June 30th, 2010 | Published in Google App Engine
Today we are happy to announce the 1.3.5 release of the App Engine SDK for both Python and Java developers.
Due to popular demand, we have increased the throughput of the Task Queue API, from 50 reqs/sec per app to 50 reqs/sec per queue. You can also now specify the amount of storage available to the taskqueue in your app, for those with very large queues with many millions of tasks. Stay tuned for even more Task Queue scalability improvements in the future.
Additionally, in this release we’ve also added support for precompilation of Python source files to match the same feature we launched for Java last year. For Python, you can now use precompilation to speed up application loading time and to reduce CPU usage for new app instances. You can enable precompilation by including the following lines in your app.yaml file:derived_file_type:
This will start offline precompilation of Python modules used by your app when you deploy your application. Currently precompliation is off by default for Python applications, but it will be enabled by default in some future release. (Java precompilation has been enabled by default since the release of 1.3.1.)
To give you a taste of what this feature is like, we tested this on a modified version of Rietveld (which included a copy of Django 1.0.4 in the app directory, and which did not use the datastore in its base url). The latency and CPU usage results for the initial load of the application, after uploading a new version of the app and requesting the homepage, were:Before precompilation enabled:
Test 1: 1450ms 1757cpu_ms
Test 2: 1298ms 1523cpu_ms
Test 3: 1539ms 1841cpu_ms
After precompilation enabled:
Test 1: 805ms 669cpu_ms
Test 2: 861ms 702cpu_ms
Test 3: 921ms 803cpu_ms
Of course, any individual app’s performance will vary, so we recommend that you experiment with the setting for your application. Please submit your feedback and results to the support group!
In addition to Task Queues and Python precompilation, we have made a few changes to the Blobstore in 1.3.5 We have added file-like interfaces for reading Blobs. In Python, this is supported through the BlobReader class