feat: allow dynamic port in scripts in preparation for parallel testing (part 1) (#4546)

Signed-off-by: Keiven Chang <keivenchang@users.noreply.github.com>

examples/backends/vllm/launch/agg_kvbm_router.sh

@@ -11,9 +11,9 @@ export PYTHONHASHSEED=0
 MODEL="Qwen/Qwen3-0.6B"
 
 # run frontend + KV router
+# dynamo.frontend accepts either --http-port flag or DYN_HTTP_PORT env var (defaults to 8000)
 python -m dynamo.frontend \
     --router-mode kv \
-    --http-port 8000 \
     --router-reset-states &
 
 # run workers with KVBM enabled

examples/backends/vllm/launch/agg_lmcache.sh

@@ -5,8 +5,9 @@ set -e
 trap 'echo Cleaning up...; kill 0' EXIT
 
 # run ingress
-python -m dynamo.frontend --http-port=8000 &
+# dynamo.frontend accepts either --http-port flag or DYN_HTTP_PORT env var (defaults to 8000)
+python -m dynamo.frontend &
 
 # run worker with LMCache enabled
-DYN_SYSTEM_PORT=8081 \
+DYN_SYSTEM_PORT=${DYN_SYSTEM_PORT:-8081} \
   python -m dynamo.vllm --model Qwen/Qwen3-0.6B --connector lmcache

examples/backends/vllm/launch/agg_multimodal.sh

@@ -45,7 +45,8 @@ done
 export DYN_REQUEST_PLANE=tcp
 
 # Start frontend with Rust OpenAIPreprocessor
-python -m dynamo.frontend --http-port=8000 &
+# dynamo.frontend accepts either --http-port flag or DYN_HTTP_PORT env var (defaults to 8000)
+python -m dynamo.frontend &
 
 # Configure GPU memory optimization for specific models
 EXTRA_ARGS=""
@@ -59,7 +60,7 @@ fi
 # Multimodal data (images) are decoded in the backend worker using ImageLoader
 # --enforce-eager: Quick deployment (remove for production)
 # --connector none: No KV transfer needed for aggregated serving
-DYN_SYSTEM_PORT=8081 \
+DYN_SYSTEM_PORT=${DYN_SYSTEM_PORT:-8081} \
     python -m dynamo.vllm --enable-multimodal --model $MODEL_NAME --enforce-eager --connector none $EXTRA_ARGS
 
 # Wait for all background processes to complete

examples/backends/vllm/launch/agg_multimodal_epd.sh

@@ -64,7 +64,8 @@ else
 fi
 
 # Start frontend (HTTP endpoint)
-python -m dynamo.frontend --http-port=8000 &
+# dynamo.frontend accepts either --http-port flag or DYN_HTTP_PORT env var (defaults to 8000)
+python -m dynamo.frontend &
 
 # To make Qwen2.5-VL fit in A100 40GB, set the following extra arguments
 EXTRA_ARGS=""

examples/backends/vllm/launch/agg_multimodal_llama.sh

@@ -8,7 +8,8 @@ trap 'echo Cleaning up...; kill 0' EXIT
 MODEL_NAME="meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8"
 
 # run ingress
-python -m dynamo.frontend --http-port=8000 &
+# dynamo.frontend accepts either --http-port flag or DYN_HTTP_PORT env var (defaults to 8000)
+python -m dynamo.frontend &
 
 # run processor
 python -m dynamo.vllm --multimodal-processor --enable-multimodal --model $MODEL_NAME --mm-prompt-template "<|image|>\n<prompt>" &

examples/backends/vllm/launch/agg_request_planes.sh

@@ -41,8 +41,9 @@ export DYN_REQUEST_PLANE=$REQUEST_PLANE
 echo "Using request plane mode: $REQUEST_PLANE"
 
 # Frontend
-python -m dynamo.frontend --http-port=8000 &
+# dynamo.frontend accepts either --http-port flag or DYN_HTTP_PORT env var (defaults to 8000)
+python -m dynamo.frontend &
 
-DYN_SYSTEM_PORT=8081 \
+DYN_SYSTEM_PORT=${DYN_SYSTEM_PORT:-8081} \
 DYN_HEALTH_CHECK_ENABLED=true \
     python -m dynamo.vllm --model Qwen/Qwen3-0.6B --enforce-eager --connector none

examples/backends/vllm/launch/agg_router.sh

@@ -12,9 +12,9 @@ MODEL="Qwen/Qwen3-0.6B"
 BLOCK_SIZE=64
 
 # run frontend + KV router
+# dynamo.frontend accepts either --http-port flag or DYN_HTTP_PORT env var (defaults to 8000)
 python -m dynamo.frontend \
     --router-mode kv \
-    --http-port 8000 \
     --router-reset-states &
 
 # run workers

examples/backends/vllm/launch/dep.sh

@@ -5,7 +5,8 @@ set -e
 trap 'echo Cleaning up...; kill 0' EXIT
 
 # run ingress
-python -m dynamo.frontend --router-mode kv --http-port=8000 &
+# dynamo.frontend accepts either --http-port flag or DYN_HTTP_PORT env var (defaults to 8000)
+python -m dynamo.frontend --router-mode kv &
 
 # Data Parallel Attention / Expert Parallelism
 # Routing to DP workers managed by Dynamo

examples/backends/vllm/launch/disagg.sh

@@ -5,7 +5,8 @@ set -e
 trap 'echo Cleaning up...; kill 0' EXIT
 
 # run ingress
-python -m dynamo.frontend --http-port=8000 &
+# dynamo.frontend accepts either --http-port flag or DYN_HTTP_PORT env var (defaults to 8000)
+python -m dynamo.frontend &
 
 # --enforce-eager is added for quick deployment. for production use, need to remove this flag
 CUDA_VISIBLE_DEVICES=0 python3 -m dynamo.vllm --model Qwen/Qwen3-0.6B --enforce-eager &

examples/backends/vllm/launch/disagg_kvbm.sh

@@ -5,7 +5,8 @@ set -e
 trap 'echo Cleaning up...; kill 0' EXIT
 
 # run ingress
-python -m dynamo.frontend --http-port=8000 &
+# dynamo.frontend accepts either --http-port flag or DYN_HTTP_PORT env var (defaults to 8000)
+python -m dynamo.frontend &
 
 # run decode worker on GPU 0, without enabling KVBM
 # NOTE: remove --enforce-eager for production use

examples/backends/vllm/launch/disagg_kvbm_2p2d.sh

@@ -5,7 +5,8 @@ set -e
 trap 'echo Cleaning up...; kill 0' EXIT
 
 # run ingress with KV router
-python -m dynamo.frontend --router-mode kv --http-port=8000 &
+# dynamo.frontend accepts either --http-port flag or DYN_HTTP_PORT env var (defaults to 8000)
+python -m dynamo.frontend --router-mode kv &
 
 # run decode workers on GPU 0 and 1, without enabling KVBM
 # NOTE: remove --enforce-eager for production use

examples/backends/vllm/launch/disagg_kvbm_router.sh

@@ -10,9 +10,9 @@ export PYTHONHASHSEED=0
 # Common configuration
 MODEL="Qwen/Qwen3-0.6B"
 
+# dynamo.frontend accepts either --http-port flag or DYN_HTTP_PORT env var (defaults to 8000)
 python -m dynamo.frontend \
     --router-mode kv \
-    --http-port 8000 \
     --router-reset-states &
 
 # two decode workers (without KVBM)

examples/backends/vllm/launch/disagg_lmcache.sh

@@ -5,7 +5,8 @@ set -e
 trap 'echo Cleaning up...; kill 0' EXIT
 
 # run ingress with KV router
-python -m dynamo.frontend --router-mode kv --http-port=8000 &
+# dynamo.frontend accepts either --http-port flag or DYN_HTTP_PORT env var (defaults to 8000)
+python -m dynamo.frontend --router-mode kv &
 
 # run decode worker on GPU 0, without enabling LMCache
 CUDA_VISIBLE_DEVICES=0 python3 -m dynamo.vllm --model Qwen/Qwen3-0.6B &

examples/backends/vllm/launch/disagg_multimodal_epd.sh

@@ -72,7 +72,8 @@ echo "=================================================="
 
 # Start frontend (no router mode)
 echo "Starting frontend..."
-python -m dynamo.frontend --http-port=8000 &
+# dynamo.frontend accepts either --http-port flag or DYN_HTTP_PORT env var (defaults to 8000)
+python -m dynamo.frontend &
 
 # Start processor
 echo "Starting processor..."

examples/backends/vllm/launch/disagg_multimodal_llama.sh

@@ -45,7 +45,8 @@ MODEL_NAME="meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8"
 
 if [[ $HEAD_NODE -eq 1 ]]; then
     # run ingress
-    python -m dynamo.frontend --http-port=8000 &
+    # dynamo.frontend accepts either --http-port flag or DYN_HTTP_PORT env var (defaults to 8000)
+    python -m dynamo.frontend &
 
     # run processor
     python -m dynamo.vllm --multimodal-processor --enable-multimodal --model $MODEL_NAME --mm-prompt-template "<|image|>\n<prompt>" &

examples/backends/vllm/launch/disagg_router.sh

@@ -13,9 +13,9 @@ BLOCK_SIZE=64
 
 # Start frontend with KV routing
 # The frontend will automatically detect prefill workers and activate an internal prefill router
+# dynamo.frontend accepts either --http-port flag or DYN_HTTP_PORT env var (defaults to 8000)
 python -m dynamo.frontend \
     --router-mode kv \
-    --http-port 8000 \
     --router-reset-states &
 
 # two decode workers

examples/backends/vllm/launch/disagg_same_gpu.sh

@@ -42,12 +42,13 @@ cleanup() {
 trap cleanup EXIT INT TERM
 
 # run ingress
-python3 -m dynamo.frontend --http-port=8000 &
+# dynamo.frontend accepts either --http-port flag or DYN_HTTP_PORT env var (defaults to 8000)
+python3 -m dynamo.frontend &
 DYNAMO_PID=$!
 
 # run decode worker with metrics on port 8081
 # --enforce-eager is added for quick deployment. for production use, need to remove this flag
-DYN_SYSTEM_PORT=8081 \
+DYN_SYSTEM_PORT=${DYN_SYSTEM_PORT:-8081} \
 CUDA_VISIBLE_DEVICES=0 \
 python3 -m dynamo.vllm \
   --model Qwen/Qwen3-0.6B \
@@ -65,7 +66,7 @@ echo "Waiting for decode worker to initialize..."
 sleep 10
 
 # run prefill worker with metrics on port 8082 (foreground)
-DYN_SYSTEM_PORT=8082 \
+DYN_SYSTEM_PORT=${DYN_SYSTEM_PORT_PREFILL:-8082} \
 DYN_VLLM_KV_EVENT_PORT=20081 \
 VLLM_NIXL_SIDE_CHANNEL_PORT=20097 \
 CUDA_VISIBLE_DEVICES=0 \

examples/backends/vllm/launch/dsr1_dep.sh

@@ -82,8 +82,9 @@ echo "  Model name: $MODEL"
 trap 'echo Cleaning up...; kill 0' EXIT
 
 # run ingress if it's node 0
+# dynamo.frontend accepts either --http-port flag or DYN_HTTP_PORT env var (defaults to 8000)
 if [ $NODE_RANK -eq 0 ]; then
-    DYN_LOG=debug python -m dynamo.frontend --router-mode kv --http-port=8000 2>&1 | tee $LOG_DIR/dsr1_dep_ingress.log &
+    DYN_LOG=debug python -m dynamo.frontend --router-mode kv 2>&1 | tee $LOG_DIR/dsr1_dep_ingress.log &
 fi
 
 mkdir -p $LOG_DIR

examples/basics/multinode/trtllm/start_frontend_services.sh

@@ -12,5 +12,6 @@ etcd --listen-client-urls http://0.0.0.0:2379 --advertise-client-urls http://0.0
 sleep 3
 
 # Start OpenAI Frontend which will dynamically discover workers when they startup
+# dynamo.frontend accepts either --http-port flag or DYN_HTTP_PORT env var (defaults to 8000)
 # NOTE: This is a blocking call.
-python3 -m dynamo.frontend --http-port 8000
+python3 -m dynamo.frontend

tests/serve/launch/template_verifier.sh

@@ -17,7 +17,8 @@ cleanup() {
 trap cleanup EXIT INT TERM
 
 # run ingress
-python3 -m dynamo.frontend --http-port=8000 &
+# dynamo.frontend accepts either --http-port flag or DYN_HTTP_PORT env var (defaults to 8000)
+python3 -m dynamo.frontend &
 FRONTEND_PID=$!
 
 # run the mock worker + template validation generate()