From 682151b166d55892de9adc99e0c66f6b2ef09391 Mon Sep 17 00:00:00 2001 From: Stephen J Day Date: Wed, 1 Nov 2017 16:11:40 -0700 Subject: [PATCH] remotes/docker: implement seekable http requests To support resumable download, the fetcher for a remote must implement `io.Seeker`. If implemented the `content.Copy` function will detect the seeker and begin from where the download was terminated by a previous attempt. Signed-off-by: Stephen J Day --- content/helpers.go | 3 +- remotes/docker/fetcher.go | 69 ++++++++++++----- remotes/docker/httpreadseeker.go | 125 +++++++++++++++++++++++++++++++ remotes/docker/resolver.go | 2 +- 4 files changed, 176 insertions(+), 23 deletions(-) create mode 100644 remotes/docker/httpreadseeker.go diff --git a/content/helpers.go b/content/helpers.go index 32efc6ca0..775583a90 100644 --- a/content/helpers.go +++ b/content/helpers.go @@ -2,7 +2,6 @@ package content import ( "context" - "fmt" "io" "sync" @@ -119,7 +118,7 @@ func seekReader(r io.Reader, offset, size int64) (io.Reader, error) { if ok { nn, err := seeker.Seek(offset, io.SeekStart) if nn != offset { - return nil, fmt.Errorf("failed to seek to offset %v", offset) + return nil, errors.Wrapf(err, "failed to seek to offset %v", offset) } if err != nil { diff --git a/remotes/docker/fetcher.go b/remotes/docker/fetcher.go index 46677e491..222cf83c0 100644 --- a/remotes/docker/fetcher.go +++ b/remotes/docker/fetcher.go @@ -2,6 +2,7 @@ package docker import ( "context" + "fmt" "io" "net/http" "path" @@ -37,32 +38,60 @@ func (r dockerFetcher) Fetch(ctx context.Context, desc ocispec.Descriptor) (io.R return nil, err } - for _, u := range urls { - req, err := http.NewRequest(http.MethodGet, u, nil) - if err != nil { - return nil, err - } + return newHTTPReadSeeker(desc.Size, func(offset int64) (io.ReadCloser, error) { + for _, u := range urls { + rc, err := r.open(ctx, u, desc.MediaType, offset) + if err != nil { + if errdefs.IsNotFound(err) { + continue // try one of the other urls. + } - req.Header.Set("Accept", strings.Join([]string{desc.MediaType, `*`}, ", ")) - resp, err := r.doRequestWithRetries(ctx, req, nil) - if err != nil { - return nil, err - } - - if resp.StatusCode > 299 { - resp.Body.Close() - if resp.StatusCode == http.StatusNotFound { - continue // try one of the other urls. + return nil, err } - return nil, errors.Errorf("unexpected status code %v: %v", u, resp.Status) + + return rc, nil } - return resp.Body, nil + return nil, errors.Wrapf(errdefs.ErrNotFound, + "could not fetch content descriptor %v (%v) from remote", + desc.Digest, desc.MediaType) + + }) +} + +func (r dockerFetcher) open(ctx context.Context, u, mediatype string, offset int64) (io.ReadCloser, error) { + req, err := http.NewRequest(http.MethodGet, u, nil) + if err != nil { + return nil, err } - return nil, errors.Wrapf(errdefs.ErrNotFound, - "could not fetch content descriptor %v (%v) from remote", - desc.Digest, desc.MediaType) + req.Header.Set("Accept", strings.Join([]string{mediatype, `*`}, ", ")) + + if offset > 0 { + // TODO(stevvooe): Only set this header in response to the + // "Accept-Ranges: bytes" header. + req.Header.Set("Range", fmt.Sprintf("bytes=%d-", offset)) + } + + resp, err := r.doRequestWithRetries(ctx, req, nil) + if err != nil { + return nil, err + } + + if resp.StatusCode > 299 { + // TODO(stevvooe): When doing a offset specific request, we should + // really distinguish between a 206 and a 200. In the case of 200, we + // can discard the bytes, hiding the seek behavior from the + // implementation. + + resp.Body.Close() + if resp.StatusCode == http.StatusNotFound { + return nil, errors.Wrapf(errdefs.ErrNotFound, "content at %v not found", u) + } + return nil, errors.Errorf("unexpected status code %v: %v", u, resp.Status) + } + + return resp.Body, nil } // getV2URLPaths generates the candidate urls paths for the object based on the diff --git a/remotes/docker/httpreadseeker.go b/remotes/docker/httpreadseeker.go new file mode 100644 index 000000000..b042a8852 --- /dev/null +++ b/remotes/docker/httpreadseeker.go @@ -0,0 +1,125 @@ +package docker + +import ( + "bytes" + "io" + "io/ioutil" + + "github.com/containerd/containerd/errdefs" + "github.com/containerd/containerd/log" + "github.com/pkg/errors" +) + +type httpReadSeeker struct { + size int64 + offset int64 + rc io.ReadCloser + open func(offset int64) (io.ReadCloser, error) + closed bool +} + +func newHTTPReadSeeker(size int64, open func(offset int64) (io.ReadCloser, error)) (io.ReadCloser, error) { + return &httpReadSeeker{ + size: size, + open: open, + }, nil +} + +func (hrs *httpReadSeeker) Read(p []byte) (n int, err error) { + if hrs.closed { + return 0, io.EOF + } + + rd, err := hrs.reader() + if err != nil { + return 0, err + } + + n, err = rd.Read(p) + hrs.offset += int64(n) + return +} + +func (hrs *httpReadSeeker) Close() error { + if hrs.closed { + return nil + } + hrs.closed = true + if hrs.rc != nil { + return hrs.rc.Close() + } + + return nil +} + +func (hrs *httpReadSeeker) Seek(offset int64, whence int) (int64, error) { + if hrs.closed { + return 0, errors.Wrap(errdefs.ErrUnavailable, "Fetcher.Seek: closed") + } + + abs := hrs.offset + switch whence { + case io.SeekStart: + abs = offset + case io.SeekCurrent: + abs += offset + case io.SeekEnd: + abs = hrs.size + offset + default: + return 0, errors.Wrap(errdefs.ErrInvalidArgument, "Fetcher.Seek: invalid whence") + } + + if abs < 0 { + return 0, errors.Wrapf(errdefs.ErrInvalidArgument, "Fetcher.Seek: negative offset") + } + + if abs != hrs.offset { + if hrs.rc != nil { + if err := hrs.rc.Close(); err != nil { + log.L.WithError(err).Errorf("Fetcher.Seek: failed to close ReadCloser") + } + + hrs.rc = nil + } + + hrs.offset = abs + } + + return hrs.offset, nil +} + +func (hrs *httpReadSeeker) reader() (io.Reader, error) { + if hrs.rc != nil { + return hrs.rc, nil + } + + if hrs.offset < hrs.size { + // only try to reopen the body request if we are seeking to a value + // less than the actual size. + if hrs.open == nil { + return nil, errors.Wrapf(errdefs.ErrNotImplemented, "cannot open") + } + + rc, err := hrs.open(hrs.offset) + if err != nil { + return nil, errors.Wrapf(err, "httpReaderSeeker: failed open") + } + + if hrs.rc != nil { + if err := hrs.rc.Close(); err != nil { + log.L.WithError(err).Errorf("httpReadSeeker: failed to close ReadCloser") + } + } + hrs.rc = rc + } else { + // There is an edge case here where offset == size of the content. If + // we seek, we will probably get an error for content that cannot be + // sought (?). In that case, we should err on committing the content, + // as the length is already satisified but we just return the empty + // reader instead. + + hrs.rc = ioutil.NopCloser(bytes.NewReader([]byte{})) + } + + return hrs.rc, nil +} diff --git a/remotes/docker/resolver.go b/remotes/docker/resolver.go index 7a1150495..a23e16e82 100644 --- a/remotes/docker/resolver.go +++ b/remotes/docker/resolver.go @@ -298,7 +298,7 @@ func (r *dockerBase) authorize(req *http.Request) { func (r *dockerBase) doRequest(ctx context.Context, req *http.Request) (*http.Response, error) { ctx = log.WithLogger(ctx, log.G(ctx).WithField("url", req.URL.String())) - log.G(ctx).WithField("request.headers", req.Header).WithField("request.method", req.Method).Debug("Do request") + log.G(ctx).WithField("request.headers", req.Header).WithField("request.method", req.Method).Debug("do request") r.authorize(req) resp, err := ctxhttp.Do(ctx, r.client, req) if err != nil {