Random thought:

Instead of trying to find a way to specify in XPath a sortable expression that 
matches up with a set of range indexes, it might be cleaner to specify the 
ordering range index directly, such as:

for $item in cts:search(...)
order by cts:field-reference("fieldname")
return $item

Perk: the behavior when the index doesn't exist will be an explanatory error 
instead of slow execution.

Note this doesn't work today.  Just thinking aloud.

-jh-

On Aug 2, 2013, at 12:45 PM, Gajanan Chinchwadkar 
<[email protected]> wrote:

> In general I agree with Mike that you will be able to use an index of kindof 
> /a/(b|c).
> 
> But creating a good path range index for 
> $result/(old/path/date|new/path/m:sort-date) may not be easy. ML 6 doesn't 
> allow you to create an index with a top level grouping operator, i.e. 
> (old/path/date|new/path/m:sort-date). You can create an index of type say, 
> indexroot/(old/path/date|new/path/m:sort-date). But then in order to get that 
> index used by fast order by the "indexroot" must match against $result (in 
> fact it should be a leaf of $result).
> 
> It will be a good idea for Ron to contact Stephen Buxton and submit an RFE.
> 
> -----Original Message-----
> From: [email protected] 
> [mailto:[email protected]] On Behalf Of Michael Blakeley
> Sent: Friday, August 02, 2013 10:49 AM
> To: MarkLogic Developer Discussion
> Subject: Re: [MarkLogic Dev General] Fast order-by with multiple range 
> indexes?
> 
> With ML6 I think you could create a single, useful path range index on both 
> elements. There is a table of permitted XPath syntax at 
> http://docs.marklogic.com/guide/admin/range_index#id_54948 and syntax like 
> /a/(b|c) is supported.
> 
> So you're answer might be "this will be slow now, but we have a plan to make 
> it faster when we upgrade".
> 
> -- Mike
> 
> On 2 Aug 2013, at 10:40 , Ron Hitchens <[email protected]> wrote:
> 
>> 
>>  This one is on 5.x.  I plan to move them to 6.x for the next phase 
>> (where everything goes to a consistent model), but it's not possible 
>> now.
>> 
>>  If you know a way on 6.x I'd love to see it.  We can't use path 
>> range indexes, but how would that work?  Something like this?:
>> 
>> order by xs:date ($result/(old/path/date|new/path/m:sort-date))
>> 
>> On Aug 2, 2013, at 6:13 PM, Michael Blakeley <[email protected]> wrote:
>> 
>>> Which release?
>>> 
>>> So you're using an element-range index? What about using a path-range index?
>>> 
>>> -- Mike
>>> 
>>> On 2 Aug 2013, at 09:26 , Ron Hitchens <[email protected]> wrote:
>>> 
>>>> 
>>>> I have a sorting problem that I can't find a good solution for.  I'm 
>>>> working on a project where a lot of content exists in one form which 
>>>> was not designed for efficient searching or sorting.
>>>> In fact, MarkLogic is not used at all for search at the moment, 
>>>> that's what I'm adding.
>>>> 
>>>> This existing format has multiple versions of the content in each 
>>>> document, with an element range index on an xs:date field.
>>>> I can do efficient sorts on this content alone using the ranged date 
>>>> field in an "order by" clause.
>>>> 
>>>> Here's the complication: a new type of content is being added in a 
>>>> newer, more MarkLogic-friendly schema.  These documents all have a 
>>>> common metadata section with a ranged date field.  Different name 
>>>> and namespace, but serving the same purpose.
>>>> 
>>>> My problem is that I need to do searches across both types of 
>>>> content and sort them together.  Searching one kind or the other and 
>>>> sorting by their respective date fields works great for massive 
>>>> result sets.  But doing them together blows the expanded tree cache 
>>>> if the result set is large.
>>>> 
>>>> Because of the odd layout of the old content, my searchable 
>>>> expression is rather funky and looks something like this:
>>>> 
>>>> cts:search 
>>>> (fn:doc()/(/container/group[@state="live]/doc[fn:not(@foo)]|x:new1|x
>>>> :new2), $q, "unfiltered")
>>>> 
>>>> Note that the first one returns a sub-element of the document, which 
>>>> is actually a fragment root.  The other two on the end return root 
>>>> elements.
>>>> 
>>>> A FLWOR like this doesn't work:
>>>> 
>>>> for $result in cts:search ( . . .)
>>>> order by xs:date (($result/old/path/date, 
>>>> $result/new/path/m:sort-date)[1]) return $result
>>>> 
>>>> It runs but ok and will do the right thing if the result set is 
>>>> reasonably small (a few thousand) will blow the cache if there are 
>>>> too many results.  Trying to ignore one of the dates also blows he 
>>>> cache:
>>>> 
>>>> for $result in cts:search ( . . .)
>>>> order by xs:date ($result/old/path/date) return $result
>>>> 
>>>> But removing the last two components of the XPath (|x:new1|x:new2) 
>>>> will then run fast.  I'm not sure why this prevents the range index 
>>>> from kicking in, probably because of the complexity of the XPath.
>>>> 
>>>> Sorting combined results by relevance in either direction is fast.
>>>> 
>>>> Does anyone have a voodoo trick to enable fast sorting using values 
>>>> from two different range indexes?  I don't need to look into the 
>>>> documents the get the sort keys, it seems like it shouldn't have to 
>>>> consume expanded tree cache space for this.
>>>> 
>>>> ---
>>>> Ron Hitchens {mailto:[email protected]}   Ronsoft Technologies
>>>>  +44 7879 358 212 (voice)          http://www.ronsoft.com
>>>>  +1 707 924 3878 (fax)              Bit Twiddling At Its Finest
>>>> "No amount of belief establishes any fact." -Unknown
>>>> 
>>>> 
>>>> 
>>>> 
>>>> _______________________________________________
>>>> General mailing list
>>>> [email protected]
>>>> http://developer.marklogic.com/mailman/listinfo/general
>>>> 
>>> 
>>> _______________________________________________
>>> General mailing list
>>> [email protected]
>>> http://developer.marklogic.com/mailman/listinfo/general
>> 
>> _______________________________________________
>> General mailing list
>> [email protected]
>> http://developer.marklogic.com/mailman/listinfo/general
>> 
> 
> _______________________________________________
> General mailing list
> [email protected]
> http://developer.marklogic.com/mailman/listinfo/general
> _______________________________________________
> General mailing list
> [email protected]
> http://developer.marklogic.com/mailman/listinfo/general

_______________________________________________
General mailing list
[email protected]
http://developer.marklogic.com/mailman/listinfo/general

Reply via email to