Test against existing endpoint

automation/test-execution/ansible/inventory/README.md

@@ -34,6 +34,12 @@ export LOADGEN_HOSTNAME=192.168.1.200
 export ANSIBLE_SSH_USER=ec2-user
 export ANSIBLE_SSH_KEY=~/.ssh/my-key.pem
 export HF_TOKEN=hf_xxxxx  # If using gated models
+
+# Optional: Use external vLLM endpoint instead of managed container
+export VLLM_ENDPOINT_MODE=external
+export VLLM_ENDPOINT_URL=http://my-vllm-instance.example.com:8000
+export VLLM_API_KEY_ENABLED=true  # If endpoint requires authentication
+export VLLM_API_KEY=your-api-key
 ```
 
 The inventory file automatically uses these environment variables with sensible defaults if not set.
@@ -198,6 +204,38 @@ The framework will automatically skip token setup for these models.
 
 ### Scenario 3: External vLLM Endpoint (AWS/K8s)
 
+**Option A: Environment Variables (Recommended)**
+
+```bash
+# Load generator connection (required - GuideLLM runs here)
+export LOADGEN_HOSTNAME=192.168.1.200  # or your load generator hostname/IP
+export ANSIBLE_SSH_USER=ec2-user
+export ANSIBLE_SSH_KEY=~/.ssh/my-key.pem
+
+# External vLLM endpoint (replaces DUT configuration)
+export VLLM_ENDPOINT_MODE=external
+export VLLM_ENDPOINT_URL=http://my-vllm-lb.example.com:8000
+
+# Optional: If endpoint requires API key authentication
+export VLLM_API_KEY_ENABLED=true
+export VLLM_API_KEY=your-api-key
+
+# Optional: HuggingFace token (if needed)
+export HF_TOKEN=hf_xxxxx
+```
+
+**Note:** In external mode, you only need load generator access. DUT connection is not required since vLLM is already running externally.
+
+Then run tests **without specifying `test_model`** - it will be auto-detected from the endpoint:
+
+```bash
+ansible-playbook -i inventory/hosts.yml llm-benchmark-concurrent-load.yml \
+  -e "base_workload=chat" \
+  -e "requested_cores=16"
+```
+
+**Option B: Edit Configuration File**
+
 ```yaml
 # group_vars/all/endpoints.yml
 vllm_endpoint:
@@ -210,7 +248,7 @@ vllm_endpoint:
       env_var: "VLLM_API_KEY"
 ```
 
-Then run tests normally - playbooks will skip vLLM container management.
+Then run tests normally - playbooks will skip vLLM container management and connect to your external endpoint.
 
 ### Scenario 4: Host-Based Execution (No Containers)
 

automation/test-execution/ansible/inventory/group_vars/all/endpoints.yml

@@ -16,21 +16,25 @@ vllm_endpoint:
   # Mode: "managed" (default) or "external"
   # - managed: Ansible starts/stops vLLM container on DUT (traditional behavior)
   # - external: Use existing vLLM endpoint (skip container management)
-  mode: "managed"
+  # Can be set via environment variable: export VLLM_ENDPOINT_MODE=external
+  mode: "{{ lookup('env', 'VLLM_ENDPOINT_MODE') | default('managed', true) }}"
 
   # External endpoint configuration (only used when mode=external)
   external:
     # Full URL to vLLM endpoint (required when mode=external)
+    # Can be set via environment variable:
+    #   export VLLM_ENDPOINT_URL=http://my-vllm-instance.example.com:8000
     # Examples:
     #   - http://my-vllm-instance.example.com:8000
     #   - https://vllm-prod.company.com
     #   - http://10.0.1.50:8000 (direct IP)
     #   - http://vllm-lb.k8s.local:8000 (Kubernetes LoadBalancer)
-    url: null
+    url: "{{ lookup('env', 'VLLM_ENDPOINT_URL') | default(None, true) }}"
 
     # Optional API key authentication for secured endpoints
     api_key:
-      enabled: false
+      # Can be enabled via environment variable: export VLLM_API_KEY_ENABLED=true
+      enabled: "{{ lookup('env', 'VLLM_API_KEY_ENABLED') | default('false', true) | bool }}"
       source: "env"              # Options: env, file, vault, prompt
       env_var: "VLLM_API_KEY"    # Used when source=env
       file_path: null            # Used when source=file (e.g., /etc/secrets/vllm-key)