Wayne,
Thanks for this. It's a useful code pattern for this sort of thing and I
will probably use it for the specific requirement I have at the moment (I was
planning to do something similar anyway).
But this code, or any user-level code, does not fully implement the
uniqueness guarantee I'd like to have and that I think a specialized range
index could easily provide. This will work, but as you say it would be
necessary to always use this code convention. It would not prevent creation of
duplicate values by code that doesn't follow the convention. If uniqueness
were enforced by the index, then I could be confident that uniqueness is
absolutely guaranteed and I don't need to trust anyone (including my future
self) to always follow the same locking protocol.
---
Ron Hitchens {r...@overstory.co.uk} +44 7879 358212
On Jun 4, 2014, at 9:19 PM, Wayne Feick <wayne.fe...@marklogic.com> wrote:
> The simplest is to have the document URI correspond to the element value, and
> if you can use a random value it's good for concurrency.
>
> If you can't do that, but you want to ensure only one document can have a
> particular value for an element, I think it's pretty easy using
> xdmp:lock-for-update() on an URI that corresponds to the element value. You
> don't actually need to create a document at that URI, just use it to
> serialize transactions. Here's one way to do it.
> declare function lock-element-value($qn as xs:QName, $v as item)
> {
> xdmp:lock-for-update(
> "http://acme.com/";
> || xdmp:hash64(fn:namespace-uri-from-QName($qn))
> || "/"
> || xdmp:hash64(fn:localname-from-QName($qn)))
> };
> You'd then do something like the following.
> let $lock := lock-element-value($qn, $v)
> let $existing := cts:search(fn:collection(), cts:element-range-query($qn,
> "=", $v, "unfiltered"))
> return
> if (fn:exists($existing))
> then ... do whatever you need to do with the existing document
> else ... create a new document, safe from a race with another transaction
> You'd want to use lock-element-value() in any updates that could affect a
> change in the element value (insert, update, delete). I think you could get
> away with ignoring deletes since those would automatically serialize with any
> transaction that would modify the existing document.
>
> We use this sort of pattern internally to ensure uniqueness of IDs.
>
> Wayne.
>
>
> On 06/04/2014 12:49 PM, Whitby, Rob wrote:
>> I thought 2 simultaneous transactions would both get read locks on the uri,
>> then one would get a write lock and the other would fail and retry. Maybe
>> I'm missing something though.
>>
>> But anyway, I agree unique indexes would be a handy feature. e.g. our docs
>> have a DOI element which *should* be unique but occasionally aren't, would
>> be nice to enforce that rather than have to code defensively.
>>
>> Rob
>> ________________________________________
>> From: general-boun...@developer.marklogic.com
>> [general-boun...@developer.marklogic.com] on behalf of Ron Hitchens
>> [r...@ronsoft.com]
>> Sent: 04 June 2014 19:31
>> To: MarkLogic Developer Discussion
>> Subject: Re: [MarkLogic Dev General] New Feature Request: Unique Value Range
>> Indexes
>>
>> Rob,
>>
>> I believe there is a race condition here. A document may not exit as-of
>> the timestamp when this request starts running, but some other request could
>> create one while it's running. This request would then over-write that
>> document.
>>
>> I'm actually more concerned about element values inside documents than
>> generating unique document URIs. It's easy to generate document URIs with
>> 64-bit random numbers that are very unlikely to collide. But I want to
>> guarantee that some meaningful value inside a document is unique across all
>> documents.
>>
>> In my case, the naming space is actually quite small because I want the
>> IDs to be meaningful but unique. For example "images:cats:fluffy:XX.png",
>> where XX can increment or be set randomly until the ID is unique. One way
>> to check for uniqueness is to make the document URI from this ID, then test
>> for an existing document.
>>
>> But this doesn't solve the general problem. I could conceivably have
>> multiple elements in the document that I want to be unique. To check for
>> unique element values it's necessary to run a cts query against the
>> element(s). And I'm not sure if you can completely close the race window
>> between checking for an existing instance and inserting a new one if the
>> query comes back empty.
>>
>> Someone from ML pointed out privately that checking for uniqueness in the
>> index would require cross-cluster communication. I'm sure that's true, but
>> I'm also pretty sure that any user-level code solution is going to be far
>> less efficient. I'd be happy to pay that ingestion time penalty for the
>> guarantee that indexed element values are unique. At query time, such a
>> unique value index should perform like any other range index.
>>
>> ---
>> Ron Hitchens {r...@overstory.co.uk} +44 7879 358212
>>
>> On Jun 4, 2014, at 6:59 PM, "Whitby, Rob" <rob.whi...@springer.com> wrote:
>>
>>> How about something like this?
>>>
>>> declare function unique-uri() {
>>> let $uri := "/doc/" || xdmp:random() || ".xml"
>>> return if (fn:not(fn:doc-available($uri))) then $uri else unique-uri()
>>> };
>>>
>>> I guess because indexes are distributed across forests, ensuring uniqueness
>>> is not that easy?
>>>
>>> Rob
>>> ________________________________________
>>> From: general-boun...@developer.marklogic.com
>>> [general-boun...@developer.marklogic.com] on behalf of Ron Hitchens
>>> [r...@ronsoft.com]
>>> Sent: 04 June 2014 18:01
>>> To: MarkLogic Developer Discussion
>>> Subject: [MarkLogic Dev General] New Feature Request: Unique Value Range
>>> Indexes
>>>
>>> I'm working on a project, one aspect of which requires minting unique IDs
>>> and assuring that no two documents with the same ID wind up in the
>>> database. I know how to accomplish this using locks (I'm pretty sure) but
>>> any such implementation is awkward and prone to subtle edge case errors,
>>> and can be difficult to test.
>>>
>>> It seems to me that this is something that MarkLogic could do much more
>>> reliably and quickly than any user-level code. The thought that occurred
>>> to me is a variation on range indexes which only allow a single instance of
>>> any given value.
>>>
>>> Conventional range indexes work by creating term lists that look like
>>> this (see Jason Hunter's ML Architecture paper), where each term list
>>> contains an element (or attribute) value and a list of fragment IDs where
>>> that term exists.
>>>
>>> aardvark | 23, 135, 469, 611
>>> ant | 23, 469, 558, 611, 750
>>> baboon | 53, 97, 469, 621
>>> etc...
>>>
>>> By making a range index like this but which only allows a single fragment
>>> ID in the list, that would ensure that no two documents in the database
>>> contain a given element with the same value. That is, attempting to add a
>>> second document with the same element or attribute value would cause an
>>> exception. And being a range index, it would provide a fast lexicon of all
>>> the current unique values in the DB.
>>>
>>> Such an index would look something like this:
>>>
>>> abc3vk34 | 17
>>> bkx46lkd | 52
>>> bz1d34nm | 37
>>> etc...
>>>
>>> Usage could be something like this:
>>>
>>> declare function create-new-id-doc ($id-root as xs:string) as xs:string
>>> {
>>> try {
>>> let $id := $id-root || "-" || mylib:random-string(8)
>>> let $uri := "/idregistry/id-" || $id
>>> let $_ :=
>>> xdmp:document-insert ($uri,
>>> <registered-id>
>>> <id>{ $id }</id>
>>> <created>{ fn:current-dateTime() }</created>
>>> </registered-id>
>>> return $id
>>> } catch (e) {
>>> create-new-id-doc ($id-root)
>>> }
>>> };
>>>
>>> This doesn't require that I write any (possibly buggy) mutual exclusion
>>> code and I can be confident that once the xdmp:document-insert succeeds
>>> that the ID is unique in the database and that the type (as configured for
>>> the range index) is correct.
>>>
>>> Any love for Unique Value Range Indexes in the next version of MarkLogic?
>>>
>>> ---
>>> Ron Hitchens {r...@overstory.co.uk} +44 7879 358212
>>>
>>> _______________________________________________
>>> General mailing list
>>> General@developer.marklogic.com
>>> http://developer.marklogic.com/mailman/listinfo/general
>>> _______________________________________________
>>> General mailing list
>>> General@developer.marklogic.com
>>> http://developer.marklogic.com/mailman/listinfo/general
>> _______________________________________________
>> General mailing list
>> General@developer.marklogic.com
>> http://developer.marklogic.com/mailman/listinfo/general
>> _______________________________________________
>> General mailing list
>> General@developer.marklogic.com
>> http://developer.marklogic.com/mailman/listinfo/general
>
> --
> Wayne Feick
> Principal Engineer
> MarkLogic Corporation
> wayne.fe...@marklogic.com
> Phone: +1 650 655 2378
> www.marklogic.com
>
> This e-mail and any accompanying attachments are confidential. The
> information is intended solely for the use of the individual to whom it is
> addressed. Any review, disclosure, copying, distribution, or use of this
> e-mail communication by others is strictly prohibited. If you are not the
> intended recipient, please notify us immediately by returning this message to
> the sender and delete all copies. Thank you for your cooperation.
> _______________________________________________
> General mailing list
> General@developer.marklogic.com
> http://developer.marklogic.com/mailman/listinfo/general
_______________________________________________
General mailing list
General@developer.marklogic.com
http://developer.marklogic.com/mailman/listinfo/general
