Odtnhj

In one of my projects, MarkLogic consultant advised me to use collection() in cts:search, and in another project, ML consultants have advised to use root element in cts:search. In both projects, we had the same volume of documents. Which one is better with respect to performance?

Let's say we have a document (I am taking a small document just to explain the scenario). It has collection named "demo":

<root>
<child1>ABC</child1>
<child2>DEF</child2>
<child3>GHI</child3>
<child4>JKL</child4>
</root>

Which case is better/more efficient:

cts:search(/root, cts:and-query((....some cts:queries..)))

cts:search(collection("demo"), cts:and-query((....some cts:queries..)))

Please help me with an explanation which one is better than other.

They are both single term lookups, as far as search execution goes, so performance should be the same.

The real distinction is about how you want to manage your content. You can have more than one collection on the same document, so you can slice the same content multiple way, but you can only have one root element. Collections also let you abstract away from the details of document structure: you could have multiple different root elements within the same collection.

As per MarkLogic documentation "MarkLogic's implementation of collections is designed to optimize query performance against large volumes of documents.". So, it means you can identify the difference on only huge database.

I tried to identify this by practical, so I created two XQuery, one with collection and one with element as you suggested. But, I put xdmp:query-trace(fn:true()) at the top of both the XQuery. I ran both the query one by one and analysed my MarkLogic log file.

If it is element XQuery:

2018-11-12 15:16:58.448 Info: App-Services: at 5:12: xdmp:eval("declare namespace sem = &quot;http://marklogic.com/semantics&quo...", (), <options xmlns="xdmp:eval"><database>5310618057872024096</database>...</options>)
2018-11-12 15:16:58.448 Info: App-Services: at 5:12: Analyzing path for search: fn:collection()/sem:triples
2018-11-12 15:16:58.448 Info: App-Services: at 5:12: Step 1 is searchable: fn:collection()
2018-11-12 15:16:58.448 Info: App-Services: at 5:12: Step 2 is searchable: sem:triples
2018-11-12 15:16:58.448 Info: App-Services: at 5:12: Path is fully searchable.
2018-11-12 15:16:58.448 Info: App-Services: at 5:12: Gathering constraints.
2018-11-12 15:16:58.448 Info: App-Services: at 5:12: Step 2 contributed 1 constraint: sem:triples
2018-11-12 15:16:58.449 Info: App-Services: at 5:12: Search query contributed 1 constraint: cts:element-value-query(xs:QName("sem:object"), "taxonomy", ("lang=en"), 1)
2018-11-12 15:16:58.449 Info: App-Services: at 5:12: Executing search.
2018-11-12 15:16:58.464 Info: App-Services: at 5:12: Selected 65964 fragments to filter

and if it is collection XQuery:

2018-11-12 15:20:07.871 Info: App-Services: at 5:11: xdmp:eval("declare namespace sem = &quot;http://marklogic.com/semantics&quo...", (), <options xmlns="xdmp:eval"><database>5310618057872024096</database>...</options>)
2018-11-12 15:20:07.871 Info: App-Services: at 5:11: Analyzing path for search: fn:collection("/triples")
2018-11-12 15:20:07.871 Info: App-Services: at 5:11: Step 1 is searchable: fn:collection("/triples")
2018-11-12 15:20:07.871 Info: App-Services: at 5:11: Path is fully searchable.
2018-11-12 15:20:07.871 Info: App-Services: at 5:11: Gathering constraints.
2018-11-12 15:20:07.871 Info: App-Services: at 5:11: Step 1 contributed 1 constraint: fn:collection("/triples")
2018-11-12 15:20:07.875 Info: App-Services: at 5:11: Search query contributed 1 constraint: cts:element-value-query(xs:QName("sem:object"), "taxonomy", ("lang=en"), 1)
2018-11-12 15:20:07.875 Info: App-Services: at 5:11: Executing search.
2018-11-12 15:20:07.891 Info: App-Services: at 5:11: Selected 65964 fragments to filter

The difference is clearly noticable. If we are using collection query, MarkLogic is doing everthing almost in single step "1" but if it is element query, MarkLogic is doing two step process.