fix: handle DeadlineExceeded in IsRetryable, fix ctx shadow, speed up tests

Signed-off-by: Zhao Chen <winters.zc@antgroup.com>

Author: aftersnow

pkg/backend/fetch.go

@@ -130,8 +130,8 @@ func (b *backend) Fetch(ctx context.Context, target string, cfg *config.Fetch) e
 			}
 
 			logrus.Debugf("fetch: processing layer %s", layer.Digest)
-			if err := retrypolicy.Do(ctx, func(ctx context.Context) error {
-				return pullAndExtractFromRemote(ctx, pb, internalpb.NormalizePrompt("Fetching blob"), client, cfg.Output, layer)
+			if err := retrypolicy.Do(ctx, func(rctx context.Context) error {
+				return pullAndExtractFromRemote(rctx, pb, internalpb.NormalizePrompt("Fetching blob"), client, cfg.Output, layer)
 			}, retrypolicy.DoOpts{
 				FileSize: layer.Size,
 				FileName: annoFilepath,

pkg/backend/fetch_by_d7y.go

@@ -177,10 +177,10 @@ func fetchLayerByDragonfly(ctx context.Context, pb *internalpb.ProgressBar, clie
 		}
 	}
 
-	err := retrypolicy.Do(ctx, func(ctx context.Context) error {
+	err := retrypolicy.Do(ctx, func(rctx context.Context) error {
 		logrus.Debugf("fetch: processing layer %s", desc.Digest)
 		cfg.Hooks.BeforePullLayer(desc, manifest) // Call before hook
-		err := downloadAndExtractFetchLayer(ctx, pb, client, ref, desc, authToken, cfg)
+		err := downloadAndExtractFetchLayer(rctx, pb, client, ref, desc, authToken, cfg)
 		cfg.Hooks.AfterPullLayer(desc, err) // Call after hook
 		if err != nil {
 			err = fmt.Errorf("pull: failed to download and extract layer %s: %w", desc.Digest, err)

pkg/backend/pull_by_d7y.go

@@ -201,10 +201,10 @@ func processLayer(ctx context.Context, pb *internalpb.ProgressBar, client dfdaem
 		}
 	}
 
-	err := retrypolicy.Do(ctx, func(ctx context.Context) error {
+	err := retrypolicy.Do(ctx, func(rctx context.Context) error {
 		logrus.Debugf("pull: processing layer %s", desc.Digest)
 		cfg.Hooks.BeforePullLayer(desc, manifest) // Call before hook
-		err := downloadAndExtractLayer(ctx, pb, client, ref, desc, authToken, cfg)
+		err := downloadAndExtractLayer(rctx, pb, client, ref, desc, authToken, cfg)
 		cfg.Hooks.AfterPullLayer(desc, err) // Call after hook
 		if err != nil {
 			err = fmt.Errorf("pull: failed to download and extract layer %s: %w", desc.Digest, err)

pkg/retrypolicy/retrypolicy.go

@@ -51,6 +51,8 @@ const (
 type Config struct {
 	MaxRetryTime time.Duration // 0 = dynamic based on file size
 	NoRetry      bool          // disable retry entirely
+	InitialDelay time.Duration // 0 = use default (5s), for testing
+	MaxJitter    time.Duration // -1 = no jitter, 0 = use default (5s), for testing
 }
 
 // DoOpts configures a single Do call.
@@ -93,16 +95,27 @@ func Do(ctx context.Context, fn func(ctx context.Context) error, opts DoOpts) er
 
 	sizeStr := humanizeBytes(opts.FileSize)
 
+	delay := initialDelay
+	jitter := maxJitter
+	if cfg.InitialDelay > 0 {
+		delay = cfg.InitialDelay
+	}
+	if cfg.MaxJitter < 0 {
+		jitter = 0
+	} else if cfg.MaxJitter > 0 {
+		jitter = cfg.MaxJitter
+	}
+
 	return retry.Do(
 		func() error {
 			return fn(deadlineCtx)
 		},
 		retry.Attempts(0),
 		retry.Context(deadlineCtx),
 		retry.DelayType(retry.BackOffDelay),
-		retry.Delay(initialDelay),
+		retry.Delay(delay),
 		retry.MaxDelay(maxBackoff),
-		retry.MaxJitter(maxJitter),
+		retry.MaxJitter(jitter),
 		retry.LastErrorOnly(true),
 		retry.WrapContextErrorWithLastError(true),
 		retry.RetryIf(func(err error) bool {
@@ -180,8 +193,8 @@ func IsRetryable(err error) bool {
 		return false
 	}
 
-	// context.Canceled is never retryable — it means user/system cancellation.
-	if errors.Is(err, context.Canceled) {
+	// context.Canceled and context.DeadlineExceeded are never retryable.
+	if errors.Is(err, context.Canceled) || errors.Is(err, context.DeadlineExceeded) {
 		return false
 	}
 

pkg/retrypolicy/retrypolicy_test.go

@@ -132,6 +132,16 @@ func TestIsRetryable(t *testing.T) {
 			err:  fmt.Errorf("operation failed: %w", context.Canceled),
 			want: false,
 		},
+		{
+			name: "context.DeadlineExceeded",
+			err:  context.DeadlineExceeded,
+			want: false,
+		},
+		{
+			name: "wrapped context.DeadlineExceeded",
+			err:  fmt.Errorf("operation timed out: %w", context.DeadlineExceeded),
+			want: false,
+		},
 		{
 			name: "HTTP 500 server error",
 			err:  fmt.Errorf("PUT /blobs/uploads: response status code 500: internal server error"),
@@ -344,6 +354,7 @@ func TestDo_RetryOnTransientError(t *testing.T) {
 	}, DoOpts{
 		FileSize: 100,
 		FileName: "test.bin",
+		Config:   &Config{InitialDelay: 10 * time.Millisecond, MaxJitter: -1},
 	})
 
 	if err != nil {
@@ -406,6 +417,7 @@ func TestDo_ParentContextCancel(t *testing.T) {
 	}, DoOpts{
 		FileSize: 100,
 		FileName: "test.bin",
+		Config:   &Config{InitialDelay: 10 * time.Millisecond, MaxJitter: -1},
 	})
 
 	if err == nil {
@@ -427,6 +439,7 @@ func TestDo_OnRetryCallback(t *testing.T) {
 	}, DoOpts{
 		FileSize: 100,
 		FileName: "test.bin",
+		Config:   &Config{InitialDelay: 10 * time.Millisecond, MaxJitter: -1},
 		OnRetry: func(attempt uint, reason string, backoff time.Duration) {
 			retryAttempts = append(retryAttempts, attempt)
 			retryReasons = append(retryReasons, reason)
@@ -458,7 +471,9 @@ func TestDo_ConfigMaxRetryTimeOverride(t *testing.T) {
 		FileSize: 100,
 		FileName: "test.bin",
 		Config: &Config{
-			MaxRetryTime: 8 * time.Second,
+			MaxRetryTime: 1 * time.Second,
+			InitialDelay: 50 * time.Millisecond,
+			MaxJitter:    -1,
 		},
 	})
 
@@ -467,9 +482,9 @@ func TestDo_ConfigMaxRetryTimeOverride(t *testing.T) {
 	if err == nil {
 		t.Fatal("expected error after retry timeout, got nil")
 	}
-	// Should have run for approximately MaxRetryTime (8s), not the dynamic default (10min).
-	if elapsed > 30*time.Second {
-		t.Errorf("expected retry to terminate within ~8s, but elapsed %v", elapsed)
+	// Should have run for approximately MaxRetryTime (1s), not the dynamic default (10min).
+	if elapsed > 5*time.Second {
+		t.Errorf("expected retry to terminate within ~1s, but elapsed %v", elapsed)
 	}
 	if atomic.LoadInt32(&callCount) < 2 {
 		t.Errorf("expected at least 2 attempts, got %d", callCount)