Twiggle, a Twitter and Google Maps mashup, running on an Android tablet (Motorola Xoom). Twiggle was written in Flapjax
Documents List API
There are a number of ways to add resources to your Google Documents List using the API. Most commonly, clients need to upload an existing resource, rather than create a new, empty one. Legacy clients may be doing this in an inefficient way. In this post, we’ll walk through why using resumable uploads makes your client more efficient.
The resumable upload process allows your client to send small segments of an upload over time, and confirm that each segment arrived intact. This has a number of advantages.
Resumable uploads have a customizable memory footprint on client systems
Since only one small segment of data is sent to the API at a time, clients can store less data in memory as they send data to the API. For example, consider a client uploading a PDF via a regular, non-resumable upload in a single request. The client might follow these steps:
- Open file pointer to PDF
- Pass file pointer and PDF to client library
- Client library starts request
- Client library reads 100,000 bytes and immediately sends 100,000 bytes
- Client library repeats until all bytes sent
- Client library returns response
But that 100,000 bytes isn’t a customizable value in most client libraries. In some environments, with limited memory, applications need to choose a custom chunk size that is either smaller or larger.
The resumable upload mechanism allows for a custom chunk size. That means that if your application only has 500KB of memory available, you can safely choose a chunk size of 256KB.
Resumable uploads are reliable even though a connection may not be
In the previous example, if any of the bytes fail to transmit, this non-resumable upload fails entirely. This often happens in mobile environments with unreliable connections. Uploading 99% of a file, failing, and restarting the entire upload creates a bad user experience. A better user experience is to resume and upload only the remaining 1%.
Resumable uploads support larger files
Traditional non-resumable uploads via HTTP have size limits depending on both the client and server systems. These limits are not applicable to resumable uploads with reasonable chunk sizes, as individual HTTP requests are sent for each chunk of a file. Since the Documents List API now supports file sizes up to 10GB, this is very important.
Resumable upload support is already in the client libraries for Google Data APIs
The Java, Python, Objective-C, and .NET Google Data API client libraries all include a mechanism by which you can initiate a resumable upload session. Examples of uploading a document with resumable upload using the client libraries is detailed in the documentation. Additionally, the new Documents List API Python client library now uses only the resumable upload mechanism. To use that version, make sure to follow these directions.
The Pictarine: Building a mashup in the cloud with Google App Engine
Pictarine is a photo management web application, launched in 2010, that allows people to easily manage and share all of their photos from Flickr, Picasa, Facebook, Twitter and other sites. Pictarine developers Guillaume Martin and Maxime Rafalimanana have contributed the following post discussing their experiences using Google App Engine and Google Web Toolkit.
From the start, we used Google technologies in developing Pictarine and we wanted to share our experience with them so far. In this post, we will shed some light on the weaknesses and strengths we found in Google Web Toolkit (GWT) and Google App Engine. We will also discuss how we leveraged GWT to build a new technology that allows Pictarine to seamlessly display photos from the computer directly into the browser. The following diagram is an overview of how our application works.
Building a mashup in the cloud with Google App Engine
The Pictarine team is made of a web designer and two developers who previously worked mainly with Java based enterprise technologies and had a little experience with web technologies. When we started the project in early 2009, we were quite open on learning new languages like Python or Ruby, but when App Engine announced that Java would be supported, we were really excited to give Google App Engine a try.
The first few months, learning about the App Engine environment was quite easy and dare I say fun. Testing our code on Google’s servers from Eclipse IDE was only one click away. So we built our first prototype fast and we quickly decided to adopt App Engine. Then we started to build the core of our application: the engine that uses the API from Flickr, Picasa, Facebook to fetch the users’ photos. This is where we hit the first limitations of App Engine. Most users have a lot of photos on these services and retrieving them can take some time. But App Engine has strict limits on how long a request should last: an outgoing HTTP request cannot last more than 10 seconds and cannot process a request for more than 30 seconds. So while building our architecture we found ourselves writing approximately one third of our code dealing with these limitations: paginating our requests, creating background tasks to store data in small batches, etc.
In early 2010, when we launched our alpha version, everything went smoothly. We had some good press coverage and App Engine met our expectations in handling our first users. During 2010, we worked on implementing new features requested by our users, and during this period of time we were really impressed by the way App Engine evolved. Many of the limitations were lifted and great new features were added. We are now able to use Task Queues for requests that last up to 10 minutes, which we fully use to sync our users’ photos and albums. One of the features we like the most is the Channel API, a push notification system that allows us to instantly show a photo in every connected browser as soon as it is uploaded.
App Engine is still not perfect but has greatly improved and when we see its roadmap, we are quite confident it will continue to improve.