OCPBUGS-80952: perf: latenyc: compute memory resources dynamically

test/e2e/performanceprofile/functests/4_latency/latency.go

@@ -43,6 +43,8 @@ const (
 	defaultTestRuntime   = "300"
 	defaultMaxLatency    = -1
 	defaultTestCpus      = -1
+	defaultTestMemory    = "1Gi"
+	dynamicMemory        = "dynamic"
 	minCpuAmountForOslat = 2
 )
 
@@ -51,13 +53,14 @@ var (
 	latencyTestRuntime = defaultTestRuntime
 	maximumLatency     = defaultMaxLatency
 	latencyTestCpus    = defaultTestCpus
+	latencyTestMemory  = defaultTestMemory
 )
 
 // LATENCY_TEST_DELAY delay the run of the binary, can be useful to give time to the CPU manager reconcile loop
 // to update the default CPU pool
 // LATENCY_TEST_RUNTIME: the amount of time in seconds that the latency test should run
 // LATENCY_TEST_CPUS: the amount of CPUs the pod which run the latency test should request
-
+// LATENCY_TEST_MEMORY: the amount of memory the pod which run the latency test should request
 var _ = Describe("[performance] Latency Test", Ordered, func() {
 	var workerRTNode *corev1.Node
 	var profile *performancev2.PerformanceProfile
@@ -278,6 +281,29 @@ func getLatencyTestCpus() (int, error) {
 	return defaultTestCpus, nil
 }
 
+func getLatencyTestMemory(cpus int) (string, error) {
+	if val, ok := os.LookupEnv("LATENCY_TEST_MEMORY"); ok {
+		if val == dynamicMemory {
+			// Defensive check: fall back to default memory if the cpus was not normalized
+			if cpus == defaultTestCpus {
+				return defaultTestMemory, nil
+			}
+			// 32Mi per requested CPU should be reasonable for the test; if needed more, it can
+			// be overridden by explicitly setting the environment variable
+			memoryFactor := 32
+			memory := memoryFactor * cpus
+			return fmt.Sprintf("%dMi", memory), nil
+		}
+
+		_, err := resource.ParseQuantity(val)
+		if err != nil {
+			return val, fmt.Errorf("the environment variable LATENCY_TEST_MEMORY has incorrect value %q, it must be a valid quantity: %w", val, err)
+		}
+		return val, nil
+	}
+	return defaultTestMemory, nil
+}
+
 // getMaximumLatency should look for one of the following environment variables:
 // OSLAT_MAXIMUM_LATENCY: the expected maximum latency for all buckets in us
 // CYCLICTEST_MAXIMUM_LATENCY: the expected maximum latency for all buckets in us
@@ -325,6 +351,10 @@ func getLatencyTestPod(profile *performancev2.PerformanceProfile, node *corev1.N
 		latencyTestCpus = cpus.Size() - 1
 	}
 
+	var err error
+	latencyTestMemory, err = getLatencyTestMemory(latencyTestCpus)
+	Expect(err).ToNot(HaveOccurred())
+
 	latencyTestRunnerArgs := []string{
 		"-logtostderr=false",
 		"-alsologtostderr=true",
@@ -361,7 +391,7 @@ func getLatencyTestPod(profile *performancev2.PerformanceProfile, node *corev1.N
 					Resources: corev1.ResourceRequirements{
 						Limits: corev1.ResourceList{
 							corev1.ResourceCPU:    resource.MustParse(strconv.Itoa(latencyTestCpus)),
-							corev1.ResourceMemory: resource.MustParse("1Gi"),
+							corev1.ResourceMemory: resource.MustParse(latencyTestMemory),
 						},
 					},
 					SecurityContext: &corev1.SecurityContext{