[MarkLogic Dev General] RE: tail-recursion with xdmp:spawn

[Kelly Stirman]

Hi Mike,

Yes, that's a good approach overall for one-off processing. It doesn't provide the robustness of CPF, but it can be easier to set up. It allows for a multi-threaded approach to processing your documents by configuring the number of threads on the task server. 

Your termination condition could use properties. When you have completed an update on a document, add a property flag. Then your processing can look for documents that do not have the property in place.

/foo[not(property::bar)] is fast

/foo[not(property::bar = "baz")] is also fast.

cts:property-query()

Hope this helps.

Also, I think there is new functionality coming in 4.2 that you will appreciate. :-) Hope to see you at the user conference.

Kelly

Message: 4
Date: Wed, 07 Apr 2010 09:26:06 -0400
From: Mike Sokolov <sokolov at ifactory.com>
Subject: [MarkLogic Dev General] tail-recursion with xdmp:spawn
To: General Mark Logic Developer Discussion
	<general at developer.marklogic.com>
Message-ID: <4BBC87EE.9060305 at ifactory.com>
Content-Type: text/plain; charset=ISO-8859-1; format=flowed

Perhaps this won't be news to others on the list, but I was so excited 
to finally stumble on a solution to a problem I have been struggling 
with for years, that I just had to share.

The problem: how to process a large number of documents using xquery only?

This can't be done easily because if all the work is done in a single 
transaction, it eventually runs out of time and space.  But xquery 
modules don't provide an obvious mechanism for flow control across 
multiple transactions.

In the past I've done this by writing an "outer loop" in Java, and more 
recently I tried using CPF.  The problem with Java is that it's 
cumbersome to set up and requires some configuration to link it to a 
database.  I had some success  with CPF, but I found it to be somewhat 
inflexible since it requires a database insert or update to trigger 
processing.  It also requires a bit of configuration to get going.  
Often I find I just want to run through a set of existing documents and 
patch them up in some way or another, (usually to clean up some earlier 
mistake!)

Finally I hit on the solution: I wrote a simple script that fetches a 
batch of documents to be updated, processes the updates, and then, using 
a new statement after ";" to separate multiple transactions, re-spawns 
the same script if there is more work to be done after logging some 
indication of progress.  Presto - an iterative processor.  This 
technique is a little sensitive to running away into an infinite loop if 
you're not careful about the termination condition, but it has many 
advantages over the other methods.

What do you think?


Michael Sokolov
Engineering Director
www.ifactory.com
@iFactoryBoston

PubFactory: the revolutionary e-publishing platform from iFactory



------------------------------