[CI] Fix PD job (#1129)

Fix e2e test for Pd job

Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>

Author: wangxiyuan

tests/e2e/common.sh

@@ -14,7 +14,7 @@ _err() { _red "Error: $*" && exit 1; }
 
 CURL_TIMEOUT=1
 CURL_COOLDOWN=5
-CURL_MAX_TRIES=120
+CURL_MAX_TRIES=180
 
 function wait_url_ready() {
   local serve_name="$1"
@@ -31,7 +31,7 @@ function wait_url_ready() {
       break
     fi
     if [ "$i" -gt "$CURL_MAX_TRIES" ]; then
-      _info "===> \$CURL_MAX_TRIES exceeded waiting for ${serve_name} to be ready"
+      _info "===> ${CURL_MAX_TRIES}s exceeded waiting for ${serve_name} to be ready"
       return 1
     fi
     sleep "$CURL_COOLDOWN"

tests/e2e/pd_disaggreate/setup_pd.sh

@@ -66,7 +66,7 @@ function run_prefill_instance() {
   --served-model-name Deepseek \
   --max-model-len 2000 \
   --trust-remote-code \
-  --kv-transfer-config "$KV_CONFIG" &
+  --kv-transfer-config "$KV_CONFIG"
 }
 
 
@@ -119,7 +119,7 @@ function run_decode_instance() {
     --max-num-batched-tokens 2000 \
     --trust-remote-code \
     --gpu-memory-utilization 0.9 \
-    --kv-transfer-config "$KV_CONFIG" &
+    --kv-transfer-config "$KV_CONFIG"
 }
 
 function run_proxy_server() {

tests/e2e/run_disagg_pd.sh

@@ -43,16 +43,16 @@ _info "Started pd disaggregated proxy server"
 
 PREFILL_PROC_NAME="Prefill-instance"
 PREFILL_PORT=8001
-run_prefill_instance $MODEL_NAME $TP_SIZE $PREFILL_PORT $REGISTER_PORT $PREFILL_DEVICE_IPS $DECODE_DEVICE_IPS
 _info "Starting prefill instance"
-
+run_prefill_instance $MODEL_NAME $TP_SIZE $PREFILL_PORT $REGISTER_PORT $PREFILL_DEVICE_IPS $DECODE_DEVICE_IPS &
+_info "Waiting for prefill instance ready"
 wait_url_ready $PREFILL_PROC_NAME "http://localhost:${PREFILL_PORT}/v1/completions"
 
 DECODE_PROC_NAME="Decode-instance"
 DECODE_PORT=8002
-run_decode_instance  $MODEL_NAME $TP_SIZE $DECODE_PORT $REGISTER_PORT $PREFILL_DEVICE_IPS $DECODE_DEVICE_IPS
 _info "Starting decode instance"
-
+run_decode_instance  $MODEL_NAME $TP_SIZE $DECODE_PORT $REGISTER_PORT $PREFILL_DEVICE_IPS $DECODE_DEVICE_IPS &
+_info "Waiting for decode instance ready"
 wait_url_ready $DECODE_PROC_NAME "http://localhost:${DECODE_PORT}/v1/completions"
 
 _info "pd disaggregated system is ready for handling request"