[Zope] Zope vs. Microsoft Dev Days 99 - mini-report

[Brad Clements]

Yesterday I attended an all-day Microsoft Windows 2000 ra-ra event 
called Developer Days 99. It was pretty much what I expected, though 
there were some useful ideas/concepts that might be applicable to Zope.


The track I attended focused on COM/OLE and XML.

1. In Windows 2000, MS has extended the concept of web-server load 
balancing to include COM object load balancing. You can combine 
multiple Windows 2000 machines into a COM "pool". When creating a 
business COM object, the pool manager creates the object on the first 
available/least loaded system.

On the backend, multiple machines are running SQL 7 in a distributed 
manner. This gives a three tear approach (actually 4 if you count the 
client).

Note that Microsoft encourages developers to encapsulate the business 
logic of an app into a COM component. This allows the component to 
be used by WIn32 apps, and by ASP web pages. The web pages only 
serve to present the data, they don't handle any of the logic:

a. multiple web servers load-balanced by the router, each server runs IIS 
5 with ASP

b. multiple "business logic" COM servers instantiating objects to 
process requests made by the ASP pages

c. multiple back-end SQL servers handling requests from the COM 
servers.

Of course all of these layers could run on one physical machine, but 
Windows 2000 is supposed to make it easier to split up the processing 
if you want to.

How does this apply to Zope?

DC has a commercial product which I think applies to multiple web 
servers talking to a single ODB3, but I'm not sure exactly how that 
works. Since DTML methods/documents are not really Objects, I can't 
see a direct correlation with the "business logic" server aspect.

However, the idea of seperating "business logic" from "presentation" 
sounds good. Presumably Python methods represent "business logic" 
and dtml methods are the presentation layer. Making it easier to use 
"external procedures" might be a big win (no flames on PythonMethods 
vs. External Methods, ok).


2. Tracking Session state for web users. By session, I mean 
remembering what the user has selected/set as properties between web 
requests. The example they used was a purchase order application. 
When adding a line item to a PO, the current PO number was saved as 
a session variable.

Although not specific to Windows 2000, one technique they 
demonstrated seemed interesting. Variables or properties specific to a 
given session were packaged up into a COM Stream object (think of it 
as a pickled dict) and saved to an SQL server.

Later requests used a cookie to retrieve the associated session 
information from the SQL server, unpack it and use the saved 
properties. Going back to item 1 above, using a back-end SQL 
database was the only way to save session properties because multiple 
web servers were handling the web requests. Hence the only common 
location for all of the servers was the back-end SQL.

How does this apply to Zope?

Someone could create a package that allows for quick and easy 
storage of session properties to a backend DB (perhaps using a DB 
Connection of some kind). Session retrieval could be via cookie, or 
using the URL insert method mentioned previously on this list.

Using an external DB makes it easy to support multiple web servers (if 
they use the same backend database) for load balancing. A non-
transactional ZODB3 would also be useful, but perhaps not available for 
load-balancing..

There are two ways I can think of to use such a product.

<dtml-with>, which is explicit, or

implicitely pushing the session dict into the namespace stack. The 
question of which comes first, REQUEST, or session dict, would be 
important.

Another other issue is how to clean out old session objects from the 
database using ZCron.. 

Thoughts?

3. XML and OLE DB

ADO 2.5 (apparently only available with Windows 2000 they said) can 
now export and import XML.

They demonstrated using MS's XMLDOM component in conjunction 
with an ADO recordset (the result of a select). The ADO recordset now 
has a GetText and PutText function which exports/imports XML.

By passing an XSL sheet to the XMLDOM component, the ADO XML 
output could be massaged for the target client. For example, IE5 clients 
got pretty much the raw XML, whereas non-XML capable clients got 
basic HTML.

I haven't looked at the Zope XMLDocument, but ADO's built-in support of 
XML to represent data sets seems compelling. For example, could a 
ZClass property sheet have a GetXML or PutXML feature? Would that 
make sense?

ZODB knows how to pickle objects and restore them. Seems to me that 
ZODB might also be able to represent python objects in XML. I'm not 
sure how this would be useful, but you could imagine creating a python 
class that consists of just data members, and being able to easily get it 
as XML. I think the dtml record feature addresses this for HTML.. 

Without data type information, XML'd python classes will be tough...


4. Finally I saw something interesting for Zope.

IIS 5 has two new features.

Server.Execute, and Server.Transfer

The first, Server.Execute, allows an ASP script to be called "as a 
subroutine". We have this now with Zope, its very easy.. <dtml-var ...>

The second, Server.Transfer, allows an ASP script to pass control to 
another script on the same server without using a redirect.

This one seems interesting and the idea might be applicable to Zope. 
Basically it performs the same action as RESPONSE.redirect(), but 
does not send the redirect to the client.

This is useful when you want to render a different page after processing 
a form, for example, but want the rendering of the second page to look 
like a new request from the user, without forcing the client to actually 
perform the redirect.

This improves the response time for the client.

I looked at HTTPResponse, and it looks like redirect() really sends the 
redirect back to the client. I didn't see an obvious function, like 
localredirect()...

Perhaps using the new RETURN tag would work, but I don't think that 
starts a new transaction. You could imagine an "add" form that updates 
a database, then wants to redisplay the master detail page.

If the "add" form redirects the client to the master detail page using 
something like the return tag, if the master detail page generates an 
exception, the add will be rolled back because its part of the same 
transaction. Right?

I think a function like RESPONSE.localredirect() would be handy. It 
would accept URLs that only reference the local server. Internally it 
would complete the current transaction, then "fake up" a new request 
from the client to the new URL, as if the client had actually requested 
the new URL.

Of course, RESPONSE.redirect() could be made smarter, such that it 
would automatically do this for local URLs, but perhaps need an option 
to force a client-side redirect if desired.

Thoughts?



Brad Clements,                bkc@murkworks.com   (315)268-1000
http://www.murkworks.com                          (315)268-9812 Fax
netmeeting: ils://ils.murkworks.com               ICQ: 14856937