[Zope] Re: Problem with keep-alive timeout

[Andrew Langmead]

If you really need handle an arbitrary processing time. You might  
need to separate the request submission from the processing, and the  
processing from the results display.

Roughly the way it would work would be like dropping your laundry off  
at the cleaners. You bring in the dirty clothes and then get a ticket  
back and an expected due date. You come back around the due date.  
Very likely your clothes are ready and when you present the ticket  
you receive your clean clothes. Occasionally, you get told that due  
to some sort of delay your clothes aren't ready and you are given a  
new due date.

A similar sort of thing could be done with a long running request.  
The initial request gets bundled into some sort of "job" object, and  
a Job ID is returned and a "please wait" page. That page can have a  
delayed redirect to a results page which can take a job ID, determine  
if it is complete and display the result. Meanwhile, you have an  
entirely separate process (perhaps run by the Scheduler product  
<http://cvs.zope.org/Products/Scheduler/> that takes jobs, processes  
them and inserts its results.

Of course, I'm leaving off a lot of details here. Off the top of my  
head, I can think of the following issues that I'm just glossing  
over. I'm sure there are many more:  You don't want job IDs to be  
easily guessable or forgeable, or people might be able to steal each  
others laundry. You have to think about what you do when jobs get  
abandoned, (eventually the clothing racks get full) Finally, (and  
thankfully one that I don't have a laundry analogy for) you may need  
to concern yourself with the fact that the Zope user that is doing  
the job processing is different than the one doing the requesting.