Fix failed request count

Signed-off-by: Rashid Kaleem <230885705+arekay-nv@users.noreply.github.com>

Author: arekay-nv

src/inference_endpoint/commands/benchmark.py

@@ -697,7 +697,7 @@ def signal_handler(signum, frame):
                     "results": {
                         "total": total,
                         "successful": success_count,
-                        "failed": total - success_count,
+                        "failed": report.n_samples_failed,
                         "elapsed_time": elapsed_time,
                         "qps": estimated_qps,
                     },

src/inference_endpoint/metrics/reporter.py

@@ -354,6 +354,7 @@ class Report:
     test_started_at: int
     n_samples_issued: int
     n_samples_completed: int
+    n_samples_failed: int
     duration_ns: int
 
     # For the following metrics, the key is a rollup statistic (i.e. mean, median, etc.)
@@ -495,6 +496,7 @@ def display(
         )
         fn(f"Total samples issued: {self.n_samples_issued}{newline}")
         fn(f"Total samples completed: {self.n_samples_completed}{newline}")
+        fn(f"Total samples failed: {self.n_samples_failed}{newline}")
         if self.duration_ns is not None:
             fn(f"Duration: {self.duration_ns / 1e9:.2f} seconds{newline}")
         else:
@@ -1312,6 +1314,7 @@ def create_report(
             test_started_at=test_started_at,
             n_samples_issued=sample_statuses["total_sent"],
             n_samples_completed=sample_statuses["completed"],
+            n_samples_failed=self.get_error_count(),
             duration_ns=self.derive_duration(),
             ttft=ttft_summary,
             tpot=tpot_summary,

tests/unit/metrics/test_reporter.py

@@ -388,6 +388,7 @@ def test_reporter_create_report(events_db, fake_outputs, tokenizer):
 
     assert report.n_samples_issued == 3
     assert report.n_samples_completed == 2
+    assert report.n_samples_failed == 3  # 3 ERROR events for uuid3 in fixture
     assert report.duration_ns == (10300 - 5000)
 
     for k, expected in ttft_rollup.summarize().items():
@@ -428,6 +429,7 @@ def test_reporter_json(events_db):
         "git_sha",
         "n_samples_issued",
         "n_samples_completed",
+        "n_samples_failed",
         "duration_ns",
         "ttft",
         "tpot",
@@ -441,6 +443,7 @@ def test_reporter_json(events_db):
     assert set(json_dict.keys()) == set(expected_keys)
     assert json_dict["n_samples_issued"] == report.n_samples_issued
     assert json_dict["n_samples_completed"] == report.n_samples_completed
+    assert json_dict["n_samples_failed"] == report.n_samples_failed
     assert json_dict["duration_ns"] == report.duration_ns
     assert json_dict["qps"] == report.qps
     assert json_dict["tps"] == report.tps