|
|
|
@ -16,18 +16,24 @@
|
|
|
|
|
|
|
|
|
|
set -e
|
|
|
|
|
|
|
|
|
|
server_port_0=${PADDLE_DIST_UT_PORT}
|
|
|
|
|
server_port_1=$(( PADDLE_DIST_UT_PORT + 1 ))
|
|
|
|
|
worker_port_0=$(( PADDLE_DIST_UT_PORT + 2 ))
|
|
|
|
|
worker_port_1=$(( PADDLE_DIST_UT_PORT + 3 ))
|
|
|
|
|
heter_worker_port_0=$(( PADDLE_DIST_UT_PORT + 4 ))
|
|
|
|
|
heter_worker_port_1=$(( PADDLE_DIST_UT_PORT + 5 ))
|
|
|
|
|
server_port_00=${PADDLE_DIST_UT_PORT}
|
|
|
|
|
server_port_10=$(( PADDLE_DIST_UT_PORT + 1 ))
|
|
|
|
|
worker_port_00=$(( PADDLE_DIST_UT_PORT + 2 ))
|
|
|
|
|
worker_port_10=$(( PADDLE_DIST_UT_PORT + 3 ))
|
|
|
|
|
|
|
|
|
|
server_port_01=$(( PADDLE_DIST_UT_PORT + 4 ))
|
|
|
|
|
server_port_11=$(( PADDLE_DIST_UT_PORT + 5 ))
|
|
|
|
|
worker_port_01=$(( PADDLE_DIST_UT_PORT + 6 ))
|
|
|
|
|
worker_port_11=$(( PADDLE_DIST_UT_PORT + 7 ))
|
|
|
|
|
|
|
|
|
|
heter_worker_port_0=$(( PADDLE_DIST_UT_PORT + 8 ))
|
|
|
|
|
heter_worker_port_1=$(( PADDLE_DIST_UT_PORT + 9 ))
|
|
|
|
|
|
|
|
|
|
function test_launch_ps(){
|
|
|
|
|
|
|
|
|
|
python -m paddle.distributed.fleet.launch \
|
|
|
|
|
--servers="127.0.0.1:${server_port_0},127.0.0.1:${server_port_1}" \
|
|
|
|
|
--workers="127.0.0.1:${worker_port_0},127.0.0.1:${worker_port_1}" \
|
|
|
|
|
--servers="127.0.0.1:${server_port_00},127.0.0.1:${server_port_10}" \
|
|
|
|
|
--workers="127.0.0.1:${worker_port_00},127.0.0.1:${worker_port_10}" \
|
|
|
|
|
fleet_ps_training.py 2> ut.elog
|
|
|
|
|
if grep -q "server are killed" ut.elog; then
|
|
|
|
|
echo "test pserver launch succeed"
|
|
|
|
@ -39,8 +45,8 @@ function test_launch_ps(){
|
|
|
|
|
|
|
|
|
|
function test_launch_ps_heter(){
|
|
|
|
|
python -m paddle.distributed.fleet.launch \
|
|
|
|
|
--servers="127.0.0.1:${server_port_0},127.0.0.1:${server_port_1}" \
|
|
|
|
|
--workers="127.0.0.1:${worker_port_0},127.0.0.1:${worker_port_1}" \
|
|
|
|
|
--servers="127.0.0.1:${server_port_01},127.0.0.1:${server_port_11}" \
|
|
|
|
|
--workers="127.0.0.1:${worker_port_01},127.0.0.1:${worker_port_11}" \
|
|
|
|
|
--heter_workers="127.0.0.1:${heter_worker_port_0},127.0.0.1:${heter_worker_port_1}" \
|
|
|
|
|
fleet_ps_training.py 2> ut.elog
|
|
|
|
|
if grep -q "server are killed" ut.elog; then
|
|
|
|
|