Twisted integration (was Re: [Zope3-dev] Re: uniformly configurable request factories: bugfix or proposal?)

[Gary Poster]

On Aug 5, 2005, at 6:27 AM, Stephan Richter wrote:

> On Thursday 04 August 2005 23:34, Tres Seaver wrote:
>
>> Can you point me to the discussion of the tradeoffs involved in  
>> moving
>> to Twisted?
>>
>
> I have not written the proposal yet, but you can play around with  
> the branch
> if you like.

Then perhaps these comments should wait for the proposal, but I'll  
make 'em anyway. :-)

>> I have a (perhaps unfounded) sense that its HTTP server
>> implementation has serious scaling problems, even compared to  
>> ZServer.
>
> Not the new web2 server. Itamar and I did some very quick (non- 
> scientific)
> profiling and web2 was just slightly slower than zserver. It was  
> just a very
> small fraction in comparison to the time the publisher requires.

...

> web2 is a reimplementation from James Knight that concentrates more  
> on the
> HTTP standard itself and provides a solid WSGI interface.

...

> Jim already convinced me to leave zserver has an alternative. Not  
> quiet sure
> how I will do this, but I hope it will be as easy as switching a  
> few options
> in zope.conf.

...

> It will not be the only way, but the default one.

This looks pretty problematic to me.  Going from 3.1 to 3.2, we are  
going to change the default publisher implementation from a proven-in- 
the-field publisher (arguably coming from a long line of proven-in- 
the-field publishers from Zope 2), to an unproven new publisher that  
is actually slower than the current implementation.  I'm -1 on that.

I am +1 on making Zope 3 WSGI-aware.  I am +1 on making Twisted an  
optional back end.  I am potentially +1 on making Twisted the default  
back end after we have a reasonably significant amount of concrete in- 
the-field experience with it.

Gary