From 004aff531446c77c68a5caae80aa5c1947467bce Mon Sep 17 00:00:00 2001
From: Alexander Bayandin <alexander@neon.tech>
Date: Wed, 3 Jan 2024 16:51:45 +0000
Subject: [PATCH] test_pageserver: report duration

---
 test_runner/fixtures/benchmark_fixture.py  | 28 ++++++++++++++--------
 test_runner/performance/test_pageserver.py |  5 ++--
 2 files changed, 21 insertions(+), 12 deletions(-)

diff --git a/test_runner/fixtures/benchmark_fixture.py b/test_runner/fixtures/benchmark_fixture.py
index f94cc9979c..1f08c4ed26 100644
--- a/test_runner/fixtures/benchmark_fixture.py
+++ b/test_runner/fixtures/benchmark_fixture.py
@@ -410,34 +410,42 @@ class NeonBenchmarker:
             report=MetricReport.LOWER_IS_BETTER,
         )
 
-    def record_pagebench_results(self, name: str, results: Dict[str, Any]):
+    def record_pagebench_results(self, name: str, results: Dict[str, Any], duration: str):
         total = results["total"]
 
         metric = "request_count"
         self.record(
-            f"{name}.{metric}",
-            total[metric],
-            "",
+            metric_name=f"{name}.{metric}",
+            metric_value=total[metric],
+            unit="",
             report=MetricReport.HIGHER_IS_BETTER,
         )
 
         metric = "latency_mean"
         self.record(
-            f"{name}.{metric}",
-            humantime_to_ms(total[metric]),
-            "ms",
+            metric_name=f"{name}.{metric}",
+            metric_value=humantime_to_ms(total[metric]),
+            unit="ms",
             report=MetricReport.LOWER_IS_BETTER,
         )
 
         metric = "latency_percentiles"
         for k, v in total[metric].items():
             self.record(
-                f"{name}.{metric}.{k}",
-                humantime_to_ms(v),
-                "ms",
+                metric_name=f"{name}.{metric}.{k}",
+                metric_value=humantime_to_ms(v),
+                unit="ms",
                 report=MetricReport.LOWER_IS_BETTER,
             )
 
+        metric = "duration"
+        self.record(
+            metric_name=f"{name}.{metric}",
+            metric_value=humantime_to_ms(duration) / 1000,
+            unit="s",
+            report=MetricReport.LOWER_IS_BETTER,
+        )
+
 
 @pytest.fixture(scope="function")
 def zenbenchmark(record_property: Callable[[str, object], None]) -> Iterator[NeonBenchmarker]:
diff --git a/test_runner/performance/test_pageserver.py b/test_runner/performance/test_pageserver.py
index 5a1342380c..3d91a52b1e 100644
--- a/test_runner/performance/test_pageserver.py
+++ b/test_runner/performance/test_pageserver.py
@@ -152,6 +152,7 @@ def test_getpage_throughput(
     ps_http = env.pageserver.http_client()
 
     # run the benchmark with one client per timeline, each doing 10k requests to random keys.
+    duration = "10s"
     cmd = [
         str(env.neon_binpath / "pagebench"),
         "get-page-latest-lsn",
@@ -160,7 +161,7 @@ def test_getpage_throughput(
         "--page-service-connstring",
         env.pageserver.connstr(password=None),
         "--runtime",
-        "10s",
+        duration,
         *[f"{tenant}/{template_timeline}" for tenant in tenants],
     ]
     log.info(f"command: {' '.join(cmd)}")
@@ -173,4 +174,4 @@ def test_getpage_throughput(
 
     log.info(f"Results:\n{json.dumps(results, sort_keys=True, indent=2)}")
 
-    zenbenchmark.record_pagebench_results("get-page-latest-lsn", results)
+    zenbenchmark.record_pagebench_results("get-page-latest-lsn", results, duration)