Eric Bidelman, Google Apps APIs team
February 2010
Introduction
Current web standards provide no reliable mechanism to facilitate the HTTP upload of large files. As a result, file uploads at Google and other sites have traditionally been limited to moderate sizes (e.g. 100 MB). For services like the YouTube and the Google Documents List APIs which support large file uploads, this presents a major hurdle.
The Google Data resumable protocol directly addresses the aforementioned issues by supporting resumable POST/PUT HTTP requests in HTTP/1.0. The protocol was modeled after the ResumableHttpRequestsProposal suggested by Google Gears team.
This document describes how to incorporate Google Data's resumable upload feature into your applications. The examples below use the Google Documents List Data API. Note that additional Google APIs that implement this protocol may have slightly different requirements/response codes/etc. Please consult the service's documentation for the specifics.
The Resumable Protocol
Initiating a resumable upload request
To initiate a resumable upload session, send an HTTP POST request to the resumable-post link. This link is found at the feed level. The DocList API's resumable-post link looks like:
<link rel="http://schemas.google.com/g/2005#resumable-create-media" type="application/atom+xml"
href="https://docs.google.com/feeds/upload/create-session/default/private/full"/>
The body of your POST request should be empty or contain an Atom XML entry and must not include the actual file contents. The example below creates a resumable request to upload a large PDF, and includes a title for the future document using the Slug header.
POST /feeds/upload/create-session/default/private/full HTTP/1.1
Host: docs.google.com
GData-Version: <version number>
Authorization: <your authorization header here> Content-Length: 0 Slug: MyTitle X-Upload-Content-Type: application/pdf X-Upload-Content-Length: 1234567 <empty body>
The X-Upload-Content-Type and X-Upload-Content-Length headers should be set to the mimetype and size of the file you will eventually upload.
Here is another example request that instead uploads a word document. This time, Atom metadata is included and will be applied to the final document entry.
POST /feeds/upload/create-session/default/private/full?convert=false HTTP/1.1
Host: docs.google.com
GData-Version: <version number>
Authorization: <your authorization header here> Content-Length: 292 Content-Type: application/atom+xml X-Upload-Content-Type: application/msword X-Upload-Content-Length: 7654321 <?xml version='1.0' encoding='UTF-8'?> <entry xmlns="http://www.w3.org/2005/Atom" xmlns:docs="http://schemas.google.com/docs/2007"> <category scheme="http://schemas.google.com/g/2005#kind" term="http://schemas.google.com/docs/2007#document"/> <title>MyTitle</title> <docs:writersCanInvite value="false"/> </entry>
The server's response from the initial POST is a unique upload URI in the Location header and an empty response body:
HTTP/1.1 200 OK
Location: <upload_uri>
The unique upload URI will be used to upload the file chunks.
Note: The initial POST request does not create a new entry in the feed. This only happens when the entire upload operation has completed.
Uploading a file
The resumable protocol allows, but doesn't require, content to be uploaded in 'chunks', because there are no inherent restrictions in HTTP on request sizes. Your client is free to choose its chunk size or just upload the file as a whole. This example uses the unique upload URI to issue a resumablePUT. The following example sends the first 100000 bytes of 1234567 byte PDF file:
PUT <upload_uri> HTTP/1.1
Host: docs.google.com
Content-Length: 100000 Content-Type: application/pdf Content-Range: bytes 0-99999/1234567 <bytes 0-99999>
Read more information on the Content-Range header here.
Server responds with the current byte range that has been stored:
HTTP/1.1 308 Resume Incomplete Content-Length: 0 Range: 0-99999
Your client should continue to PUT each chunk of the file until the entire file has been uploaded. Until the upload is complete, the server will respond with an HTTP 308 Resume Incomplete and the byte range it knows about in the Range header. Clients must use the Range header to determine where to start the next chunk. Therefore, do not assume that the server received all bytes originally sent in the PUT request.
Note: The server may issue a new unique upload URI in the Location header during a chunk. Your client should check for an updated Locationand use that URI to send the remaining chunks to the server.
When the upload is complete, the response will be the same as if the upload had been made using the API's non-resumable upload mechanism. That is to say, a 201 Created will be returned along with the <atom:entry>, as created by the server. Subsequent PUTs to the unique upload URI will return the same response as what was returned when the upload completed. After a period of time, the response will be 410 Gone or 404 Not Found.
Resuming an upload
If your request is terminated prior to receiving a response from the server or if you receive an HTTP 503 response from the server, you can query the current status of the upload by issuing an empty PUT request on the unique upload URI.
Client polls the server to determine which bytes it has received:
PUT <upload_uri> HTTP/1.1
Host: docs.google.com
Content-Length: 0
Content-Range: bytes */100
Server responds with the current byte range:
HTTP/1.1 308 Resume Incomplete Content-Length: 0 Range: 0-42
Note: The server may issue a new unique upload URI in the Location header during a chunk. Your client should check for an updated Locationand use that URI to send the remaining chunks to the server.
Finally, the client resumes where the server left off:
PUT <upload_uri> HTTP/1.1
Host: docs.google.com
Content-Length: 57
Content-Range: 43-99/100
<bytes 43-99>
Updating an existing resource
Similar to initiating a resumable upload session, you can utilize the resumable upload protocol to replace an existing file's content. To start a resumable update request, send an HTTP PUT to the entry's link with rel='...#resumable-edit-media'. Each media entry will contain such a link if the API supports updating the resource's content.
As an example, a document entry in the DocList API will contain a link similar to:
<link rel="http://schemas.google.com/g/2005#resumable-edit-media" type="application/atom+xml"
href="https://docs.google.com/feeds/upload/create-session/default/private/full/document%3A12345"/>
Thus, the initial request would be:
PUT /feeds/upload/create-session/default/private/full/document%3A12345 HTTP/1.1
Host: docs.google.com
GData-Version: <version number>
Authorization: <your authorization header here> If-Match: <ETag or * here> Content-Length: 0 X-Upload-Content-Length: 1000 X-Upload-Content-Type: text/plain <empty body>
To update a resource's metadata and content at the same time, include Atom XML instead of an empty body. See the example in the Initiating a resumable upload request section.
When the server responds with the unique upload URI, send a PUT with your payload. Once you have the unique upload URI, the process for updating the file's content is the same as uploading a file.
This particular example will update the existing document's content in one shot:
PUT <upload_uri> HTTP/1.1
Host: docs.google.com
Content-Length: 1000
Content-Range: 0-999/1000
<bytes 0-999>
Client library examples
Below are samples of uploading a movie file to Google Docs (using the resumable upload protocol) in the Google Data client libraries. Note, that not all of the libraries support the resumable feature at this time.
int MAX_CONCURRENT_UPLOADS = 10;
int PROGRESS_UPDATE_INTERVAL = 1000;
int DEFAULT_CHUNK_SIZE = 10485760;
DocsService client = new DocsService("yourCompany-yourAppName-v1");
client.setUserCredentials("[email protected]", "pa$$word");
// Create a listener
FileUploadProgressListener listener = new FileUploadProgressListener(); // See the sample for details on this class.
// Pool for handling concurrent upload tasks
ExecutorService executor = Executors.newFixedThreadPool(MAX_CONCURRENT_UPLOADS);
// Create {@link ResumableGDataFileUploader} for each file to upload
List uploaders = Lists.newArrayList();
File file = new File("test.mpg");
String contentType = DocumentListEntry.MediaType.fromFileName(file.getName()).getMimeType();
MediaFileSource mediaFile = new MediaFileSource(file, contentType);
URL createUploadUrl = new URL("https://docs.google.com/feeds/upload/create-session/default/private/full");
ResumableGDataFileUploader uploader = new ResumableGDataFileUploader(createUploadUrl, mediaFile, client, DEFAULT_CHUNK_SIZE,
executor, listener, PROGRESS_UPDATE_INTERVAL);
uploaders.add(uploader);
listener.listenTo(uploaders); // attach the listener to list of uploaders
// Start the upload(s)
for (ResumableGDataFileUploader uploader : uploaders) {
uploader.start();
}
// wait for uploads to complete
while(!listener.isDone()) {
try {
Thread.sleep(100);
} catch (InterruptedException ie) {
listener.printResults();
throw ie; // rethrow
}
// Chunk size in MB
int CHUNK_SIZE = 1;
ClientLoginAuthenticator cla = new ClientLoginAuthenticator(
"yourCompany-yourAppName-v1", ServiceNames.Documents, "[email protected]", "pa$$word");
// Set up resumable uploader and notifications
ResumableUploader ru = new ResumableUploader(CHUNK_SIZE);
ru.AsyncOperationCompleted += new AsyncOperationCompletedEventHandler(this.OnDone);
ru.AsyncOperationProgress += new AsyncOperationProgressEventHandler(this.OnProgress);
// Set metadata for our upload.
Document entry = new Document()
entry.Title = "My Video";
entry.MediaSource = new MediaFileSource("c:\\test.mpg", "video/mpeg");
// Add the upload uri to document entry.
Uri createUploadUrl = new Uri("https://docs.google.com/feeds/upload/create-session/default/private/full");
AtomLink link = new AtomLink(createUploadUrl.AbsoluteUri);
link.Rel = ResumableUploader.CreateMediaRelation;
entry.DocumentEntry.Links.Add(link);
ru.InsertAsync(cla, entry.DocumentEntry, userObject);
- (void)uploadAFile {
NSString *filePath = @"~/test.mpg";
NSString *fileName = [filePath lastPathComponent];
// get the file's data
NSData *data = [NSData dataWithContentsOfMappedFile:filePath];
// create an entry to upload
GDataEntryDocBase *newEntry = [GDataEntryStandardDoc documentEntry];
[newEntry setTitleWithString:fileName];
[newEntry setUploadData:data];
[newEntry setUploadMIMEType:@"video/mpeg"];
[newEntry setUploadSlug:fileName];
// to upload, we need the entry, our service object, the upload URL,
// and the callback for when upload has finished
GDataServiceGoogleDocs *service = [self docsService];
NSURL *uploadURL = [GDataServiceGoogleDocs docsUploadURL];
SEL finishedSel = @selector(uploadTicket:finishedWithEntry:error:);
// now start the upload
GDataServiceTicket *ticket = [service fetchEntryByInsertingEntry:newEntry
forFeedURL:uploadURL
delegate:self
didFinishSelector:finishedSel];
// progress monitoring is done by specifying a callback, like this
SEL progressSel = @selector(ticket:hasDeliveredByteCount:ofTotalByteCount:);
[ticket setUploadProgressSelector:progressSel];
}
// callback for when uploading has finished
- (void)uploadTicket:(GDataServiceTicket *)ticket
finishedWithEntry:(GDataEntryDocBase *)entry
error:(NSError *)error {
if (error == nil) {
// upload succeeded
}
}
- (void)pauseOrResumeUploadForTicket:(GDataServiceTicket *)ticket {
if ([ticket isUploadPaused]) {
[ticket resumeUpload];
} else {
[ticket pauseUpload];
}
}
import os.path
import atom.data
import gdata.client
import gdata.docs.client
import gdata.docs.data
CHUNK_SIZE = 10485760
client = gdata.docs.client.DocsClient(source='yourCompany-yourAppName-v1')
client.ClientLogin('[email protected]', 'pa$$word', client.source);
f = open('test.mpg')
file_size = os.path.getsize(f.name)
uploader = gdata.client.ResumableUploader(
client, f, 'video/mpeg', file_size, chunk_size=CHUNK_SIZE, desired_class=gdata.docs.data.DocsEntry)
# Set metadata for our upload.
entry = gdata.docs.data.DocsEntry(title=atom.data.Title(text='My Video'))
new_entry = uploader.UploadFile('/feeds/upload/create-session/default/private/full', entry=entry)
print 'Document uploaded: ' + new_entry.title.text
print 'Quota used: %s' % new_entry.quota_bytes_used.text
For complete samples and source code reference, see the following resources:
- Java library sample app and source
- Objective-C library sample app
- .NET library source