# The number of clients run MPI loads is configured by parameter
# ha_mpi_instances. Only one client runs MPI workloads by default.
#
+# MPI workloads can be run from several users. The list of users to use is
+# configured by parameter ha_mpi_users, default is "mpiuser".
+#
# PROCESS STRUCTURE AND IPC
#
# On the node where this script is run, the processes look like this:
declare ha_mpi_loads=${ha_mpi_loads="ior simul mdtest"}
declare -a ha_mpi_load_tags=($ha_mpi_loads)
+declare -a ha_mpiusers=(${ha_mpi_users="mpiuser"})
declare ha_ior_params=${IORP:-'" -b $ior_blockSize -t 2m -w -W -T 1"'}
declare ha_simul_params=${SIMULP:-'" -n 10"'}
declare ha_mpirun_options=${MPIRUN_OPTIONS:-""}
declare ha_clients_stripe=${CLIENTSSTRIPE:-'"$STRIPEPARAMS"'}
declare ha_nclientsset=${NCLIENTSSET:-1}
+declare ha_ninstmustfail=${NINSTMUSTFAIL:-0}
declare ha_racer_params=${RACERP:-"MDSCOUNT=1"}
local parameter=$4
local machines=$5
local stripeparams=$6
+ local mpiuser=$7
+ local mustpass=$8
local tag=${ha_mpi_load_tags[$load]}
local cmd=${ha_mpi_load_cmds[$tag]}
local dir=$ha_test_dir/$client-$tag
cmd=${cmd//"{}"/$dir}
cmd=${cmd//"{params}"/$parameter}
+ [[ -n "$ha_postcmd" ]] && ha_postcmd=${ha_postcmd//"{}"/$dir}
ha_info "Starting $tag"
machines="-machinefile $machines"
while [ ! -e "$ha_stop_file" ] && ((rc == 0)); do
+ ha_info "$client Starts: $cmd" 2>&1 | tee -a $log
{
local mdt_index
if $ha_mdt_index_random && [ $ha_mdt_index -ne 0 ]; then
ha_on $client $LFS setstripe $stripeparams $dir &&
ha_on $client $LFS getstripe $dir &&
ha_on $client chmod a+xwr $dir &&
- ha_on $client "su mpiuser sh -c \" $mpirun $ha_mpirun_options \
+ ha_on $client "su $mpiuser sh -c \" $mpirun $ha_mpirun_options \
-np $((${#ha_clients[@]} * mpi_threads_per_client )) \
- $machines $cmd \" " &&
+ $machines $cmd \" " || rc=$?
+ [[ -n "$ha_postcmd" ]] && ha_info "$ha_postcmd" &&
+ ha_on $client $ha_postcmd >>"$log" 2>&1
+ (( ((rc == 0)) && (( mustpass != 0 )) )) ||
+ (( ((rc != 0)) && (( mustpass == 0 )) )) &&
ha_on $client rm -rf "$dir";
} >>"$log" 2>&1 || rc=$?
- ha_info rc=$rc
-
- if ((rc != 0)); then
+ ha_info rc=$rc mustpass=$mustpass
+
+ # mustpass=0 means that failure is expected
+ if (( rc !=0 )); then
+ if (( mustpass != 0 )); then
+ touch "$ha_fail_file"
+ touch "$ha_stop_file"
+ ha_dump_logs "${ha_clients[*]} ${ha_servers[*]}"
+ else
+ # Ok to fail
+ rc=0
+ fi
+ elif (( mustpass == 0 )); then
touch "$ha_fail_file"
touch "$ha_stop_file"
ha_dump_logs "${ha_clients[*]} ${ha_servers[*]}"
fi
- echo $rc >"$status"
+ echo rc=$rc mustpass=$mustpass >"$status"
nr_loops=$((nr_loops + 1))
done
[ $nr_loops -ne 0 ] &&
avg_loop_time=$((($(date +%s) - start_time) / nr_loops))
- ha_info "$tag stopped: rc $rc avg loop time $avg_loop_time"
+ ha_info "$tag stopped: rc=$rc mustpass=$mustpass \
+ avg loop time $avg_loop_time"
}
ha_start_mpi_loads()
local machines
local m
local -a mach
+ local mpiuser
# ha_mpi_instances defines the number of
# clients start mpi loads; should be <= ${#ha_clients[@]}
for ((n = 0; n < $inst; n++)); do
client=${ha_clients[n]}
+ mpiuser=${ha_mpiusers[$((n % ${#ha_mpiusers[@]}))]}
for ((load = 0; load < ${#ha_mpi_load_tags[@]}; load++)); do
tag=${ha_mpi_load_tags[$load]}
status=$ha_status_file_prefix-$tag-$client
local stripe=${!aref}
local m=$(( n % ha_nclientsset))
machines=${mach[m]}
- ha_repeat_mpi_load $client $load $status "$parameter" $machines "$stripe" &
+ local mustpass=1
+ [[ $ha_ninstmustfail == 0 ]] ||
+ mustpass=$(( n % ha_ninstmustfail ))
+ ha_repeat_mpi_load $client $load $status "$parameter" \
+ $machines "$stripe" "$mpiuser" "$mustpass" &
ha_status_files+=("$status")
done
done
local avg_loop_time=0
local start_time=$(date +%s)
- cmd=${cmd//"{}"/$dir}
+ cmd=${cmd//"{}"/$dir}
- ha_info "Starting $tag on $client"
+ ha_info "Starting $tag on $client"
while [ ! -e "$ha_stop_file" ] && ((rc == 0)); do
+ ha_info "$client Starts: $cmd" 2>&1 | tee -a $log
ha_on $client "mkdir -p $dir && \
$cmd && \
rm -rf $dir" >>"$log" 2>&1 || rc=$?
ha_start_loads()
{
- trap ha_trap_stop_signals $ha_stop_signals
- ha_start_nonmpi_loads
- ha_start_mpi_loads
+ $ha_lfsck_bg && ha_lfsck_bg
+ trap ha_trap_stop_signals $ha_stop_signals
+ ha_start_nonmpi_loads
+ ha_start_mpi_loads
}
ha_stop_loads()
{
- touch $ha_stop_file
- trap - $ha_stop_signals
- ha_info "Waiting for workloads to stop"
- wait
+ touch $ha_stop_file
+ # true because of lfsck_bg could be stopped already
+ $ha_lfsck_bg && wait $LFSCK_BG_PID || true
+ trap - $ha_stop_signals
+ ha_info "Waiting for workloads to stop"
+ wait
}
ha_wait_loads()
ha_on ${ha_clients[0]} " \
$LFS setstripe $ha_stripe_params $ha_test_dir"
- $ha_lfsck_bg && ha_lfsck_bg
-
ha_start_loads
ha_wait_loads
$ha_lfsck_fail_on_repaired && ha_lfsck_repaired
- # true because of lfsck_bg could be stopped already
- $ha_lfsck_bg && wait $LFSCK_BG_PID || true
-
if [ -e "$ha_fail_file" ]; then
exit 1
else