feat: GPU support for Service through ECS Managed Instances

examples/aws-ecs-gpus/.dockerignore

@@ -0,0 +1,6 @@
+
+# sst
+.sst
+node_modules
+__pycache__
+*.pyc

examples/aws-ecs-gpus/Dockerfile

@@ -0,0 +1,13 @@
+FROM python:3.11-slim
+
+WORKDIR /code
+
+COPY requirements.txt /code/requirements.txt
+
+RUN pip install --no-cache-dir --upgrade -r /code/requirements.txt
+
+COPY app.py /code/app.py
+
+EXPOSE 8000
+
+CMD ["fastapi", "run", "app.py", "--host", "0.0.0.0", "--port", "8000"]

examples/aws-ecs-gpus/app.py

@@ -0,0 +1,34 @@
+from pathlib import Path
+import glob
+import os
+
+from fastapi import FastAPI
+
+
+app = FastAPI()
+
+
+def read_file(path: str):
+    try:
+        return Path(path).read_text(encoding="utf-8").strip()
+    except FileNotFoundError:
+        return None
+
+
+@app.get("/health")
+def health():
+    return {"ok": True}
+
+
+@app.get("/")
+def index():
+    return {
+        "message": "hello from ecs managed instances",
+        "gpu": {
+            "visibleDevicesEnv": os.getenv("NVIDIA_VISIBLE_DEVICES"),
+            "deviceFiles": sorted(glob.glob("/dev/nvidia*")),
+            "procGpus": sorted(glob.glob("/proc/driver/nvidia/gpus/*")),
+            "driverVersion": read_file("/proc/driver/nvidia/version"),
+            "cudaVersion": os.getenv("CUDA_VERSION"),
+        },
+    }

examples/aws-ecs-gpus/package.json

@@ -0,0 +1,9 @@
+{
+  "name": "aws-ecs-gpus",
+  "version": "1.0.0",
+  "private": true,
+  "type": "module",
+  "dependencies": {
+    "sst": "^4"
+  }
+}

examples/aws-ecs-gpus/requirements.txt

@@ -0,0 +1 @@
+fastapi[standard]>=0.115.0,<1.0.0

examples/aws-ecs-gpus/sst.config.ts

@@ -0,0 +1,45 @@
+/// <reference path="./.sst/platform/config.d.ts" />
+
+/**
+ * ## AWS ECS GPUs
+ *
+ * A minimal ECS service running on ECS Managed Instances with a GPU-enabled host.
+ * The service uses top-level `gpu`, `cpu`, `memory`, and `storage` settings, while
+ * the managed instances IAM resources remain customizable through `transform`.
+ *
+ * A private API Gateway HTTP API is used to test the service without exposing a public
+ * load balancer.
+ */
+export default $config({
+  app(input) {
+    return {
+      name: "service-gpu-example",
+      removal: input?.stage === "production" ? "retain" : "remove",
+      home: "aws",
+    };
+  },
+  async run() {
+    const vpc = new sst.aws.Vpc("MyVpc");
+    const cluster = new sst.aws.Cluster("MyCluster", { vpc });
+
+    // Provisions g4dn.xlarge
+    const service = new sst.aws.Service("MyService", {
+      cluster,
+      image: { context: "./" },
+      gpu: "nvidia/t4",
+      cpu: "4 vCPU",
+      memory: "10 GB",
+      serviceRegistry: {
+        port: 8000,
+      },
+    });
+
+    const api = new sst.aws.ApiGatewayV2("MyApi", { vpc });
+    api.routePrivate("$default", service.nodes.cloudmapService.arn);
+
+    return {
+      service: service.service,
+      api: api.url,
+    };
+  },
+});