I started with a simple Python 2.5 app engine app and migrated it to Python 2.7 in hopes of taking advantage of its multithreaded abilities. After migrating, I noticed that webapp2.RequestHandler instances are all being called from the Main Thread.
I have an AJAX client firing up multiple asynchronous requests. One of the requests I'd like to respond only when certain event occurs on server side. Let's just say that event sleeps for 10 seconds for now. The problem is that sleep occurs in the Main Thread and occupies the thread before processing my second ASYNC request from AJAX. What am I missing?
Here's a stack trace:
PyDevDebug [PyDev Google App Run] dev_appserver.py
MainThread - pid4276_seq4
post [test1.py:53]
dispatch [webapp2.py:570]
call [webapp2.py:1102]
default_dispatcher [webapp2.py:1278]
call [webapp2.py:1529]
Handle [wsgi.py:223]
HandleRequest [wsgi.py:298] HandleRequest [runtime.py:151]
ExecutePy27Handler [dev_appserver.py:1525]
ExecuteCGI [dev_appserver.py:1701]
Dispatch [dev_appserver.py:1803]
Dispatch [dev_appserver.py:719] _Dispatch [dev_appserver.py:2870]
_HandleRequest [dev_appserver.py:3001]
do_POST [dev_appserver.py:2794] handle_one_request [BaseHTTPServer.py:328]
handle [BaseHTTPServer.py:340]
init [SocketServer.py:638]
init [dev_appserver.py:2780]
finish_request [SocketServer.py:323]
process_request [SocketServer.py:310]
_handle_request_noblock [SocketServer.py:284]
handle_request [dev_appserver.py:3991]
serve_forever [dev_appserver.py:4028]
main [dev_appserver_main.py:721]
[dev_appserver_main.py:747]
run_file [dev_appserver.py:167] [dev_appserver.py:171] run [pydevd.py:1090]
[pydevd.py:1397]
Thread-4 - pid4276_seq5 dev_appserver.py
It seems that in the dev (local) server parallel threads are not executed in parallel, but serial.
There is an "experimental dev server" here which you just reminded me of:
An experimental new development server for Google App Engine.
Multithreaded serving for better performance for complex applications and more correct semantics e.g. accessing your own application through urlfetch no longer deadlocks.
That might solve it locally, but I've not tried it personally.