This implements the Open Web Application synchronization protocol. The client is implemented in the main openwebapps repository.
To run the server in a configuration that is fairly equivalent to the https://myapps.mozillalabs.com setup, use:
$ git clone https://github.com/mozilla/openwebapps.git $ export OPENWEBAPPS=$(pwd)/openwebapps $ git clone https://github.com/mozilla/apps.mozillalabs.com.git $ export APPS=$(pwd)/apps.mozillalabs.com $ git clone https://github.com/mozilla/appsync.git $ cd appsync $ pip -E test-env install -r prod-reqs.txt $ ./test-env/bin/paster serve etc/appsync-dev.ini -n myapps --reload
Now you can load http://localhost:5000/ to get the dashboard, and
to get some apps to install go to
http://localhost:5000/apps/appdir/
You may want to change [storage] in etc/appsync-dev.ini to use
a SQLite database (there are comments in that file to show how).
To perform a stress test on the server, install Funkload:
$ pip -E test-env install Funkload
Then run the Appsync server:
$ ./test-env/bin/paster serve etc/appsync-dev.ini --daemon
And verify the server is properly set by running a single test:
$ make loadonce cd loadtest; ../bin/fl-run-test simple.py . ----------------------------------------------------- Ran 1 test in 2.742s OK
You can now run a full stress test:
$ make load Benching ======== * setUpBench hook: ... done. Cycle #0 with 5 virtual users ----------------------------- * setUpCycle hook: ... done. * Current time: 2011-12-05T14:47:14.737983 * Starting threads: ..... done. * Logging for 30s (until 2011-12-05T14:47:44.796607): ........ ...
You can configure the load test with a few options:
- HOST: the AppSync Server the test is run against (default: http://localhost:5000)
- DURATION: the duration of a cycle in seconds (default: 30)
- CYCLES: cycles to run. Number of virtual users to run per cycle, separated by columns. (default: 5:10:20)
Let's run 50, 100, then 200 users for a duration of 1 minute on the myapps.example.com
$ make load HOST=http://myapps.example.com DURATION=60 CYCLES=50:100:200
To slow down the clients requests, you can set up a special value in Memcache that will add a X-Sync-Poll header to all GETs' responses.
The appsync-backoff script is provided for this:
$ appsync-backoff get The Backoff is currently set to 10 seconds. $ appsync-backoff set 20 Backoff set to 20 seconds. $ appsync-backoff del Backoff removed $ appsync-backoff get No Backoff has been set in Memcached $ appsync-backoff -m 127.0.0.1 get No Backoff has been set in Memcached