After poking around, I've found a presentation by John Muellerleile at
Boundary's meetup on October 6, 2011. He addresses Riak, Solr and
other technologies in use at showyou.com.

http://blog.boundary.com/2011/10/11/Boundary-Tech-Talk-1.html
http://thinkdifferent.ly/stuff/scaling-at-showyou-jmm.pdf
http://twitter.com/jrecursive

Around 11:10 in the video, he addresses the Pre/Post Commit Hook
question. He uses a Post Commit Hook for the reason Ryan outlined
above... he doesn't want a write to be blocked. :)

Hope this helps others.

Thanks,
Simeon

On Wed, Dec 28, 2011 at 11:13 AM, Simeon F. Willbanks
<[email protected]> wrote:
> Ryan,
>
> I really appreciate the detailed answer and breakdown of Pre/Post
> Commit Hook options.
>
> Thanks,
> Simeon
>
> On Wed, Dec 28, 2011 at 10:42 AM, Ryan Zezeski <[email protected]> wrote:
>> Simeon,
>>
>> I think the assumption is that if it's a precommit hook you "know" that if
>> an object was written then it has also been indexed.  Whereas if it's a
>> postcommit the object may be written but the indexing fails.  I think the
>> guiding decision would probably be what you favor, 1) that for any object
>> it's index will exist first, therefore your query results are "consistent"
>> with your data or 2) that data persistence is the foremost priority and
>> inconsistent or eventually consistent query results are acceptable (because
>> maybe objects are constantly updated and eventually they will get indexed).
>>  The thing to keep in mind is that the precommit will block the write, and
>> if you're using a persistent backend like bitcask or level you may not want
>> your data sitting in memory waiting for the indexing to finish.  Another
>> thing to think about is using a postcommit that fires off into a queue that
>> is indexed thus reducing latency on your clients writes to Riak.  My point
>> is that there are many factors and if you are going to go down the Riak/Solr
>> route you should take your applications requirements into consideration.
>>
>> HTH,
>> -Ryan
>>
>>
>> On Wed, Dec 28, 2011 at 12:56 PM, Simeon F. Willbanks
>> <[email protected]> wrote:
>>>
>>> Ryan,
>>>
>>> Thanks for the timely answer.
>>>
>>> Yea, faceting in code doesn't sound like the right solution. :)
>>>
>>> In the talk 'Riak Search Explained,' Daniel Reverri described how Riak
>>> Search integrates with Riak KV via a Pre-Commit Hook. If you were to
>>> roll your own search (Solr Cluster), is the recommendation to initiate
>>> a indexing process from a Pre-Commit Hook as opposed to a Post-Commit
>>> Hook?
>>>
>>> Thanks again,
>>> Simeon
>>>
>>>
>>> On Wed, Dec 28, 2011 at 8:04 AM, Ryan Zezeski <[email protected]> wrote:
>>> > Simeon,
>>> >
>>> > There are currently no plans to implement faceted queries.  The first
>>> > thought is that you pull back all results and do the faceting yourself,
>>> > but
>>> > what's the point of that, right?  I think other than that some Erlang
>>> > would
>>> > need to be written for Search to properly support faceting.
>>> >
>>> > -Ryan
>>> >
>>> > On Tue, Dec 27, 2011 at 4:54 PM, Simeon F. Willbanks
>>> > <[email protected]> wrote:
>>> >>
>>> >> Are there plans to implement Faceted Queries?
>>> >>
>>> >> "Facet querying through the Solr interface is not yet supported."
>>> >>
>>> >>
>>> >> http://wiki.basho.com/Riak-Search---Querying.html#Faceted-Queries-via-the-Solr-Interface
>>> >>
>>> >> In the meantime, is there a recommended alternative way to build
>>> >> facets with Riak-Search?
>>> >>
>>> >> Thanks,
>>> >> Simeon
>>> >>
>>> >> _______________________________________________
>>> >> riak-users mailing list
>>> >> [email protected]
>>> >> http://lists.basho.com/mailman/listinfo/riak-users_lists.basho.com
>>> >
>>> >
>>
>>

_______________________________________________
riak-users mailing list
[email protected]
http://lists.basho.com/mailman/listinfo/riak-users_lists.basho.com

Reply via email to