+
+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"
+ test_mkdir $DIR/$tdir
+ setstripe_getstripe $DIR/$tdir $rr_alloc_STRIPEPARAMS
+
+ chmod 0777 $DIR/$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} \
+ { $1<min ? min=$1 : min; $1>max ? 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:=$(which fs_test.x 2> /dev/null || true)}
+
+ local clients=${CLIENTS:-$(hostname)}
+ local testdir=$DIR/d0.fs_test
+ local file=${testdir}/fs_test
+ fs_test_threads=${fs_test_threads:-2}
+ fs_test_type=${fs_test_type:-1}
+ fs_test_nobj=${fs_test_nobj:-10}
+ fs_test_check=${fs_test_check:-3}
+ fs_test_strided=${fs_test_strided:-1}
+ fs_test_touch=${fs_test_touch:-3}
+ fs_test_supersize=${fs_test_supersize:-1}
+ fs_test_op=${fs_test_op:-write}
+ fs_test_barriers=${fs_test_barriers:-bopen,bwrite,bclose}
+ fs_test_io=${fs_test_io:-mpi}
+ fs_test_objsize=${fs_test_objsize:-100}
+ fs_test_objunit=${fs_test_objunit:-1048576} # 1 mb
+ fs_test_ndirs=${fs_test_ndirs:-80000}
+
+ [ x$FS_TEST = x ] && skip "FS_TEST not found"
+
+ # Space estimation in bytes
+ local space=$(df -B 1 -P $dir | tail -n 1 | awk '{ print $4 }')
+ local total_threads=$((num_clients * fs_test_threads))
+ echo "+ $fs_test_objsize * $fs_test_objunit * $total_threads "
+ if [ $((space / 2)) -le \
+ $((fs_test_objsize * fs_test_objunit * total_threads)) ]; then
+ fs_test_objsize=$((space / 2 / fs_test_objunit / \
+ total_threads))
+ [ $fs_test_objsize -eq 0 ] &&
+ skip_env "Need free space more than \
+ $((2 * total_threads * fs_test_objunit)) \
+ : have $((space / fs_test_objunit))"
+
+ echo "(reduced objsize to \
+ $((fs_test_objsize * fs_test_objunit)) bytes)"
+ fi
+
+ print_opts FS_TEST clients fs_test_threads fs_test_objsize MACHINEFILE
+
+ test_mkdir $testdir
+ setstripe_getstripe $testdir $fs_test_STRIPEPARAMS
+
+ # mpi_run uses mpiuser
+ chmod 0777 $testdir
+
+ # --nodb Turn off the database code at runtime
+ # -g --target The path to the data file
+ # -t --type Whether to do N-N (1) or N-1 (2)
+ # -n --nobj The number of objects written/read by each proc
+ # -z --size The size of each object
+ # -d ---num_nn_dirs Number of subdirectories for files
+ # -C --check Check every byte using argument 3.
+ # --collective Whether to use collective I/O (for N-1, mpi-io only)
+ # -s --strided Whether to use a strided pattern (for N-1 only)
+ # -T --touch Touch every byte using argument 3
+ # -o --op Whether to read only (read) or write only (write)
+ # -b --barriers When to barrier.
+ # -i --io Use POSIX, MPI, or PLFS IO routines (mpi|posix|plfs)
+ # -S --supersize Specify how many objects per superblock
+
+ local cmd="$FS_TEST -nodb -g $file -t $fs_test_type -n $fs_test_nobj \
+ -z $((fs_test_objsize * fs_test_objunit)) -d $fs_test_ndirs \
+ -C $fs_test_check -collective -s $fs_test_strided \
+ -T $fs_test_touch -o $fs_test_op -b $fs_test_barriers \
+ -i $fs_test_io -S $fs_test_supersize"
+
+ echo "+ $cmd"
+ mpi_run "-np $((num_clients * fs_test_threads))" $cmd
+
+ local rc=$?
+ if [ $rc != 0 ] ; then
+ error "fs_test failed! $rc"
+ fi
+
+ rm -rf $testdir
+}
+
+ior_mdtest_parallel() {
+ local rc1=0
+ local rc2=0
+ local type=$1
+
+ run_ior $type &
+ local pids=$!
+
+ run_mdtest $type || rc2=$?
+ [[ $rc2 -ne 0 ]] && echo "mdtest failed with error $rc2"
+
+ wait $pids || rc1=$?
+ [[ $rc1 -ne 0 ]] && echo "ior failed with error $rc1"
+
+ [[ $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"
+
+ test_mkdir $testdir
+ setstripe_getstripe $testdir $fio_STRIPEPARAMS
+
+ # use fio job file if exists,
+ # create a simple one if missing
+ if ! [ -f $fio_jobFile ]; then
+ cat >> $fio_jobFile <<EOF
+[global]
+rw=randwrite
+size=128m
+time_based=1
+runtime=$runtime
+filename=${file}_\$(hostname)
+EOF
+ # bs size increased by $i for each job
+ for ((i=1; i<=fio_jobNum; i++)); do
+ cat >> $fio_jobFile <<EOF
+
+[job$i]
+bs=$(( fio_bs * i ))m
+EOF
+ done
+ # job file is created, should be propagated to all clients
+ propagate=true
+ fi
+
+
+ # propagate the job file if not all clients have it yet or
+ # if the job file was created during the test run
+ if ! do_nodesv $clients " [ -f $fio_jobFile ] " ||
+ $propagate; then
+ local cfg=$(cat $fio_jobFile)
+ do_nodes $clients "echo \\\"$cfg\\\" > ${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
+
+ test_mkdir $testdir
+ setstripe_getstripe $testdir $xdd_STRIPEPARAMS
+
+ 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
+}