LU-8865 tests: add fs_test test

Patch adds parallel-scale fs_test test.

Formerly it is MPI-IO test and it provides I/O performance
results for Effective bandwidths. We used fs_test to generate
a maximum IO write and read bandwidth scenario where a lot of
data is written to amortize away the overhead of open/sync/close
operations.

Test-Parameters: trivial testlist=parallel-scale
Seagate-bug-id: MRP-3914
Signed-off-by: Chennaiah Palla <chennaiah.palla@seagate.com>
Change-Id: I6057c8269fa72a151a792dcf1d05d30f4882204d
Reviewed-on: https://review.whamcloud.com/23932
Tested-by: Jenkins
Tested-by: Maloo <hpdd-maloo@intel.com>
Reviewed-by: Andreas Dilger <andreas.dilger@intel.com>
Reviewed-by: Elena Gryaznova <elena.gryaznova@seagate.com>
Reviewed-by: Oleg Drokin <oleg.drokin@intel.com>

diff --git a/lustre/tests/functions.sh b/lustre/tests/functions.sh
index af268ca..01f3741 100644 (file)
--- a/lustre/tests/functions.sh
+++ b/lustre/tests/functions.sh
@@ -957,3 +957,83 @@ run_statahead () {
     rm -rf $testdir
     cleanup_statahead $clients $mntpt_root $num_mntpts
 }
     rm -rf $testdir
     cleanup_statahead $clients $mntpt_root $num_mntpts
 }

+
+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" && return; }
+
+       # 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))" &&
+                               return
+
+                       echo "(reduced objsize to \
+                               $((fs_test_objsize * fs_test_objunit)) bytes)"
+       fi
+
+       print_opts FS_TEST clients fs_test_threads fs_test_objsize MACHINEFILE
+
+       mkdir -p $testdir
+       # 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
+}

diff --git a/lustre/tests/parallel-scale.sh b/lustre/tests/parallel-scale.sh
index 5f138c8..4530b3e 100644 (file)
--- a/lustre/tests/parallel-scale.sh
+++ b/lustre/tests/parallel-scale.sh
@@ -56,6 +56,12 @@ fi
 # write_disjoint
 [ "$SLOW" = "no" ] && wdisjoint_REP=${wdisjoint_REP:-100}
 
 # write_disjoint
 [ "$SLOW" = "no" ] && wdisjoint_REP=${wdisjoint_REP:-100}
 

+# fs_test
+if [ "$SLOW" = "no" ]; then
+       fs_test_ndirs=${fs_test_ndirs:-10000}
+       fs_test_nobj=${fs_test_nobj:-2}
+fi
+

 . $LUSTRE/tests/functions.sh
 
 build_test_filter
 . $LUSTRE/tests/functions.sh
 
 build_test_filter


@@ -135,6 +141,11 @@ test_statahead () {
 }
 run_test statahead "statahead test, multiple clients"
 
 }
 run_test statahead "statahead test, multiple clients"
 

+test_fs_test () {
+       run_fs_test
+}
+run_test fs_test "fs_test"
+

 [ $(facet_fstype $SINGLEMDS) = zfs -o $(facet_fstype "ost1") = zfs ] &&
        SLOW=$ZFSSLOW
 
 [ $(facet_fstype $SINGLEMDS) = zfs -o $(facet_fstype "ost1") = zfs ] &&
        SLOW=$ZFSSLOW