From 65d658a05948c6c6d8eb9189db635f44777b7bed Mon Sep 17 00:00:00 2001 From: Wenkai Yin Date: Sun, 23 Aug 2020 21:31:17 +0800 Subject: [PATCH] retry several times when failed to copy blob during the replication Signed-off-by: Wenkai Yin --- src/replication/transfer/image/transfer.go | 34 +++++++++++++++++++--- 1 file changed, 30 insertions(+), 4 deletions(-) diff --git a/src/replication/transfer/image/transfer.go b/src/replication/transfer/image/transfer.go index 57ccf86ac..f8acfa7c9 100644 --- a/src/replication/transfer/image/transfer.go +++ b/src/replication/transfer/image/transfer.go @@ -17,7 +17,10 @@ package image import ( "errors" "net/http" + "os" + "strconv" "strings" + "time" "github.com/docker/distribution/manifest/manifestlist" "github.com/docker/distribution/manifest/schema1" @@ -32,7 +35,15 @@ import ( trans "github.com/goharbor/harbor/src/replication/transfer" ) +var ( + retry int +) + func init() { + retry, _ = strconv.Atoi(os.Getenv("COPY_BLOB_RETRY_COUNT")) + if retry <= 0 { + retry = 5 + } if err := trans.RegisterFactory(model.ResourceTypeImage, factory); err != nil { log.Errorf("failed to register transfer factory: %v", err) } @@ -244,17 +255,33 @@ func (t *transfer) copyContent(content distribution.Descriptor, srcRepo, dstRepo // the media type of the layer or config can be "application/octet-stream", // schema1.MediaTypeManifestLayer, schema2.MediaTypeLayer, schema2.MediaTypeImageConfig default: - return t.copyBlob(srcRepo, dstRepo, digest, content.Size) + return t.copyBlobWithRetry(srcRepo, dstRepo, digest, content.Size) } } +func (t *transfer) copyBlobWithRetry(srcRepo, dstRepo, digest string, sizeFromDescriptor int64) error { + var err error + for i, backoff := 1, 2*time.Second; i <= retry; i, backoff = i+1, backoff*2 { + t.logger.Infof("copying the blob %s(the %dth running)...", digest, i) + if err = t.copyBlob(srcRepo, dstRepo, digest, sizeFromDescriptor); err == nil { + t.logger.Infof("copy the blob %s completed", digest) + return nil + } + if i == retry { + break + } + t.logger.Infof("will retry %v later", backoff) + time.Sleep(backoff) + } + return err +} + // copy the layer or artifact config from the source registry to destination // the size parameter is taken from manifests. func (t *transfer) copyBlob(srcRepo, dstRepo, digest string, sizeFromDescriptor int64) error { if t.shouldStop() { return nil } - t.logger.Infof("copying the blob %s...", digest) exist, err := t.dst.BlobExist(dstRepo, digest) if err != nil { t.logger.Errorf("failed to check the existence of blob %s on the destination registry: %v", digest, err) @@ -278,10 +305,9 @@ func (t *transfer) copyBlob(srcRepo, dstRepo, digest string, sizeFromDescriptor } if err = t.dst.PushBlob(dstRepo, digest, size, data); err != nil { - t.logger.Errorf("failed to pushing the blob %s: %v", digest, err) + t.logger.Errorf("failed to pushing the blob %s, size %d: %v", digest, size, err) return err } - t.logger.Infof("copy the blob %s completed", digest) return nil }