Merge pull request #5598 from vjpai/make_qps_great_again

Improve perf smoke test stability

1 files changed, 48 insertions, 11 deletions

diff --git a/tools/jenkins/run_performance.sh b/tools/jenkins/run_performance.sh
index fbc078330f..b06fce7acb 100755
--- a/tools/jenkins/run_performance.sh
+++ b/tools/jenkins/run_performance.sh
@@ -42,17 +42,10 @@ config=opt
 
 make CONFIG=$config qps_worker qps_driver -j8
 
-bins/$config/qps_worker -driver_port 10000 &
-PID1=$!
-bins/$config/qps_worker -driver_port 10010 &
-PID2=$!
-
 #
 # Put a timeout on these tests
 #
-((sleep 900; kill $$ && killall qps_worker && rm -f /tmp/qps-test.$$ )&)
-
-export QPS_WORKERS="localhost:10000,localhost:10010"
+((sleep 900; killall qps_worker && rm -f /tmp/qps-test.$$ && kill $$)&)
 
 # big is the size in bytes of large messages (0 is the size otherwise)
 big=65536
@@ -69,40 +62,74 @@ deep=100
 cores=`grep -c ^processor /proc/cpuinfo`
 halfcores=`expr $cores / 2`
 
+#
+# Keep changing the ports
+port1=10000
+port2=10100
+
 for secure in true false; do
   # Scenario 1: generic async streaming ping-pong (contentionless latency)
+  bins/$config/qps_worker -driver_port $port1 &
+  bins/$config/qps_worker -driver_port $port2 &
+  export QPS_WORKERS="localhost:$port1,localhost:$port2"
+  port1=`expr $port1 + 1`
+  port2=`expr $port2 + 1`
   bins/$config/qps_driver --rpc_type=STREAMING --client_type=ASYNC_CLIENT \
     --server_type=ASYNC_GENERIC_SERVER --outstanding_rpcs_per_channel=1 \
     --client_channels=1 --bbuf_req_size=0 --bbuf_resp_size=0 \
     --async_client_threads=1 --async_server_threads=1 --secure_test=$secure \
     --num_servers=1 --num_clients=1 \
     --server_core_limit=$halfcores --client_core_limit=0
+  bins/$config/qps_driver --quit=true
 
   # Scenario 2: generic async streaming "unconstrained" (QPS)
+  bins/$config/qps_worker -driver_port $port1 &
+  bins/$config/qps_worker -driver_port $port2 &
+  export QPS_WORKERS="localhost:$port1,localhost:$port2"
+  port1=`expr $port1 + 1`
+  port2=`expr $port2 + 1`
   bins/$config/qps_driver --rpc_type=STREAMING --client_type=ASYNC_CLIENT \
     --server_type=ASYNC_GENERIC_SERVER --outstanding_rpcs_per_channel=$deep \
     --client_channels=$wide --bbuf_req_size=0 --bbuf_resp_size=0 \
     --async_client_threads=0 --async_server_threads=0 --secure_test=$secure \
     --num_servers=1 --num_clients=0 \
     --server_core_limit=$halfcores --client_core_limit=0 |& tee /tmp/qps-test.$$
+  bins/$config/qps_driver --quit=true
 
   # Scenario 2b: QPS with a single server core
+  bins/$config/qps_worker -driver_port $port1 &
+  bins/$config/qps_worker -driver_port $port2 &
+  export QPS_WORKERS="localhost:$port1,localhost:$port2"
+  port1=`expr $port1 + 1`
+  port2=`expr $port2 + 1`
   bins/$config/qps_driver --rpc_type=STREAMING --client_type=ASYNC_CLIENT \
     --server_type=ASYNC_GENERIC_SERVER --outstanding_rpcs_per_channel=$deep \
     --client_channels=$wide --bbuf_req_size=0 --bbuf_resp_size=0 \
     --async_client_threads=0 --async_server_threads=0 --secure_test=$secure \
     --num_servers=1 --num_clients=0 --server_core_limit=1 --client_core_limit=0
+  bins/$config/qps_driver --quit=true
 
   # Scenario 2c: protobuf-based QPS
+  bins/$config/qps_worker -driver_port $port1 &
+  bins/$config/qps_worker -driver_port $port2 &
+  export QPS_WORKERS="localhost:$port1,localhost:$port2"
+  port1=`expr $port1 + 1`
+  port2=`expr $port2 + 1`
   bins/$config/qps_driver --rpc_type=STREAMING --client_type=ASYNC_CLIENT \
     --server_type=ASYNC_SERVER --outstanding_rpcs_per_channel=$deep \
     --client_channels=$wide --simple_req_size=0 --simple_resp_size=0 \
     --async_client_threads=0 --async_server_threads=0 --secure_test=$secure \
     --num_servers=1 --num_clients=0 \
     --server_core_limit=$halfcores --client_core_limit=0
+  bins/$config/qps_driver --quit=true
 
   # Scenario 3: Latency at sub-peak load (all clients equally loaded)
   for loadfactor in 0.7; do
+    bins/$config/qps_worker -driver_port $port1 &
+    bins/$config/qps_worker -driver_port $port2 &
+    export QPS_WORKERS="localhost:$port1,localhost:$port2"
+    port1=`expr $port1 + 1`
+    port2=`expr $port2 + 1`
     bins/$config/qps_driver --rpc_type=STREAMING --client_type=ASYNC_CLIENT \
       --server_type=ASYNC_GENERIC_SERVER --outstanding_rpcs_per_channel=$deep \
       --client_channels=$wide --bbuf_req_size=0 --bbuf_resp_size=0 \
@@ -110,27 +137,37 @@ for secure in true false; do
       --num_servers=1 --num_clients=0 --poisson_load=`awk -v lf=$loadfactor \
       '$5 == "QPS:" {print int(lf * $6); exit}' /tmp/qps-test.$$` \
       --server_core_limit=$halfcores --client_core_limit=0
+    bins/$config/qps_driver --quit=true
   done
 
   rm /tmp/qps-test.$$
 
   # Scenario 4: Single-channel bidirectional throughput test (like TCP_STREAM).
+  bins/$config/qps_worker -driver_port $port1 &
+  bins/$config/qps_worker -driver_port $port2 &
+  export QPS_WORKERS="localhost:$port1,localhost:$port2"
+  port1=`expr $port1 + 1`
+  port2=`expr $port2 + 1`
   bins/$config/qps_driver --rpc_type=STREAMING --client_type=ASYNC_CLIENT \
     --server_type=ASYNC_GENERIC_SERVER --outstanding_rpcs_per_channel=$deep \
     --client_channels=1 --bbuf_req_size=$big --bbuf_resp_size=$big \
     --async_client_threads=1 --async_server_threads=1 --secure_test=$secure \
     --num_servers=1 --num_clients=1 \
     --server_core_limit=$halfcores --client_core_limit=0
+  bins/$config/qps_driver --quit=true
 
   # Scenario 5: Sync unary ping-pong with protobufs
+  bins/$config/qps_worker -driver_port $port1 &
+  bins/$config/qps_worker -driver_port $port2 &
+  export QPS_WORKERS="localhost:$port1,localhost:$port2"
+  port1=`expr $port1 + 1`
+  port2=`expr $port2 + 1`
   bins/$config/qps_driver --rpc_type=UNARY --client_type=SYNC_CLIENT \
     --server_type=SYNC_SERVER --outstanding_rpcs_per_channel=1 \
     --client_channels=1 --simple_req_size=0 --simple_resp_size=0 \
     --secure_test=$secure --num_servers=1 --num_clients=1 \
     --server_core_limit=$halfcores --client_core_limit=0
-
+  bins/$config/qps_driver --quit=true
 done
 
-bins/$config/qps_driver --quit=true
-
 wait