From 2fc473dcf2456c703892e796dd568312da0e10a8 Mon Sep 17 00:00:00 2001
From: DABH <DABH@users.noreply.github.com>
Date: Thu, 10 Sep 2026 00:25:00 -0500
Subject: [PATCH] Bound workflow task latency metric by wall-clock time in
 metrics test

test_workflow_metrics_other_types asserted that the workflow task execution latency was under 3s (seconds format) and over 3ms (milliseconds format) as a sanity check of the duration unit. On a loaded CI runner the first workflow task of the HelloWorkflow run took over 3s, failing the seconds assertion ('assert False' from 'any(...)'). Bound the recorded latency by how long the worker actually ran instead, which still distinguishes the two duration formats without assuming a fast machine.
---
 tests/worker/test_workflow.py | 17 ++++++++++-------
 1 file changed, 10 insertions(+), 7 deletions(-)

diff --git a/tests/worker/test_workflow.py b/tests/worker/test_workflow.py
index 42ba0b69c..d6024dfd2 100644
--- a/tests/worker/test_workflow.py
+++ b/tests/worker/test_workflow.py
@@ -5343,11 +5343,12 @@ async def test_workflow_buffered_metrics(client: Client, env: WorkflowEnvironmen
 
 
 async def test_workflow_metrics_other_types(env: WorkflowEnvironment):
-    async def do_stuff(buffer: MetricBuffer) -> None:
+    async def do_stuff(buffer: MetricBuffer) -> float:
         runtime = Runtime(telemetry=TelemetryConfig(metrics=buffer))
         new_client = await env.connect_client(
             runtime=runtime,
         )
+        start = time.monotonic()
         async with new_worker(new_client, HelloWorkflow) as worker:
             await new_client.execute_workflow(
                 HelloWorkflow.run,
@@ -5355,21 +5356,23 @@ async def do_stuff(buffer: MetricBuffer) -> None:
                 id=f"wf-{uuid.uuid4()}",
                 task_queue=worker.task_queue,
             )
+        worker_seconds = time.monotonic() - start
         # Also, add some manual types beyond the defaults tested in other tests
         runtime.metric_meter.create_histogram_float("my-histogram-float").record(1.23)
         runtime.metric_meter.create_histogram_timedelta(
             "my-histogram-timedelta"
         ).record(timedelta(days=2, seconds=3, milliseconds=4))
         runtime.metric_meter.create_gauge_float("my-gauge-float").set(4.56)
+        return worker_seconds
 
     # Create a buffer, do stuff, check the metrics
     buffer = MetricBuffer(10000)
-    await do_stuff(buffer)
+    worker_seconds = await do_stuff(buffer)
     updates = buffer.retrieve_updates()
     assert any(
         u.metric.name == "temporal_workflow_task_execution_latency"
-        # Took more than 3ms
-        and u.value > 3
+        # Milliseconds, bounded by how long the worker ran
+        and 0 < u.value <= worker_seconds * 1000
         and isinstance(u.value, int)
         and u.metric.unit == "ms"
         for u in updates
@@ -5396,12 +5399,12 @@ async def do_stuff(buffer: MetricBuffer) -> None:
 
     # Do it again with seconds
     buffer = MetricBuffer(10000, duration_format=MetricBufferDurationFormat.SECONDS)
-    await do_stuff(buffer)
+    worker_seconds = await do_stuff(buffer)
     updates = buffer.retrieve_updates()
     assert any(
         u.metric.name == "temporal_workflow_task_execution_latency"
-        # Took less than 3s
-        and u.value < 3
+        # Seconds, bounded by how long the worker ran
+        and 0 < u.value <= worker_seconds
         and isinstance(u.value, float)
         and u.metric.unit == "s"
         for u in updates