Python Asynchronous Services

[Graeme Matthew]

Hi all, I have hit a serious snag ...

I am writing an app server. I am happy with pythons threading etc however I
am totally confused with handling large request on a socket. For example I
can create a threadingTCPSer ver, the script that calls the app server might
send a file or an xml request. Where I am stuck is how do you handle large
files ?

for example:

while 1:
data = recv(1024)
if not data: break

surely in a thread this will block all other threads until all data is
recieved. I have also been told that one cannot rely on this mechanism as
the socket might end up in a continious loop, i.e you need to send a header
included in the data that states the length (bytes) that is been sent, or
you need some terminator, what happens if the terminator is within the
requests data and not at the end ?

please any help or examples as I have now been on this for 2 days without
any luck. I have looked at asyncore and asynchat, problem is once the server
socket input is completed then their will be an area that is CPU bound where
it will need to call the model, database etc and produce the html request,
this means that control will only be handed back to IO operations when this
is finished, effectively blocking all other requests ....

I suppose another question is can python handle large files and requests at
the same time or is one better off simply forking the process even though
its costly ?

Many thanks .....

[Moshe Zadka]

On Thu, 17 Jul 2003, "Graeme Matthew" <gs*******@ozem ail.com.au> wrote:

Hi all, I have hit a serious snag ... .... please any help or examples as I have now been on this for 2 days without
any luck. I have looked at asyncore and asynchat, problem is once the server
socket input is completed then their will be an area that is CPU bound where
it will need to call the model, database etc and produce the html request,
this means that control will only be handed back to IO operations when this
is finished, effectively blocking all other requests ....

Well, I can recommend Twisted. Twisted uses a producer/consumer model
to send large file, which is fairly transparent [there is a good
cut'n'pastable example of usage of this in twisted.web.sta tic.File].
This should be enough to handle the network part. You claim there
is a CPU-bound operation after that -- if this is really true [and
in my experience, in many of the cases the operation are short enough
that it is not required] -- you can use Twisted's deferToThread to
treat those operations as though they were completely asynchronous.
This uses Twisted's internal threadpool support.

More information: http://twistedmatrix.com/
[Disclaimer: I am part of the Twisted developement team]
--
Moshe Zadka -- http://moshez.org/
Buffy: I don't like you hanging out with someone that... short.
Riley: Yeah, a lot of young people nowadays are experimenting with shortness.
Agile Programming Language -- http://www.python.org/

[Peter Hansen]

Graeme Matthew wrote:

while 1:
data = recv(1024)
if not data: break

surely in a thread this will block all other threads until all data is
recieved.
Why "surely"? It's actually not the case. A blocking call such as to
socket.recv() means that the *calling* thread is blocked until the call
returns, not that other threads are blocked. In fact, the other threads
will immediately be free to run since the calling thread will release
the Global Interpreter Lock before it blocks in the OS socket call.
I have also been told that one cannot rely on this mechanism as
the socket might end up in a continious loop, i.e you need to send a header
included in the data that states the length (bytes) that is been sent, or
you need some terminator, what happens if the terminator is within the
requests data and not at the end ?
Now you're getting into a higher level. You certainly *can* "rely" on
sockets, or receiving data in chunks like the above, but it's certainly
not the easiest way to approach things at this point. You know about
asyncore, and it would work, and Moshe has mentioned Twisted (and I have
to agree that looking there is your best bet); no point reinventing the
wheel, and even if you're trying to learn, you'll probably learn a lot
by examining the source to either of those packages.
please any help or examples as I have now been on this for 2 days without
any luck. I have looked at asyncore and asynchat, problem is once the server
socket input is completed then their will be an area that is CPU bound where
it will need to call the model, database etc and produce the html request,
this means that control will only be handed back to IO operations when this
is finished, effectively blocking all other requests ....
Knowing about how to use a Queue to communicate with a pool of threads, or
a single worker thread, for CPU bound requests, is a good skill...
I suppose another question is can python handle large files and requests at
the same time or is one better off simply forking the process even though
its costly ?

Python is quite capable of this... and forking is definitely not required.
On the other hand, you're better off in general not worrying about performance
issues until you've figured out how to make something work properly. Forking
is not the best approach, but mainly because of its relative complexity, not
because of "cost".

-Peter