feat: macos llm-d simulator dockerfile

Signed-off-by: Alon Kellner <[email protected]>

Author: AlonKellner-RedHat

tests/e2e/README.md

@@ -6,10 +6,10 @@ The E2E tests in GuideLLM use the [vLLM simulator by llm-d](https://llm-d.ai/doc
 docker build . -f tests/e2e/vllm-sim.Dockerfile -o type=local,dest=./
 ```
 
-On MacOS run:
+For MacOS native:
 
 ```shell
-docker build . -f tests/e2e/vllm-sim.Dockerfile -o type=local,dest=./ --build-arg BUILDOS=darwin
+docker build . -f tests/e2e/vllm-sim-macos.Dockerfile -o type=local,dest=./
 ```
 
 Then to run the tests:

tests/e2e/test_over_saturated_benchmark.py

@@ -39,7 +39,7 @@ def test_over_saturated_benchmark(server: VllmSimServer):
     Another example test interacting with the server.
     """
     report_path = Path("tests/e2e/over_saturated_benchmarks.json")
-    rate = 100
+    rate = 1
 
     # Create and configure the guidellm client
     client = GuidellmClient(target=server.get_url(), output_path=report_path)
@@ -80,7 +80,7 @@ def test_over_saturated_benchmark_with_dict_config(server: VllmSimServer):
     Test over-saturation detection with dictionary configuration instead of boolean.
     """
     report_path = Path("tests/e2e/over_saturated_benchmarks_dict.json")
-    rate = 100
+    rate = 1
 
     # Create and configure the guidellm client
     client = GuidellmClient(target=server.get_url(), output_path=report_path)

tests/e2e/vllm-sim-macos.Dockerfile

@@ -0,0 +1,18 @@
+FROM golang AS base
+
+WORKDIR /app
+
+ARG BUILDOS
+ARG BUILDARCH
+
+RUN apt-get update && \
+    apt-get install -y libzmq3-dev pkg-config && \
+    git clone https://github.com/llm-d/llm-d-inference-sim.git && \
+    cd llm-d-inference-sim && \
+    git checkout v0.3.0 && \
+    GOOS=${BUILDOS} GOARCH=${BUILDARCH} make build
+
+WORKDIR /app/llm-d-inference-sim
+
+FROM scratch
+COPY --from=base /app/llm-d-inference-sim/bin /bin