Querying metadata vs data

Question

Hi There,
Say I had the option of coming up with whatever schema that works best for BDB XML from a performance standpoint to persist xml documents. That said, I now have the option of storing metadata as metadata or as part of the regular data/content. Given that both metadata and data/content can be indexed, is one option preferable over the other ?
Just to be really clear, I am considering b/w options (1) and (2) below. Assume that 'custCreatedBy' will be indexed for both (1) and (2).
(1)
Metadata: <custCreatedBy>bar</custCreatedBy>
Data/Content:
<cust>
<id>123</id>
<name>foo</name>
</cust>
(2)
Metadata: None
Data/Content:
<cust>
<custCreatedBy>bar</custCreatedBy>
<content>
<id>123</id>
<name>foo</name>
</content>
</cust>
Cheers,
Shashi

Answer 1

Whether you opt to put data in the document or in
metadata depends mainly on whether you need that data
available outside of BDB XML. If you do, it should go
in the document. There is not a major performance
difference, because both can be looked up directly
using indexes if they exist. I use metadata primarily
when I need to store data that cannot exist in
my XML, for such reasons as schema constraints, etc.
Keep in mind that metadata is not XML, it's just
name-value pairs.
Agreed. Makes sense.
When you can create your own XML, some of the things
to keep in mind for ease of use with BDB XML are:
I guess in my case, I wasn't very clear but I do not have as much control over the content as much as the rest of it which would be the metadata. But regardless, all of the below points are very valid and useful (some of which I gathered from Chapter 7). Thanks for reiterating these anyways.
• Indexes are not used for negative presence. That
is, you should avoid XML where you need to test for
the absence of certain nodes. If you wanted to query
for all documents missing an <id/> element, for
example, indexes would not help. You'd need to
instead test for documents with <id/> elements with
an empty string value, and look that up with a
query.Interesting. I find it hard to understand why a "presence" index cannot be leveraged for this as well. Anyways, just to satisfy my curiosity, I played with the two queries (exist, not exist) - below are query plans for both usecases with a "presence" index for "id" node.
dbxml> queryPlan collection('junk.dbxml')/cust[id]
<XQuery>
<Navigation>
<QueryPlanFunction result="collection" container="junk.dbxml">
<OQPlan>P(node-element-presence-none,=,id)</OQPlan>
</QueryPlanFunction>
<Step axis="child" name="cust" nodeType="element">
<Predicates>
<Step axis="child" name="id" nodeType="element">
<OQPlan>P(node-element-presence-none,=,id)</OQPlan>
</Step>
</Predicates>
</Step>
</Navigation>
</XQuery>
dbxml> queryPlan collection('junk.dbxml')/cust[not(id)]
<XQuery>
<Navigation>
<QueryPlanFunction result="collection" container="junk.dbxml">
<OQPlan>U</OQPlan>
</QueryPlanFunction>
<Step axis="child" name="cust" nodeType="element">
<Predicates>
<Function name="{http://www.w3.org/2005/04/xpath-functions}:not">
<Step axis="child" name="id" nodeType="element">
<OQPlan>P(node-element-presence-none,=,id)</OQPlan>
</Step>
</Function>
</Predicates>
</Step>
</Navigation>
</XQuery>
I presume that you are referring to the first bolded line (indicating that the index is not kicking in for "cust[not(id)]" query. However, I am curious to learn about the second bolded line above which seems to indicate that the "presence" index is indeed kicking in. In my limited experience with db xml, this is the first query plan where I have seen the <OQPlan> node appear twice. Wondering if someone could shed more light or pointers to understand this query plan more. I did see the "Caution" line in your book citing that it is not documented anywhere - but i am hoping against hope that maybe that is not true anymore - since I see myself banking on it to do some tuning of my own.
Thanks again for all your insightful comments/tips. Cheers.