> Hi all,
>
> Utter uwsgi newbie here. I'm running a uwsgi server in a JSON RPC client
> server architecture. Some of the calls can take a few minutes to
> complete, and the problem I'm having is that the client and server lose
> their connection after about a minute. The server continues to process
> the data for another 30s to a minute, writes the data to another server,
> and then gets a broken pipe when trying to write to the client. When I
> run the application with wsgiref.simple_server.make_server everything
> works correctly. Pertinent information is below:
> _
> uwsgi command line:_
> uwsgi --master --processes 20 --cheaper 4 --gevent 5 --http :10001
> --wsgi-file [file]
>
> _bottom of the stack trace client side:_
> /usr/lib/python2.7/httplib.pyc in _read_status(self)
> 369 # Presumably, the server closed the connection before
> 370 # sending a valid response.
> --> 371 raise BadStatusLine(line)
> 372 try:
> 373 [version, status, reason] = line.split(None, 2)
>
> BadStatusLine: ''
>
> _Error in the server logs/std*:_
> SIGPIPE: writing to a closed pipe/socket/fd (probably the client
> disconnected) on request / (ip 127.0.0.1) !!!
> IOError: write error
> [pid: 3138|app: 0|req: 1/1] 127.0.0.1 () {34 vars in 897 bytes} [Fri Feb
> 15 09:13:15 2013] POST / => generated 0 bytes in 44679 msecs (HTTP/1.1
> 200) 4 headers in 149 bytes (5 switches on core 4)
>
> Obviously the client and server are on the same machine. I've tried
> running with -t at 0 and 300 and without gevent. Looking here:
> http://uwsgi-docs.readthedocs.org/en/latest/ThingsToKnow.html ...the
> most pertinent note seems to be:
>
> If an HTTP request has a body (like a POST request generated by
> a form), you /have/ to read (consume) it in your application....
>
> The method params are in JSON in the request body, so I presume the body
> is being consumed since the method is getting the params.
>
> There's a brief (~5s) CPU spike during the method, usually to around
> 30-50% per top. Memory, as far as I can tell, never gets above 1%.
>
> It's not clear to me what to do next to figure out why this is happening
> - it seems it must be related to the uwsgi container as the wsgiref
> container doesn't have this problem. Any advice would be appreciated.
>
> Thanks, Gavin
>
>
If there is no data transferred between the client and the server in 60
seconds (default http router timeout), the HTTP router will close the
connection.
This is the behaviour you will get on basically all of the "sane" servers
out there :)
You can increase the timeout with --http-timeout <n>, but the best
approach (expecially if the generation time is not deterministic) is
sending chunks continuosly. Obviously if your specific app could not be
rewritten in that way, just increase timeouts (you are under gevent so if
you have done right you will be able to manage lot of concurrent requests
even with really big timeouts in place)
--
Roberto De Ioris
http://unbit.it
_______________________________________________
uWSGI mailing list
[email protected]
http://lists.unbit.it/cgi-bin/mailman/listinfo/uwsgi
