage of read to write operations + # -verbose will print out statistics on each pass + + local cmd="$XDD -targets $xdd_ntargets $files -reqsize $xdd_reqsize \ + -mbytes $xdd_mbytes -blocksize $xdd_blocksize \ + -passes $xdd_passes -queuedepth $xdd_queuedepth \ + -rwratio $xdd_rwratio -verbose $xdd_custom_params" + echo "+ $cmd" + + local rc=0 + do_nodesv $clients "$cmd " + rc=$? + + [ $rc = 0 ] || error "xdd failed: $rc" + + rm -rf $testdir +}