load test schema

asciiart.txt

+-----------------------+      HTTP Requests      +---------------------------------+      Trace Data       +-------------------+
| Client de Load Test   | -------------------> |   vLLM OpenAI API Endpoint      | -------------------> |   Langfuse Cloud  |
| (k6, Locust, script)  |      (Load Profile)      |   (Tournant sur H100 ou L40)    |      (SDK/API Call)     |   (UI & API)      |
|                       |                        |   (Modèle: Qwen / Llama3)       |                     |                   |
| - Génère QPS          | <------------------- |                                 | <------------------- |   - Visualisation |
| - Mesure Latence (Client) |      Responses         | - Traite les requêtes           |      User Query     |   - Analyse       |
| - Mesure Taux Succès  |                        | - Mesure Latence (Server)       |                     |   - Export        |
+-----------------------+                        | - Mesure Tokens In/Out          |                     +-------------------+
                                                 +---------------------------------+
                                                          |        ^
                                                          |        | (Optionnel)
                                                 nvidia-smi query  | GPU Metrics (Util, Mem)
                                                          V        |
                                                      +-------+
                                                      |  GPU  |
                                                      +-------+