b2: Make NewObject use less expensive API calls

Before this change when NewObject was called the b2 backend would list
the directory that the object was in in order to find it.

Unfortunately list calls are Class C transactions and cost more.

This patch switches to using HEAD requests instead. These are Class B
transactions. It is then necessary to parse the headers from response
back into the data that we get from the listing. However B2 returns
exactly the same data, just in a different form.

Rclone will use the old directory listing method when looking for
files with versions as these can't be found via a HEAD request.

This change will particularly benefit --files-from, rclone serve
restic but most operations will see some benefit.
This commit is contained in:
Nick Craig-Wood 2020-12-03 17:24:46 +00:00
parent 7ae84a3c91
commit cb16f42075

View File

@ -44,8 +44,10 @@ const (
timeHeader = headerPrefix + timeKey
sha1Key = "large_file_sha1"
sha1Header = "X-Bz-Content-Sha1"
sha1InfoHeader = headerPrefix + sha1Key
testModeHeader = "X-Bz-Test-Mode"
idHeader = "X-Bz-File-Id"
nameHeader = "X-Bz-File-Name"
timestampHeader = "X-Bz-Upload-Timestamp"
retryAfterHeader = "Retry-After"
minSleep = 10 * time.Millisecond
maxSleep = 5 * time.Minute
@ -1496,8 +1498,11 @@ func (o *Object) decodeMetaDataFileInfo(info *api.FileInfo) (err error) {
return o.decodeMetaDataRaw(info.ID, info.SHA1, info.Size, info.UploadTimestamp, info.Info, info.ContentType)
}
// getMetaData gets the metadata from the object unconditionally
func (o *Object) getMetaData(ctx context.Context) (info *api.File, err error) {
// getMetaDataListing gets the metadata from the object unconditionally from the listing
//
// Note that listing is a class C transaction which costs more than
// the B transaction used in getMetaData
func (o *Object) getMetaDataListing(ctx context.Context) (info *api.File, err error) {
bucket, bucketPath := o.split()
maxSearched := 1
var timestamp api.Timestamp
@ -1530,6 +1535,19 @@ func (o *Object) getMetaData(ctx context.Context) (info *api.File, err error) {
return info, nil
}
// getMetaData gets the metadata from the object unconditionally
func (o *Object) getMetaData(ctx context.Context) (info *api.File, err error) {
// If using versions and have a version suffix, need to list the directory to find the correct versions
if o.fs.opt.Versions {
timestamp, _ := api.RemoveVersion(o.remote)
if !timestamp.IsZero() {
return o.getMetaDataListing(ctx)
}
}
_, info, err = o.getOrHead(ctx, "HEAD", nil)
return info, err
}
// readMetaData gets the metadata if it hasn't already been fetched
//
// Sets
@ -1659,12 +1677,11 @@ func (file *openFile) Close() (err error) {
// Check it satisfies the interfaces
var _ io.ReadCloser = &openFile{}
// Open an object for read
func (o *Object) Open(ctx context.Context, options ...fs.OpenOption) (in io.ReadCloser, err error) {
fs.FixRangeOption(options, o.size)
func (o *Object) getOrHead(ctx context.Context, method string, options []fs.OpenOption) (resp *http.Response, info *api.File, err error) {
opts := rest.Opts{
Method: "GET",
Options: options,
Method: method,
Options: options,
NoResponse: method == "HEAD",
}
// Use downloadUrl from backblaze if downloadUrl is not set
@ -1682,37 +1699,67 @@ func (o *Object) Open(ctx context.Context, options ...fs.OpenOption) (in io.Read
bucket, bucketPath := o.split()
opts.Path += "/file/" + urlEncode(o.fs.opt.Enc.FromStandardName(bucket)) + "/" + urlEncode(o.fs.opt.Enc.FromStandardPath(bucketPath))
}
var resp *http.Response
err = o.fs.pacer.Call(func() (bool, error) {
resp, err = o.fs.srv.Call(ctx, &opts)
return o.fs.shouldRetry(ctx, resp, err)
})
if err != nil {
return nil, errors.Wrap(err, "failed to open for download")
// 404 for files, 400 for directories
if resp != nil && (resp.StatusCode == http.StatusNotFound || resp.StatusCode == http.StatusBadRequest) {
return nil, nil, fs.ErrorObjectNotFound
}
return nil, nil, errors.Wrapf(err, "failed to %s for download", method)
}
// Parse the time out of the headers if possible
err = o.parseTimeString(resp.Header.Get(timeHeader))
// NB resp may be Open here - don't return err != nil without closing
// Convert the Headers into an api.File
var uploadTimestamp api.Timestamp
err = uploadTimestamp.UnmarshalJSON([]byte(resp.Header.Get(timestampHeader)))
if err != nil {
fs.Debugf(o, "Bad "+timestampHeader+" header: %v", err)
}
var Info = make(map[string]string)
for k, vs := range resp.Header {
k = strings.ToLower(k)
for _, v := range vs {
if strings.HasPrefix(k, headerPrefix) {
Info[k[len(headerPrefix):]] = v
}
}
}
info = &api.File{
ID: resp.Header.Get(idHeader),
Name: resp.Header.Get(nameHeader),
Action: "upload",
Size: resp.ContentLength,
UploadTimestamp: uploadTimestamp,
SHA1: resp.Header.Get(sha1Header),
ContentType: resp.Header.Get("Content-Type"),
Info: Info,
}
return resp, info, nil
}
// Open an object for read
func (o *Object) Open(ctx context.Context, options ...fs.OpenOption) (in io.ReadCloser, err error) {
fs.FixRangeOption(options, o.size)
resp, info, err := o.getOrHead(ctx, "GET", options)
if err != nil {
return nil, err
}
// Don't check length or hash or metadata on partial content
if resp.StatusCode == http.StatusPartialContent {
return resp.Body, nil
}
err = o.decodeMetaData(info)
if err != nil {
_ = resp.Body.Close()
return nil, err
}
// Read sha1 from header if it isn't set
if o.sha1 == "" {
o.sha1 = resp.Header.Get(sha1Header)
fs.Debugf(o, "Reading sha1 from header - %q", o.sha1)
// if sha1 header is "none" (in big files), then need
// to read it from the metadata
if o.sha1 == "none" {
o.sha1 = resp.Header.Get(sha1InfoHeader)
fs.Debugf(o, "Reading sha1 from info - %q", o.sha1)
}
o.sha1 = cleanSHA1(o.sha1)
}
// Don't check length or hash on partial content
if resp.StatusCode == http.StatusPartialContent {
return resp.Body, nil
}
return newOpenFile(o, resp), nil
}