Add guest load metrics tests.

The new test validates that guest load metrics are properly reported
when the guest-agent is up to date to version 9.6 and above.

Author: OhadRevah

tests/global_config.py

@@ -171,6 +171,12 @@ def _get_default_storage_class(sc_list):
     VirtualMachine.RunStrategy.RERUNONFAILURE,
 ]
 
+guest_load_os_matrix = [
+    "kubevirt_vmi_guest_load_1m",
+    "kubevirt_vmi_guest_load_5m",
+    "kubevirt_vmi_guest_load_15m",
+]
+
 aaq_allocation_methods_matrix = [AAQ_VIRTUAL_RESOURCES, AAQ_VMI_POD_USAGE]
 
 sysprep_source_matrix = ["ConfigMap", "Secret"]

tests/observability/metrics/conftest.py

@@ -1,4 +1,5 @@
 import logging
+import re
 import shlex
 
 import bitmath
@@ -10,6 +11,7 @@
 from ocp_resources.storage_class import StorageClass
 from ocp_resources.virtual_machine import VirtualMachine
 from ocp_resources.virtual_machine_instance_migration import VirtualMachineInstanceMigration
+from pyhelper_utils.shell import run_ssh_commands
 from pytest_testconfig import py_config
 from timeout_sampler import TimeoutExpiredError, TimeoutSampler
 
@@ -37,6 +39,8 @@
 from tests.utils import create_vms
 from utilities import console
 from utilities.constants import (
+    CENTOS_STREAM10_PREFERENCE,
+    DATA_SOURCE_NAME,
     IPV4_STR,
     KUBEVIRT_VMI_MEMORY_PGMAJFAULT_TOTAL,
     KUBEVIRT_VMI_MEMORY_PGMINFAULT_TOTAL,
@@ -58,6 +62,7 @@
     TWO_CPU_THREADS,
     VIRT_TEMPLATE_VALIDATOR,
     Images,
+    StorageClassNames,
 )
 from utilities.hco import ResourceEditorValidateHCOReconcile, enabled_aaq_in_hco
 from utilities.infra import (
@@ -69,6 +74,7 @@
 )
 from utilities.monitoring import get_metrics_value
 from utilities.network import assert_ping_successful, get_ip_from_vm_or_virt_handler_pod, ping
+from utilities.os_utils import generate_linux_instance_type_os_matrix
 from utilities.ssp import verify_ssp_pod_is_running
 from utilities.storage import (
     data_volume_template_with_source_ref_dict,
@@ -619,3 +625,63 @@ def aaq_resource_hard_limit_and_used(application_aware_resource_quota):
         for key, value in resource_used.items()
     }
     return formatted_hard_limit, formatted_used_value
+
+
+@pytest.fixture(scope="class")
+def centos_stream_10_vm(
+    unprivileged_client,
+    namespace,
+    golden_images_namespace,
+    modern_cpu_for_migration,
+):
+    instance_type_centos_os_matrix = generate_linux_instance_type_os_matrix(
+        os_name="centos.stream", preferences=[CENTOS_STREAM10_PREFERENCE]
+    )[0]
+    os_name = next(iter(instance_type_centos_os_matrix))
+    data_source_name = instance_type_centos_os_matrix[os_name][DATA_SOURCE_NAME]
+    with VirtualMachineForTests(
+        client=unprivileged_client,
+        name=f"{data_source_name}-vm-with-instance-type",
+        namespace=namespace.name,
+        vm_instance_type_infer=True,
+        vm_preference_infer=True,
+        data_volume_template=data_volume_template_with_source_ref_dict(
+            data_source=DataSource(
+                name=data_source_name,
+                namespace=golden_images_namespace.name,
+            ),
+            storage_class=StorageClassNames.CEPH_RBD_VIRTUALIZATION,
+        ),
+        cpu_model=modern_cpu_for_migration,
+    ) as vm:
+        running_vm(vm=vm)
+        yield vm
+
+
+@pytest.fixture(scope="class")
+def qemu_guest_agent_version_updated_centos(centos_stream_10_vm):
+    LOGGER.info(f"Checking qemu-guest-agent package on VM: {centos_stream_10_vm.name}")
+    output = run_ssh_commands(
+        host=centos_stream_10_vm.ssh_exec,
+        commands=shlex.split("rpm -q qemu-guest-agent"),
+    )
+    agent_version = output[0].strip() if output else ""
+    LOGGER.info(f"qemu-guest-agent version: {agent_version}")
+
+    # Parse version to ensure it's at least 9.6
+    version_match = re.search(r"qemu-guest-agent-(\d+)\.(\d+)", agent_version)
+    if version_match:
+        version_num = float(f"{version_match.group(1)}.{version_match.group(2)}")
+
+        LOGGER.info(f"Parsed qemu-guest-agent version: {version_num}")
+        if version_num >= 9.6:
+            return
+        raise ValueError(f"qemu-guest-agent version {version_num} is less than required 9.6")
+
+
+@pytest.fixture(scope="class")
+def stressed_vm_cpu_centos(centos_stream_10_vm):
+    run_ssh_commands(
+        host=centos_stream_10_vm.ssh_exec,
+        commands=shlex.split("sudo dnf install stress-ng -y; stress-ng --cpu $(nproc) --timeout 60s"),
+    )

tests/observability/metrics/test_metrics_cpu.py

@@ -3,6 +3,7 @@
 from tests.observability.metrics.utils import (
     ONE_CPU_CORES,
     ZERO_CPU_CORES,
+    validate_metric_value_greater_than_initial_value,
     wait_for_metric_vmi_request_cpu_cores_output,
 )
 
@@ -40,3 +41,20 @@ class TestMetricsCpuErrorState:
     @pytest.mark.s390x
     def test_verify_metrics_error_state_vm(self, prometheus, error_state_vm):
         wait_for_metric_vmi_request_cpu_cores_output(prometheus=prometheus, expected_cpu=ZERO_CPU_CORES)
+
+
+class TestVMIGuestLoad:
+    @pytest.mark.polarion("CNV-12369")
+    def test_kubevirt_vmi_guest_load_centos(
+        self,
+        prometheus,
+        centos_stream_10_vm,
+        qemu_guest_agent_version_updated_centos,
+        stressed_vm_cpu_centos,
+        guest_load_os_matrix__function__,
+    ):
+        validate_metric_value_greater_than_initial_value(
+            prometheus=prometheus,
+            metric_name=f"{guest_load_os_matrix__function__}{{name='{centos_stream_10_vm.name}'}}",
+            initial_value=0,
+        )

tests/observability/metrics/utils.py

@@ -805,10 +805,11 @@ def validate_metric_value_greater_than_initial_value(
         prometheus=prometheus,
         metrics_name=metric_name,
     )
+    sample = None
     try:
         for sample in samples:
             if sample:
-                if int(sample) > initial_value:
+                if float(sample) > initial_value:
                     return
     except TimeoutExpiredError:
         LOGGER.error(f"{sample} should be greater than {initial_value}")