Solr fitting in travel site context?

[Carsten Kraus <ca...@gmail.com>]

Hi all,

I'm about to develop a travel website and am wondering if Solr might fit to
be used as the search solution.
Being quite the opposite of a db guru and new to Solr, it's hard for me to
judge if for my use-case a relational db should be used in favor of Solr(or
similar indexing server). Maybe some of you guys would share their opinion
on this?

The products being searched for would be travel packages. That is: hotel
room + flight combined into one product.
I receive the products via a csv file, where each line defines a travel
package with concrete departure/return, accommodation and price data.

For example one csv row might represent:
Hotel Foo in Paris, flight departing 10/10/09 from London, ending 10/20/09,
mealplan Bar, pricing $300
..while another one might look like:
Hotel Foo in Paris, flight departing 10/10/09 from Amsterdam, ending
10/30/09, mealplan Eggs :), pricing $400

Now searches should show results in 2 steps: first step showing results
grouped by hotel(so no hotel appears twice) and second one all
date-airport-mealplan combinations for the hotel selected by the user in
step 1.

>From some first little tests, it seems to me as if I at least would need the
collapse patch(SOLR-236) to be used in step 1 above?!

What do you think? Does Solr fit into this scenario? Thoughts?

Sorry for the lengthy post & thanks a lot for any pointer!
Carsten

Re: Solr fitting in travel site context?

[Carsten Kraus <ca...@gmail.com>]

Cool, thanks a lot for sharing your experience and thoughts!
I will run a test like you suggested.

However, I've got some questions. The facet list I would retrieve for step 1
- it would 'only' contain the field values for what I faceted on such as
Hotel-ID, right?
How do I receive hotelname, description, price, etc. then? By issuing
further solr requests for each facetlist-item?

Would I be able to sort and paginate over this list?

You see, to end users I would need to show a hotel list sorted by e.g.
price(with only one search hit per hotel), showing the cheapest price
fitting to e.g. the user-specified date range for each hotel.

Would I still go the facet way with these requirements?

Thanks again!
Carsten


On Thu, Sep 10, 2009 at 9:07 AM, Constantijn Visinescu
<ba...@gmail.com>wrote:

> I'd run look into faceting and run a test.
>
> Create a schema, index the data and then run a query for *:* facteted by
> hotel to get a list of all the hotels you want followed by a query that
> returns all documents matching that hotel for your 2nd usecase.
>
> You're probably still going to want a SQL database to catch the
> reservations
> made tho.
>
> in my experience implementing Solr is more work then implementing a normal
> SQL database, and loosing the relational part of a relational database is
> something you have to wrap your head around to see how it affects your
> application.
>
> That said solr on my 4 year old single core laptop outperforms our new dual
> xeon database server running IBM DB2 when it comes to running a query on a
> 10 million record dataset and retuning the total amount of documents that
> match.
>
> Once you get it up and running properly and you need querys that are like
> "give me the total number of documents that match these criteria,
> optionally
> facted by this and that" it's amazingly fast.
>
> Note that this advantage only becomes apparent when dealing with large data
> sets. anything under a coulpe 100k records (guideline, depends heavily on
> the type of record) and a normal SQL server should also be able to give you
> the results you need near instantly.
>
> Hope this helps ;)
>
>
> On Wed, Sep 9, 2009 at 5:33 PM, Carsten Kraus <carsten.kraus@gmail.com
> >wrote:
>
> > Hi all,
> >
> > I'm about to develop a travel website and am wondering if Solr might fit
> to
> > be used as the search solution.
> > Being quite the opposite of a db guru and new to Solr, it's hard for me
> to
> > judge if for my use-case a relational db should be used in favor of
> Solr(or
> > similar indexing server). Maybe some of you guys would share their
> opinion
> > on this?
> >
> > The products being searched for would be travel packages. That is: hotel
> > room + flight combined into one product.
> > I receive the products via a csv file, where each line defines a travel
> > package with concrete departure/return, accommodation and price data.
> >
> > For example one csv row might represent:
> > Hotel Foo in Paris, flight departing 10/10/09 from London, ending
> 10/20/09,
> > mealplan Bar, pricing $300
> > ..while another one might look like:
> > Hotel Foo in Paris, flight departing 10/10/09 from Amsterdam, ending
> > 10/30/09, mealplan Eggs :), pricing $400
> >
> > Now searches should show results in 2 steps: first step showing results
> > grouped by hotel(so no hotel appears twice) and second one all
> > date-airport-mealplan combinations for the hotel selected by the user in
> > step 1.
> >
> > From some first little tests, it seems to me as if I at least would need
> > the
> > collapse patch(SOLR-236) to be used in step 1 above?!
> >
> > What do you think? Does Solr fit into this scenario? Thoughts?
> >
> > Sorry for the lengthy post & thanks a lot for any pointer!
> > Carsten
> >
>

Re: Solr fitting in travel site context?

[Constantijn Visinescu <ba...@gmail.com>]

I'd run look into faceting and run a test.

Create a schema, index the data and then run a query for *:* facteted by
hotel to get a list of all the hotels you want followed by a query that
returns all documents matching that hotel for your 2nd usecase.

You're probably still going to want a SQL database to catch the reservations
made tho.

in my experience implementing Solr is more work then implementing a normal
SQL database, and loosing the relational part of a relational database is
something you have to wrap your head around to see how it affects your
application.

That said solr on my 4 year old single core laptop outperforms our new dual
xeon database server running IBM DB2 when it comes to running a query on a
10 million record dataset and retuning the total amount of documents that
match.

Once you get it up and running properly and you need querys that are like
"give me the total number of documents that match these criteria, optionally
facted by this and that" it's amazingly fast.

Note that this advantage only becomes apparent when dealing with large data
sets. anything under a coulpe 100k records (guideline, depends heavily on
the type of record) and a normal SQL server should also be able to give you
the results you need near instantly.

Hope this helps ;)


On Wed, Sep 9, 2009 at 5:33 PM, Carsten Kraus <ca...@gmail.com>wrote:

> Hi all,
>
> I'm about to develop a travel website and am wondering if Solr might fit to
> be used as the search solution.
> Being quite the opposite of a db guru and new to Solr, it's hard for me to
> judge if for my use-case a relational db should be used in favor of Solr(or
> similar indexing server). Maybe some of you guys would share their opinion
> on this?
>
> The products being searched for would be travel packages. That is: hotel
> room + flight combined into one product.
> I receive the products via a csv file, where each line defines a travel
> package with concrete departure/return, accommodation and price data.
>
> For example one csv row might represent:
> Hotel Foo in Paris, flight departing 10/10/09 from London, ending 10/20/09,
> mealplan Bar, pricing $300
> ..while another one might look like:
> Hotel Foo in Paris, flight departing 10/10/09 from Amsterdam, ending
> 10/30/09, mealplan Eggs :), pricing $400
>
> Now searches should show results in 2 steps: first step showing results
> grouped by hotel(so no hotel appears twice) and second one all
> date-airport-mealplan combinations for the hotel selected by the user in
> step 1.
>
> From some first little tests, it seems to me as if I at least would need
> the
> collapse patch(SOLR-236) to be used in step 1 above?!
>
> What do you think? Does Solr fit into this scenario? Thoughts?
>
> Sorry for the lengthy post & thanks a lot for any pointer!
> Carsten
>