[V1][Metrics] Handle preemptions (#13169)

2025-02-27 04:04:59 +00:00
parent 378b3ef6f8
commit cd711c48b2
5 changed files with 48 additions and 19 deletions
--- a/tests/entrypoints/openai/test_metrics.py
+++ b/tests/entrypoints/openai/test_metrics.py
@@ -227,6 +227,7 @@ EXPECTED_METRICS_V1 = [
    "vllm:gpu_cache_usage_perc",
    "vllm:gpu_prefix_cache_queries",
    "vllm:gpu_prefix_cache_hits",
+    "vllm:num_preemptions_total",
    "vllm:prompt_tokens_total",
    "vllm:generation_tokens_total",
    "vllm:iteration_tokens_total",