Reduce test wall-clock time without weakening coverage

.github/workflows/test.yml

@@ -121,6 +121,7 @@ jobs:
           # cross-filesystem safe. Keeping both on /ci avoids that.
           export HYPEMAN_TEST_PREWARM_DIR="/ci/prewarm/linux-amd64"
           mkdir -p "$HYPEMAN_TEST_PREWARM_DIR"
+          sudo chown -R "$(id -u):$(id -g)" "$HYPEMAN_TEST_PREWARM_DIR"
           go run ./cmd/test-prewarm
 
       - name: Check gofmt

cmd/test-prewarm/main.go

@@ -106,17 +106,36 @@ func main() {
 		PrewarmDir: prewarmDir,
 		Images:     make([]manifestImage, 0, len(imagesToWarm)),
 	}
+	p := paths.New(prewarmDir)
+	readyImageManager, err := images.NewManager(p, 1, nil)
+	if err != nil {
+		fatalf("create ready image manager: %v", err)
+	}
 
 	for _, img := range imagesToWarm {
 		entry, err := ensureMirroredImage(ctx, inspectClient, registry, img)
 		if err != nil {
 			fatalf("prewarm image %s: %v", img.Source, err)
 		}
 		fmt.Printf("prewarm image source=%s local=%s digest=%s cache_hit=%t\n", entry.Source, entry.LocalRef, entry.Digest, entry.CacheHit)
+		created, err := readyImageManager.CreateImage(ctx, images.CreateImageRequest{Name: entry.LocalRef, Platform: img.Platform})
+		if err != nil {
+			fatalf("build ready image %s: %v", entry.LocalRef, err)
+		}
+		waitName := created.Name
+		if created.Digest != "" {
+			ref, parseErr := images.ParseNormalizedRef(entry.LocalRef)
+			if parseErr != nil {
+				fatalf("parse ready image ref %s: %v", entry.LocalRef, parseErr)
+			}
+			waitName = ref.Repository() + "@" + created.Digest
+		}
+		if err := readyImageManager.WaitForReady(ctx, waitName); err != nil {
+			fatalf("wait for ready image %s: %v", entry.LocalRef, err)
+		}
 		manifest.Images = append(manifest.Images, entry)
 	}
 
-	p := paths.New(prewarmDir)
 	systemMgr := system.NewManager(p)
 	if err := systemMgr.EnsureSystemFiles(ctx); err != nil {
 		fatalf("prewarm system files: %v", err)

lib/instances/auto_standby_integration_linux_test.go

@@ -99,7 +99,7 @@ func TestAutoStandbyCloudHypervisorActiveInboundTCP(t *testing.T) {
 	mgr, _ := setupCompressionTestManagerForHypervisor(t, hypervisor.TypeCloudHypervisor)
 	require.NoError(t, mgr.networkManager.Initialize(ctx, nil))
 	require.NoError(t, mgr.systemManager.EnsureSystemFiles(ctx))
-	createNginxImageAndWait(t, ctx, mgr.imageManager)
+	createNginxImageAndWait(t, ctx, mgr.paths, mgr.imageManager)
 
 	connSource := autostandby.NewConntrackSource()
 	if _, err := connSource.ListConnections(ctx); err != nil {
@@ -128,7 +128,7 @@ func TestAutoStandbyCloudHypervisorActiveInboundTCP(t *testing.T) {
 
 	t.Cleanup(func() {
 		logInstanceArtifactsOnFailure(t, mgr, instanceID)
-		_ = mgr.DeleteInstance(context.Background(), instanceID)
+		_ = deleteTestInstanceNow(context.Background(), mgr, instanceID)
 	})
 
 	inst, err = waitForInstanceState(ctx, mgr, instanceID, StateRunning, 30*time.Second)

lib/instances/compression_integration_linux_test.go

@@ -137,15 +137,15 @@ func runStandbyRestoreCompressionScenarios(t *testing.T, harness compressionInte
 
 	imageManager, err := images.NewManager(p, 1, nil)
 	require.NoError(t, err)
-	createNginxImageAndWait(t, ctx, imageManager)
+	createNginxImageAndWait(t, ctx, p, imageManager)
 
 	systemManager := system.NewManager(p)
 	require.NoError(t, systemManager.EnsureSystemFiles(ctx))
 
 	inst, err := mgr.CreateInstance(ctx, CreateInstanceRequest{
 		Name:           fmt.Sprintf("compression-%s", harness.name),
 		Image:          integrationTestImageRef(t, "docker.io/library/nginx:alpine"),
-		Size:           1024 * 1024 * 1024,
+		Size:           lifecycleTestMemorySize,
 		HotplugSize:    512 * 1024 * 1024,
 		OverlaySize:    10 * 1024 * 1024 * 1024,
 		Vcpus:          1,
@@ -157,7 +157,7 @@ func runStandbyRestoreCompressionScenarios(t *testing.T, harness compressionInte
 	deleted := false
 	t.Cleanup(func() {
 		if !deleted {
-			_ = mgr.DeleteInstance(context.Background(), inst.Id)
+			_ = deleteTestInstanceNow(context.Background(), mgr, inst.Id)
 		}
 	})
 
@@ -196,7 +196,7 @@ func runStandbyRestoreCompressionScenarios(t *testing.T, harness compressionInte
 		inst = runCompressionCycle(t, ctx, mgr, p, inst, harness.waitHypervisorUp, tc.name, tc.cfg, true)
 	}
 
-	require.NoError(t, mgr.DeleteInstance(ctx, inst.Id))
+	require.NoError(t, deleteTestInstanceNow(ctx, mgr, inst.Id))
 	deleted = true
 }
 

lib/instances/delete.go

@@ -17,10 +17,22 @@ import (
 
 const deleteGracefulShutdownTimeout = 2
 
+type deleteInstanceOptions struct {
+	skipGracefulShutdown bool
+}
+
 // deleteInstance stops and deletes an instance
 func (m *manager) deleteInstance(
 	ctx context.Context,
 	id string,
+) error {
+	return m.deleteInstanceWithOptions(ctx, id, deleteInstanceOptions{})
+}
+
+func (m *manager) deleteInstanceWithOptions(
+	ctx context.Context,
+	id string,
+	options deleteInstanceOptions,
 ) (retErr error) {
 	log := logger.FromContext(ctx)
 	log.InfoContext(ctx, "deleting instance", "instance_id", id)
@@ -76,7 +88,7 @@ func (m *manager) deleteInstance(
 
 	// 4. If active, try graceful guest shutdown before force kill.
 	gracefulShutdown := false
-	if inst.State == StateRunning || inst.State == StateInitializing {
+	if !options.skipGracefulShutdown && (inst.State == StateRunning || inst.State == StateInitializing) {
 		stopTimeout := resolveStopTimeout(stored)
 		if stopTimeout > deleteGracefulShutdownTimeout {
 			stopTimeout = deleteGracefulShutdownTimeout

lib/instances/docker_integration_linux_test.go

@@ -53,7 +53,7 @@ func TestDockerInVMCloudHypervisorWithAttachedVolume(t *testing.T) {
 					t.Logf("dockerd log (exit=%d):\n%s", code, output)
 				}
 			}
-			_ = manager.DeleteInstance(context.Background(), inst.Id)
+			_ = deleteTestInstanceNow(context.Background(), manager, inst.Id)
 		}
 		_ = volumeManager.DeleteVolume(context.Background(), vol.Id)
 	})

lib/instances/egress_proxy_integration_test.go

@@ -18,7 +18,7 @@ import (
 	"time"
 
 	"github.com/kernel/hypeman/lib/egressproxy"
-	"github.com/kernel/hypeman/lib/images"
+	snapshottest "github.com/kernel/hypeman/lib/snapshot/testsupport"
 	"github.com/stretchr/testify/require"
 )
 
@@ -49,20 +49,7 @@ func TestEgressProxyRewritesHTTPSHeaders(t *testing.T) {
 	require.NoError(t, err)
 
 	imageRef := integrationTestImageRef(t, "docker.io/library/nginx:alpine")
-	t.Logf("Pulling %s image...", imageRef)
-	created, err := manager.imageManager.CreateImage(ctx, images.CreateImageRequest{Name: imageRef})
-	require.NoError(t, err)
-
-	for i := 0; i < 120; i++ {
-		img, err := manager.imageManager.GetImage(ctx, created.Name)
-		if err == nil && img.Status == images.StatusReady {
-			break
-		}
-		time.Sleep(1 * time.Second)
-	}
-	img, err := manager.imageManager.GetImage(ctx, created.Name)
-	require.NoError(t, err)
-	require.Equal(t, images.StatusReady, img.Status)
+	snapshottest.EnsureImageReady(t, ctx, manager.paths, manager.imageManager, imageRef)
 
 	require.NoError(t, manager.systemManager.EnsureSystemFiles(ctx))
 	require.NoError(t, manager.networkManager.Initialize(ctx, nil))
@@ -103,7 +90,7 @@ func TestEgressProxyRewritesHTTPSHeaders(t *testing.T) {
 	deleted := false
 	t.Cleanup(func() {
 		if !deleted {
-			_ = manager.DeleteInstance(context.Background(), inst.Id)
+			_ = deleteTestInstanceNow(context.Background(), manager, inst.Id)
 		}
 	})
 
@@ -192,7 +179,7 @@ func TestEgressProxyRewritesHTTPSHeaders(t *testing.T) {
 	require.Contains(t, rotatedOutput, "Bearer rotated-key-456")
 	require.NotContains(t, rotatedOutput, "real-openai-key-123")
 
-	require.NoError(t, manager.DeleteInstance(ctx, inst.Id))
+	require.NoError(t, deleteTestInstanceNow(ctx, manager, inst.Id))
 	deleted = true
 }