|
| 1 | +//go:build integration |
| 2 | + |
| 3 | +package integration_test |
| 4 | + |
| 5 | +import ( |
| 6 | + "context" |
| 7 | + "encoding/json" |
| 8 | + "fmt" |
| 9 | + "path/filepath" |
| 10 | + "testing" |
| 11 | + "time" |
| 12 | + |
| 13 | + ctrl "sigs.k8s.io/controller-runtime" |
| 14 | + ctrlconfig "sigs.k8s.io/controller-runtime/pkg/config" |
| 15 | + |
| 16 | + metav1 "k8s.io/apimachinery/pkg/apis/meta/v1" |
| 17 | + "k8s.io/apimachinery/pkg/types" |
| 18 | + "k8s.io/apimachinery/pkg/util/intstr" |
| 19 | + k8sscheme "k8s.io/client-go/kubernetes/scheme" |
| 20 | + "sigs.k8s.io/controller-runtime/pkg/envtest" |
| 21 | + logf "sigs.k8s.io/controller-runtime/pkg/log" |
| 22 | + "sigs.k8s.io/controller-runtime/pkg/log/zap" |
| 23 | + metricsserver "sigs.k8s.io/controller-runtime/pkg/metrics/server" |
| 24 | + |
| 25 | + spannerv1alpha1 "github.com/mercari/spanner-autoscaler/api/v1alpha1" |
| 26 | + spannerv1beta1 "github.com/mercari/spanner-autoscaler/api/v1beta1" |
| 27 | + "github.com/mercari/spanner-autoscaler/internal/controller" |
| 28 | + webhookv1beta1 "github.com/mercari/spanner-autoscaler/internal/webhook/v1beta1" |
| 29 | +) |
| 30 | + |
| 31 | +// TestController_ScaledownStepSize_Percent observes, end-to-end via the |
| 32 | +// emulators, the scale-down step size when scaledownStepSize is set to "10%". |
| 33 | +// |
| 34 | +// It contrasts two paths: |
| 35 | +// - applyWebhookDefault=true: apply the mutating webhook Default() before |
| 36 | +// creation, exactly like production, to check whether "10%" is preserved |
| 37 | +// rather than being overwritten with 2000. |
| 38 | +// - applyWebhookDefault=false: pass "10%" straight to the controller without |
| 39 | +// going through Default(), to observe the step-size resolution logic alone. |
| 40 | +// |
| 41 | +// In every case the instance starts at initPU under a low CPU load so it scales |
| 42 | +// down, and the observed PU sequence (and each step delta) is recorded. |
| 43 | +func TestController_ScaledownStepSize_Percent(t *testing.T) { |
| 44 | + cases := []struct { |
| 45 | + name string |
| 46 | + projectID string |
| 47 | + instanceID string |
| 48 | + saName string |
| 49 | + initPU int |
| 50 | + maxPU int |
| 51 | + applyWebhookDefault bool |
| 52 | + }{ |
| 53 | + { |
| 54 | + name: "with_webhook_default_production_path", |
| 55 | + projectID: "sd-def-project", |
| 56 | + instanceID: "sd-def-instance", |
| 57 | + saName: "sd-def-sa", |
| 58 | + initPU: 10000, |
| 59 | + maxPU: 10000, |
| 60 | + applyWebhookDefault: true, |
| 61 | + }, |
| 62 | + { |
| 63 | + name: "without_webhook_default_direct_10pct", |
| 64 | + projectID: "sd-raw-project", |
| 65 | + instanceID: "sd-raw-instance", |
| 66 | + saName: "sd-raw-sa", |
| 67 | + initPU: 10000, |
| 68 | + maxPU: 10000, |
| 69 | + applyWebhookDefault: false, |
| 70 | + }, |
| 71 | + { |
| 72 | + // Post-fix check: even on the production-equivalent path (after the |
| 73 | + // webhook Default() is applied), "10%" is preserved and the instance |
| 74 | + // scales down by roughly 10% of the current PU from 35000. |
| 75 | + name: "from_35000_with_webhook_default", |
| 76 | + projectID: "sd-35k-project", |
| 77 | + instanceID: "sd-35k-instance", |
| 78 | + saName: "sd-35k-sa", |
| 79 | + initPU: 35000, |
| 80 | + maxPU: 35000, |
| 81 | + applyWebhookDefault: true, |
| 82 | + }, |
| 83 | + } |
| 84 | + |
| 85 | + for _, tc := range cases { |
| 86 | + tc := tc |
| 87 | + t.Run(tc.name, func(t *testing.T) { |
| 88 | + seq, effective := runScaledownScenario(t, tc.projectID, tc.instanceID, tc.saName, tc.initPU, tc.maxPU, tc.applyWebhookDefault) |
| 89 | + t.Logf("[%s] effective scaledownStepSize on the object = %q", tc.name, effective) |
| 90 | + t.Logf("[%s] observed PU sequence = %v", tc.name, seq) |
| 91 | + for i := 1; i < len(seq); i++ { |
| 92 | + t.Logf("[%s] step %d: %d -> %d (delta=%d)", |
| 93 | + tc.name, i, seq[i-1], seq[i], seq[i-1]-seq[i]) |
| 94 | + } |
| 95 | + }) |
| 96 | + } |
| 97 | +} |
| 98 | + |
| 99 | +func runScaledownScenario(t *testing.T, projectID, instanceID, saName string, initPU, maxPU int, applyWebhookDefault bool) (seq []int, effectiveStepSize string) { |
| 100 | + t.Helper() |
| 101 | + |
| 102 | + const ( |
| 103 | + referenceCPU = 0.10 // Workload = 0.10 * initPU; cpu = Workload / PU |
| 104 | + syncInterval = 1 * time.Second |
| 105 | + scaleUpInterval = 1 * time.Second |
| 106 | + scaleDownInterval = 1 * time.Second |
| 107 | + ) |
| 108 | + targetCPUVal := 40 |
| 109 | + targetCPU := &targetCPUVal |
| 110 | + |
| 111 | + logf.SetLogger(zap.New(zap.UseDevMode(true))) |
| 112 | + |
| 113 | + // Configure a low CPU load on the total metric to trigger scale-down. |
| 114 | + body, _ := json.Marshal(map[string]interface{}{ |
| 115 | + "total": map[string]interface{}{ |
| 116 | + "cpu_utilization": referenceCPU, |
| 117 | + "reference_processing_units": initPU, |
| 118 | + }, |
| 119 | + }) |
| 120 | + adminPUT(t, fmt.Sprintf("/workload/%s/%s", projectID, instanceID), body) |
| 121 | + t.Cleanup(func() { adminDELETE(t, fmt.Sprintf("/workload/%s/%s", projectID, instanceID)) }) |
| 122 | + |
| 123 | + createSpannerInstance(t, projectID, instanceID, initPU) |
| 124 | + |
| 125 | + testEnv := &envtest.Environment{ |
| 126 | + CRDDirectoryPaths: []string{filepath.Join(repoRoot(), "config", "crd", "bases")}, |
| 127 | + ErrorIfCRDPathMissing: true, |
| 128 | + } |
| 129 | + cfg, err := testEnv.Start() |
| 130 | + if err != nil { |
| 131 | + t.Fatalf("failed to start envtest: %v", err) |
| 132 | + } |
| 133 | + t.Cleanup(func() { testEnv.Stop() }) //nolint:errcheck |
| 134 | + |
| 135 | + if err := spannerv1alpha1.AddToScheme(k8sscheme.Scheme); err != nil { |
| 136 | + t.Fatalf("add v1alpha1 scheme: %v", err) |
| 137 | + } |
| 138 | + if err := spannerv1beta1.AddToScheme(k8sscheme.Scheme); err != nil { |
| 139 | + t.Fatalf("add v1beta1 scheme: %v", err) |
| 140 | + } |
| 141 | + |
| 142 | + skipValidation := true |
| 143 | + mgr, err := ctrl.NewManager(cfg, ctrl.Options{ |
| 144 | + Scheme: k8sscheme.Scheme, |
| 145 | + Metrics: metricsserver.Options{BindAddress: "0"}, |
| 146 | + Controller: ctrlconfig.Controller{SkipNameValidation: &skipValidation}, |
| 147 | + }) |
| 148 | + if err != nil { |
| 149 | + t.Fatalf("failed to create manager: %v", err) |
| 150 | + } |
| 151 | + |
| 152 | + reconciler := controller.NewSpannerAutoscalerReconciler( |
| 153 | + mgr.GetClient(), |
| 154 | + mgr.GetAPIReader(), |
| 155 | + mgr.GetScheme(), |
| 156 | + mgr.GetEventRecorderFor("sd-controller"), |
| 157 | + logf.Log.WithName("sd"), |
| 158 | + controller.WithSpannerEndpoint(spannerEmulatorAddr()), |
| 159 | + controller.WithMetricsEndpoint(monitoringGRPCAddr()), |
| 160 | + controller.WithSyncInterval(syncInterval), |
| 161 | + controller.WithScaleUpInterval(scaleUpInterval), |
| 162 | + controller.WithScaleDownInterval(scaleDownInterval), |
| 163 | + ) |
| 164 | + if err := reconciler.SetupWithManager(mgr); err != nil { |
| 165 | + t.Fatalf("failed to setup controller: %v", err) |
| 166 | + } |
| 167 | + |
| 168 | + mgrCtx, mgrCancel := context.WithCancel(context.Background()) |
| 169 | + t.Cleanup(mgrCancel) |
| 170 | + t.Cleanup(reconciler.StopAll) |
| 171 | + go func() { |
| 172 | + if err := mgr.Start(mgrCtx); err != nil { |
| 173 | + t.Logf("manager exited: %v", err) |
| 174 | + } |
| 175 | + }() |
| 176 | + |
| 177 | + k8sClient := mgr.GetClient() |
| 178 | + ctx := context.Background() |
| 179 | + nn := types.NamespacedName{Namespace: "default", Name: saName} |
| 180 | + |
| 181 | + sa := &spannerv1beta1.SpannerAutoscaler{ |
| 182 | + ObjectMeta: metav1.ObjectMeta{ |
| 183 | + Name: nn.Name, |
| 184 | + Namespace: nn.Namespace, |
| 185 | + }, |
| 186 | + Spec: spannerv1beta1.SpannerAutoscalerSpec{ |
| 187 | + TargetInstance: spannerv1beta1.TargetInstance{ |
| 188 | + ProjectID: projectID, |
| 189 | + InstanceID: instanceID, |
| 190 | + }, |
| 191 | + Authentication: spannerv1beta1.Authentication{ |
| 192 | + Type: spannerv1beta1.AuthTypeADC, |
| 193 | + }, |
| 194 | + ScaleConfig: spannerv1beta1.ScaleConfig{ |
| 195 | + ComputeType: spannerv1beta1.ComputeTypePU, |
| 196 | + ProcessingUnits: spannerv1beta1.ScaleConfigPUs{ |
| 197 | + Min: 100, |
| 198 | + Max: maxPU, |
| 199 | + }, |
| 200 | + // User intent: scale down by 10% of the current PU per step. |
| 201 | + ScaledownStepSize: intstr.FromString("10%"), |
| 202 | + ScaleupStepSize: intstr.FromInt(1000), |
| 203 | + TargetCPUUtilization: spannerv1beta1.TargetCPUUtilization{ |
| 204 | + Total: targetCPU, |
| 205 | + }, |
| 206 | + }, |
| 207 | + }, |
| 208 | + } |
| 209 | + |
| 210 | + // In production the mutating webhook Default() always runs before creation. |
| 211 | + // This envtest environment does not register the webhook, so call it |
| 212 | + // explicitly here to reproduce the same state as production. |
| 213 | + if applyWebhookDefault { |
| 214 | + d := &webhookv1beta1.SpannerAutoscalerCustomDefaulter{} |
| 215 | + if err := d.Default(ctx, sa); err != nil { |
| 216 | + t.Fatalf("webhook Default() failed: %v", err) |
| 217 | + } |
| 218 | + } |
| 219 | + effectiveStepSize = sa.Spec.ScaleConfig.ScaledownStepSize.String() |
| 220 | + |
| 221 | + if err := k8sClient.Create(ctx, sa); err != nil { |
| 222 | + t.Fatalf("failed to create SpannerAutoscaler: %v", err) |
| 223 | + } |
| 224 | + |
| 225 | + // Observe the PU sequence. status.CurrentProcessingUnits reflects the |
| 226 | + // instance's actual PU in the emulator (populated by the syncer); record it |
| 227 | + // whenever the value changes. |
| 228 | + seq = []int{} |
| 229 | + last := -1 |
| 230 | + deadline := time.Now().Add(25 * time.Second) |
| 231 | + for time.Now().Before(deadline) { |
| 232 | + var updated spannerv1beta1.SpannerAutoscaler |
| 233 | + if err := k8sClient.Get(ctx, nn, &updated); err != nil { |
| 234 | + time.Sleep(200 * time.Millisecond) |
| 235 | + continue |
| 236 | + } |
| 237 | + cur := updated.Status.CurrentProcessingUnits |
| 238 | + if cur != 0 && cur != last { |
| 239 | + seq = append(seq, cur) |
| 240 | + last = cur |
| 241 | + t.Logf("[%s] observed PU=%d (totalCPU=%d%%)", saName, cur, updated.Status.CurrentTotalCPUUtilization) |
| 242 | + } |
| 243 | + time.Sleep(200 * time.Millisecond) |
| 244 | + } |
| 245 | + return seq, effectiveStepSize |
| 246 | +} |
0 commit comments