test_cpu_mem_from_psutil fails on single-CPU systems

[sanvila]

Hello.

When I try to build (the Debian package of) this project (version 6.25.0) on AWS machines of type m7a.medium or r7a.medium, which incidentally have a single CPU, I always (i.e. every time I try) get this:

=================================== FAILURES ===================================
___________________________ test_cpu_mem_from_psutil ___________________________

elasticapm_client = <tests.fixtures.TempStoreClient object at 0x7f6ed00a8aa0>

    def test_cpu_mem_from_psutil(elasticapm_client):
        metricset = cpu_psutil.CPUMetricSet(MetricsRegistry(elasticapm_client))
        # do something that generates some CPU load
        for i in range(10**6):
            j = i * i
        data = next(metricset.collect())
        # we can't really test any specific values here as it depends on the system state.
        # Mocking is also not really a viable choice, as we would then lose the "integration testing"
        # nature of this test with different versions of psutil
>       assert 0 < data["samples"]["system.cpu.total.norm.pct"]["value"] < 1
E       assert 1.0 < 1

tests/metrics/cpu_psutil_tests.py:50: AssertionError

To reproduce I suggest configuring GRUB with GRUB_CMDLINE_LINUX="nr_cpus=1".
If that does not work I can offer a VM where this happens all the time.

Thanks.

[xrmx]

Looks like the upper limit can be 1 so the check should be <= 1

[sanvila]

I agree. Would this be the correct fix, then?

--- a/tests/metrics/cpu_psutil_tests.py
+++ b/tests/metrics/cpu_psutil_tests.py
@@ -47,8 +47,8 @@ def test_cpu_mem_from_psutil(elasticapm_client):
     # we can't really test any specific values here as it depends on the system state.
     # Mocking is also not really a viable choice, as we would then lose the "integration testing"
     # nature of this test with different versions of psutil
-    assert 0 < data["samples"]["system.cpu.total.norm.pct"]["value"] < 1
-    assert 0 < data["samples"]["system.process.cpu.total.norm.pct"]["value"] < 1
+    assert 0 < data["samples"]["system.cpu.total.norm.pct"]["value"] <= 1
+    assert 0 < data["samples"]["system.process.cpu.total.norm.pct"]["value"] <= 1
 
     assert data["samples"]["system.memory.total"]["value"] > 0
     assert data["samples"]["system.memory.actual.free"]["value"] > 0

[xrmx]

LGTM

[sanvila]

Hi. Apparently the change above is not enough. Today's Debian upload, versioned 6.25.0-2, incorporates the first <= change but not the second one, and now I get failures like this one more than half of the time:

>       assert 0 < data["samples"]["system.process.cpu.total.norm.pct"]["value"] < 1
E       assert 1.212 < 1

After many tries, I get (among the failing cases) values from E assert 1.0 < 1 to E assert 1.348 < 1, so I assume that changing the second assert from < 1 to <= 1 would not help here.