refactor: increase the retry timeout and wrap the error for retry (#221)

Signed-off-by: chlins <[email protected]>

Author: chlins

internal/pb/pb.go

@@ -24,6 +24,7 @@ import (
 	"time"
 
 	humanize "github.com/dustin/go-humanize"
+	"github.com/sirupsen/logrus"
 	mpbv8 "github.com/vbauerster/mpb/v8"
 	"github.com/vbauerster/mpb/v8/decor"
 )
@@ -154,7 +155,7 @@ func (p *ProgressBar) Abort(name string, err error) {
 	p.mu.RUnlock()
 
 	if ok {
-		// TODO: Log error message.
+		logrus.Errorf("abort the progress bar[%s] as error occurred: %v", name, err)
 		bar.Abort(true)
 	}
 }

pkg/backend/processor/base.go

@@ -145,13 +145,14 @@ func (b *base) Process(ctx context.Context, builder build.Builder, workDir strin
 					}),
 				))
 				if err != nil {
-					logrus.Errorf("processor: failed to build layer for %s file %s: %v", b.name, path, err)
+					err = fmt.Errorf("processor: failed to build layer for %s file %s: %w", b.name, path, err)
+					logrus.Error(err)
 					cancel()
 					return err
 				}
 
-				mu.Lock()
 				logrus.Debugf("processor: successfully built %s layer for file %s [digest: %s, size: %d]", b.name, path, desc.Digest, desc.Size)
+				mu.Lock()
 				descriptors = append(descriptors, desc)
 				mu.Unlock()
 

pkg/backend/processor/options.go

@@ -46,8 +46,8 @@ func WithProgressTracker(tracker *pb.ProgressBar) ProcessOption {
 }
 
 var defaultRetryOpts = []retry.Option{
-	retry.Attempts(3),
+	retry.Attempts(4),
 	retry.DelayType(retry.BackOffDelay),
-	retry.Delay(5 * time.Second),
-	retry.MaxDelay(10 * time.Second),
+	retry.Delay(10 * time.Second),
+	retry.MaxDelay(20 * time.Second),
 }

pkg/backend/pull.go

@@ -119,10 +119,10 @@ func (b *backend) Pull(ctx context.Context, target string, cfg *config.Pull) err
 				// call the after hook.
 				cfg.Hooks.AfterPullLayer(layer, err)
 				if err != nil {
-					logrus.Debugf("pull: failed to process layer %s: %v", layer.Digest, err)
-				} else {
-					logrus.Debugf("pull: successfully processed layer %s", layer.Digest)
+					err = fmt.Errorf("pull: failed to process layer %s: %w", layer.Digest, err)
+					logrus.Error(err)
 				}
+
 				return err
 			}, append(defaultRetryOpts, retry.Context(ctx))...)
 		})

pkg/backend/pull_by_d7y.go

@@ -165,11 +165,19 @@ func buildBlobURL(ref Referencer, plainHTTP bool, digest string) string {
 
 // processLayer handles downloading and extracting a single layer.
 func processLayer(ctx context.Context, pb *internalpb.ProgressBar, client dfdaemon.DfdaemonDownloadClient, ref Referencer, manifest ocispec.Manifest, desc ocispec.Descriptor, authToken string, cfg *config.Pull) error {
-	cfg.Hooks.BeforePullLayer(desc, manifest) // Call before hook
 	err := retry.Do(func() error {
-		return downloadAndExtractLayer(ctx, pb, client, ref, desc, authToken, cfg)
+		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)
+		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)
+			logrus.Error(err)
+		}
+
+		return err
 	}, append(defaultRetryOpts, retry.Context(ctx))...)
-	cfg.Hooks.AfterPullLayer(desc, err) // Call after hook
+
 	return err
 }
 

pkg/backend/retry.go

@@ -23,8 +23,8 @@ import (
 )
 
 var defaultRetryOpts = []retry.Option{
-	retry.Attempts(3),
+	retry.Attempts(4),
 	retry.DelayType(retry.BackOffDelay),
-	retry.Delay(5 * time.Second),
-	retry.MaxDelay(10 * time.Second),
+	retry.Delay(10 * time.Second),
+	retry.MaxDelay(20 * time.Second),
 }