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/sglang/launch/agg.sh

@@ -44,12 +44,13 @@ if [ "$ENABLE_OTEL" = true ]; then
 fi
 
 # run ingress
+# dynamo.frontend accepts either --http-port flag or DYN_HTTP_PORT env var (defaults to 8000)
 OTEL_SERVICE_NAME=dynamo-frontend \
-python3 -m dynamo.frontend --http-port=8000 &
+python3 -m dynamo.frontend &
 DYNAMO_PID=$!
 
-# run worker
-OTEL_SERVICE_NAME=dynamo-worker DYN_SYSTEM_PORT=8081 \
+# run worker with metrics enabled
+DYN_SYSTEM_PORT=${DYN_SYSTEM_PORT:-8081} \
 python3 -m dynamo.sglang \
   --model-path Qwen/Qwen3-0.6B \
   --served-model-name Qwen/Qwen3-0.6B \

examples/backends/sglang/launch/agg_embed.sh

@@ -44,12 +44,13 @@ if [ "$ENABLE_OTEL" = true ]; then
 fi
 
 # run ingress
+# dynamo.frontend accepts either --http-port flag or DYN_HTTP_PORT env var (defaults to 8000)
 OTEL_SERVICE_NAME=dynamo-frontend \
-python3 -m dynamo.frontend --http-port=8000 &
+python3 -m dynamo.frontend &
 DYNAMO_PID=$!
 
 # run worker
-OTEL_SERVICE_NAME=dynamo-worker-embedding DYN_SYSTEM_PORT=8081 \
+OTEL_SERVICE_NAME=dynamo-worker-embedding DYN_SYSTEM_PORT=${DYN_SYSTEM_PORT:-8081} \
 python3 -m dynamo.sglang \
   --embedding-worker \
   --model-path Qwen/Qwen3-Embedding-4B \

examples/backends/sglang/launch/agg_router.sh

@@ -44,12 +44,13 @@ if [ "$ENABLE_OTEL" = true ]; then
 fi
 
 # run ingress
+# dynamo.frontend accepts either --http-port flag or DYN_HTTP_PORT env var (defaults to 8000)
 OTEL_SERVICE_NAME=dynamo-frontend \
-python3 -m dynamo.frontend --router-mode kv --http-port=8000 &
+python3 -m dynamo.frontend --router-mode kv &
 DYNAMO_PID=$!
 
 # run worker
-OTEL_SERVICE_NAME=dynamo-worker-1 DYN_SYSTEM_PORT=8081 \
+OTEL_SERVICE_NAME=dynamo-worker-1 DYN_SYSTEM_PORT=${DYN_SYSTEM_PORT_WORKER1:-8081} \
 python3 -m dynamo.sglang \
   --model-path Qwen/Qwen3-0.6B \
   --served-model-name Qwen/Qwen3-0.6B \
@@ -60,7 +61,7 @@ python3 -m dynamo.sglang \
   --enable-metrics &
 WORKER_PID=$!
 
-OTEL_SERVICE_NAME=dynamo-worker-2 DYN_SYSTEM_PORT=8082 \
+OTEL_SERVICE_NAME=dynamo-worker-2 DYN_SYSTEM_PORT=${DYN_SYSTEM_PORT_WORKER2:-8082} \
 CUDA_VISIBLE_DEVICES=1 python3 -m dynamo.sglang \
   --model-path Qwen/Qwen3-0.6B \
   --served-model-name Qwen/Qwen3-0.6B \

examples/backends/sglang/launch/disagg.sh

@@ -44,12 +44,13 @@ if [ "$ENABLE_OTEL" = true ]; then
 fi
 
 # run ingress
+# dynamo.frontend accepts either --http-port flag or DYN_HTTP_PORT env var (defaults to 8000)
 OTEL_SERVICE_NAME=dynamo-frontend \
-python3 -m dynamo.frontend --http-port=8000 &
+python3 -m dynamo.frontend &
 DYNAMO_PID=$!
 
 # run prefill worker
-OTEL_SERVICE_NAME=dynamo-worker-prefill DYN_SYSTEM_PORT=8081 \
+OTEL_SERVICE_NAME=dynamo-worker-prefill DYN_SYSTEM_PORT=${DYN_SYSTEM_PORT_PREFILL:-8081} \
 python3 -m dynamo.sglang \
   --model-path Qwen/Qwen3-0.6B \
   --served-model-name Qwen/Qwen3-0.6B \
@@ -64,7 +65,7 @@ python3 -m dynamo.sglang \
 PREFILL_PID=$!
 
 # run decode worker
-OTEL_SERVICE_NAME=dynamo-worker-decode DYN_SYSTEM_PORT=8082 \
+OTEL_SERVICE_NAME=dynamo-worker-decode DYN_SYSTEM_PORT=${DYN_SYSTEM_PORT_DECODE:-8082} \
 CUDA_VISIBLE_DEVICES=1 python3 -m dynamo.sglang \
   --model-path Qwen/Qwen3-0.6B \
   --served-model-name Qwen/Qwen3-0.6B \

examples/backends/sglang/launch/disagg_router.sh

@@ -45,16 +45,16 @@ if [ "$ENABLE_OTEL" = true ]; then
 fi
 
 # run ingress
+# dynamo.frontend accepts either --http-port flag or DYN_HTTP_PORT env var (defaults to 8000)
 OTEL_SERVICE_NAME=dynamo-frontend \
 python3 -m dynamo.frontend \
- --http-port=8000 \
  --router-mode kv \
  --kv-overlap-score-weight 0 \
  --router-reset-states &
 DYNAMO_PID=$!
 
 # run prefill router
-OTEL_SERVICE_NAME=dynamo-router-prefill DYN_SYSTEM_PORT=8081 \
+OTEL_SERVICE_NAME=dynamo-router-prefill DYN_SYSTEM_PORT=${DYN_SYSTEM_PORT_PREFILL_ROUTER:-8081} \
 python3 -m dynamo.router \
   --endpoint dynamo.prefill.generate \
   --block-size 64 \
@@ -63,7 +63,7 @@ python3 -m dynamo.router \
 PREFILL_ROUTER_PID=$!
 
 # run prefill worker
-OTEL_SERVICE_NAME=dynamo-worker-prefill-1 DYN_SYSTEM_PORT=8082 \
+OTEL_SERVICE_NAME=dynamo-worker-prefill-1 DYN_SYSTEM_PORT=${DYN_SYSTEM_PORT_PREFILL_WORKER1:-8082} \
 python3 -m dynamo.sglang \
   --model-path deepseek-ai/DeepSeek-R1-Distill-Llama-8B \
   --served-model-name deepseek-ai/DeepSeek-R1-Distill-Llama-8B \
@@ -78,7 +78,7 @@ python3 -m dynamo.sglang \
 PREFILL_PID=$!
 
 # run prefill worker
-OTEL_SERVICE_NAME=dynamo-worker-prefill-2 DYN_SYSTEM_PORT=8083 \
+OTEL_SERVICE_NAME=dynamo-worker-prefill-2 DYN_SYSTEM_PORT=${DYN_SYSTEM_PORT_PREFILL_WORKER2:-8083} \
 CUDA_VISIBLE_DEVICES=1 python3 -m dynamo.sglang \
   --model-path deepseek-ai/DeepSeek-R1-Distill-Llama-8B \
   --served-model-name deepseek-ai/DeepSeek-R1-Distill-Llama-8B \
@@ -93,7 +93,7 @@ CUDA_VISIBLE_DEVICES=1 python3 -m dynamo.sglang \
 PREFILL_PID=$!
 
 # run decode worker
-OTEL_SERVICE_NAME=dynamo-worker-decode-1 DYN_SYSTEM_PORT=8084 \
+OTEL_SERVICE_NAME=dynamo-worker-decode-1 DYN_SYSTEM_PORT=${DYN_SYSTEM_PORT_DECODE_WORKER1:-8084} \
 CUDA_VISIBLE_DEVICES=3 python3 -m dynamo.sglang \
   --model-path deepseek-ai/DeepSeek-R1-Distill-Llama-8B \
   --served-model-name deepseek-ai/DeepSeek-R1-Distill-Llama-8B \
@@ -108,7 +108,7 @@ CUDA_VISIBLE_DEVICES=3 python3 -m dynamo.sglang \
 PREFILL_PID=$!
 
 # run decode worker
-OTEL_SERVICE_NAME=dynamo-worker-decode-2 DYN_SYSTEM_PORT=8085 \
+OTEL_SERVICE_NAME=dynamo-worker-decode-2 DYN_SYSTEM_PORT=${DYN_SYSTEM_PORT_DECODE_WORKER2:-8085} \
 CUDA_VISIBLE_DEVICES=2 python3 -m dynamo.sglang \
   --model-path deepseek-ai/DeepSeek-R1-Distill-Llama-8B \
   --served-model-name deepseek-ai/DeepSeek-R1-Distill-Llama-8B \

examples/backends/sglang/launch/disagg_same_gpu.sh

@@ -37,11 +37,12 @@ trap cleanup EXIT INT TERM
 
 
 # run ingress with KV router mode for disaggregated setup
-python3 -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)
+python3 -m dynamo.frontend --router-mode kv &
 DYNAMO_PID=$!
 
 # run prefill worker with metrics on port 8081
-DYN_SYSTEM_PORT=8081 \
+DYN_SYSTEM_PORT=${DYN_SYSTEM_PORT1:-8081} \
 python3 -m dynamo.sglang \
   --model-path Qwen/Qwen3-0.6B \
   --served-model-name Qwen/Qwen3-0.6B \
@@ -71,7 +72,7 @@ echo "Waiting for prefill worker to initialize..."
 sleep 5
 
 # run decode worker with metrics on port 8082 (foreground)
-DYN_SYSTEM_PORT=8082 \
+DYN_SYSTEM_PORT=${DYN_SYSTEM_PORT2:-8082} \
 python3 -m dynamo.sglang \
   --model-path Qwen/Qwen3-0.6B \
   --served-model-name Qwen/Qwen3-0.6B \

examples/backends/sglang/launch/multimodal_agg.sh

@@ -60,7 +60,8 @@ if [[ -n "$SERVED_MODEL_NAME" ]]; then
 fi
 
 # 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 SGLang multimodal processor

examples/backends/sglang/launch/multimodal_disagg.sh

@@ -60,7 +60,8 @@ if [[ -n "$SERVED_MODEL_NAME" ]]; then
 fi
 
 # 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 SGLang multimodal processor

examples/backends/trtllm/launch/agg.sh

@@ -22,7 +22,8 @@ trap cleanup EXIT INT TERM
 
 
 # run frontend
-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 worker

examples/backends/trtllm/launch/agg_metrics.sh

@@ -19,11 +19,12 @@ cleanup() {
 trap cleanup EXIT INT TERM
 
 # Run frontend
-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 worker
-DYN_SYSTEM_PORT=8081 \
+DYN_SYSTEM_PORT=${DYN_SYSTEM_PORT:-8081} \
 python3 -m dynamo.trtllm \
   --model-path "$MODEL_PATH" \
   --served-model-name "$SERVED_MODEL_NAME" \

examples/backends/trtllm/launch/agg_router.sh

@@ -19,7 +19,8 @@ trap cleanup EXIT INT TERM
 
 
 # run frontend
-python3 -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)
+python3 -m dynamo.frontend --router-mode kv &
 DYNAMO_PID=$!
 
 # run worker

examples/backends/trtllm/launch/disagg.sh

@@ -25,7 +25,8 @@ trap cleanup EXIT INT TERM
 
 
 # run frontend
-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 prefill worker

examples/backends/trtllm/launch/disagg_multimodal.sh

@@ -23,7 +23,8 @@ trap cleanup EXIT INT TERM
 
 
 # run frontend
-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 prefill worker

examples/backends/trtllm/launch/disagg_router.sh

@@ -22,7 +22,8 @@ trap cleanup EXIT INT TERM
 
 
 # run frontend with KV routing for cache-aware optimization
-python3 -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)
+python3 -m dynamo.frontend --router-mode kv &
 DYNAMO_PID=$!
 
 # run prefill worker

examples/backends/trtllm/launch/disagg_same_gpu.sh

@@ -48,12 +48,13 @@ trap cleanup EXIT INT TERM
 
 
 # run frontend
-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 prefill worker (shares GPU with decode)
 CUDA_VISIBLE_DEVICES=$CUDA_VISIBLE_DEVICES \
-DYN_SYSTEM_PORT=8081 \
+DYN_SYSTEM_PORT=${DYN_SYSTEM_PORT1:-8081} \
 python3 -m dynamo.trtllm \
   --model-path "$MODEL_PATH" \
   --served-model-name "$SERVED_MODEL_NAME" \
@@ -65,7 +66,7 @@ PREFILL_PID=$!
 
 # run decode worker (shares GPU with prefill)
 CUDA_VISIBLE_DEVICES=$CUDA_VISIBLE_DEVICES \
-DYN_SYSTEM_PORT=8082 \
+DYN_SYSTEM_PORT=${DYN_SYSTEM_PORT2:-8082} \
 python3 -m dynamo.trtllm \
   --model-path "$MODEL_PATH" \
   --served-model-name "$SERVED_MODEL_NAME" \

examples/backends/trtllm/launch/epd_disagg.sh

@@ -29,7 +29,8 @@ trap cleanup EXIT INT TERM
 
 
 # run frontend
-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 encode worker

examples/backends/trtllm/launch/gpt_oss_disagg.sh

@@ -14,7 +14,8 @@ trap 'echo Cleaning up...; kill 0' EXIT
 
 
 # run frontend
-python3 -m dynamo.frontend --router-mode round-robin --http-port 8000 &
+# dynamo.frontend accepts either --http-port flag or DYN_HTTP_PORT env var (defaults to 8000)
+python3 -m dynamo.frontend --router-mode round-robin &
 
 # With tensor_parallel_size=4, each worker needs 4 GPUs
 # run prefill worker

examples/backends/trtllm/performance_sweeps/scripts/start_frontend.sh

@@ -20,6 +20,7 @@ etcd --listen-client-urls http://0.0.0.0:2379 --advertise-client-urls http://0.0
 sleep 2
 
 # 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
 

examples/backends/vllm/launch/agg.sh

@@ -5,9 +5,10 @@ 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
 # --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} \
     python -m dynamo.vllm --model Qwen/Qwen3-0.6B --enforce-eager --connector none

examples/backends/vllm/launch/agg_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 worker with KVBM enabled
 # NOTE: remove --enforce-eager for production use