Friday, December 28, 2012

Say Goodbye to HTTP URI Query Parameters

New Year is approaching fast but there is still some time to make one more New Year resolution.

Traditionally, when it comes to expressing the search requirements with HTTP URI, one uses URI query name and value, for example:

1. "/search/people?age=30&age=40"
2. "/search/people?ageFrom=30&ageTill=40"

First query can be read like this: "Find all people who are either 30 or 40 years old", the second - "Find all people older than 30 but younger than 40".

Plain query parameters are very widely used and actually very 'capable', one can invent a query parameter such as "ageFrom" to indicate the conditional requirement. It is not quite perfect in that the actual operator is "=" but the name implies it is not "equal to" but "greater or equal to", but it works, and probably reads better in some cases.

That said, the bigger the number of terms, the more tedious it can become to support creating the parameters like "ageFrom" and writing the actual code for figuring out which parameter means whether it is indeed the equality check as in "age=30" or say a "greater than" check as in "ageFrom=30". Supporting the finer-level checks such as "greater or equal to", as well as combining multiple type of checks can indeed become difficult to manage at the code level.

Consider always using FIQL expressions:

1. "/search/people?_s=age==30;age=40"
1.1 "/search/people?age==30;age=40"
1.2 "/search/people/age==30;age=40"

2. "/search/people?_s=age=ge=30;age=lt=40"
2.1 "/search/people?age=ge=30;age=lt=40"
2.2 "/search/people/age=ge=30;age=lt=40"

Does it look more complex ? I don't think it does but I can agree it may look a bit unusual at first, though the syntax is actually very easy to understand after typing it  just for a bit of time.

Note a number of possible variations - in fact I guess I like options 1.2 and 2.2 most, the "optionality" aspect of URI query components does not make the whole URI represent a specific application state very cleanly (options 1, 1.1, 2, 2.1).

Also note the fine-level checks in 2.1/2.2 which are tricky to express with the plain queries, example, in this case it is "Find all people 30 years old or older but younger than 40 (exclusive)".

If you think about it, you have nothing to lose by starting using FIQL, for many cases it is nearly as simple and primitive as using plain query name and value pairs but the bonus comes immediately once you start expressing something more involved. As noted at the beginning of this post, plain queries are also capable in this regard - but the more complex the search requirements are - the less easy it becomes working with the plain queries.

Those Apache CXF users who have started experimenting with FIQL can utilize the existing FIQL converters, which is another plus point, consider the task of writing the code for converting plain query parameters to JPA2 expressions - it can be an involved task indeed - but with CXF JPA2 converter it is a walk in the park; as a side note - JPA2 converter now supports "count" extensions, for example, "find all the people living in London with more than 6 children".

Well, you may not be convinced that it is time to drop plain query name and value pairs. Never mind, just get CXF converting them internally to FIQL (which is very easy to do for CXF given that FIQL is a richer language, and still rely on the handy FIQL converters of your choice to interact with the data stores and completely generalize the search processing logic along the way. Offer different search endpoints for your users to enjoy working with, one accepting the traditional queries and another one accepting FIQL queries, with both endpoints using the same FIQL converters.

Still plenty of options to innovate in this well-explored space :-)!

Update: See also this earlier post from Abhishek Jain.

Happy New Year !






Wednesday, December 12, 2012

FIQL and JPA2 Queries In Action

I've been focusing quite a lot recently on enhancing CXF Search extension module, by improving the existing converters and creating the new ones, making sure the parser is configurable, flexible and capable of mapping arbitrary property names to the properties of the bean capturing the search expression, and improving the documentation.

Andy Michalec created a FIQL parser quite a long time ago, it's been around for a while really, but it is only since Jeff Wang provided an initial FIQL to JPA2 converter patch that it kind of struck me how important it was to ensure the utility converters for mainstream and most popular technologies for querying the data stores were available. 

With all the documentation and code improvements, it is still not quite good till it is actually demonstrated somehow, and this is exactly what I've spent the last few days upon, on enhancing the existing Talend ESB jaxrs-advanced demo to show how the client can type the queries with SearchConditionBuilder without having to type FIQL expressions manually and get the expressions seamlessly converted to JPA2 TypedQuery and CriteriaQuery, with Tuple mixed in, and the matching data returned to the client, all with the help of FIQL to JPA2 converter.

Let me comment on the actual demo enhancements in more detail.

First check the client code, the "useSearchService" function, starting from the line 194 and more specifically from line 205. IMHO expressing the query with the help of the fluent query builder (also created by Andy) is quite cool, it is definitely more descriptive with respect to what is actually expected, and it is the specific query language agnostic, it is FIQL by default but can be something else.
The builder can build whatever advanced query expression is needed, though practically speaking I would not expect massive expressions being created.

Now lets move to the server code. SearchService is one which actually handles the client requests and all it does it delegates to PersonInfoStorage which deals with querying the data. Note, SearchService works with CXF SearchContext  by expecting it to extract the search expression from the current URI query component (default mode) or getting the expression from URI path and submitting it to SearchContext.

The main and in fact the single demo domain entity is Person but it is the one which is recursive, with children and parents, ancestors and descendants linked to. I haven't modified these relationships to get CXF JPA2 converter working properly, the only thing I did I added a couple of missing Person setters, added JPA2 annotations and also a hack required for JAXB RI capable of dealing with recursive structures (see the end of Person code).

The Person model is initialized in PersonInfoStorage init method (see the end of the file), where the injected JPA2 EntityManager is used - it could've been done elsewhere but was good enough for me for the purpose of the demo.

Next check getTypedQueryPerson (line 68) - see how straightforward it is  to get the FIQL expression transparently converted to JPA2 TypedQuery, the comments in the code should make it very easy to understand.

Note, Person is a recursive and possibly a very deep structure and one may not necessarily want to fetch all the Person representation back to the client. A number of options exists for dealing with this issue including using the intrusive JAXB XmlTransient annotation but using JPA2 Tuple is one of the most elegant ones.

Using Tuple is one of JPA2 options for having the response shaped into simpler or different structures and getCriteriaQueryPerson  method shows how it can be easily done by having the query returning the data sufficient for initializing a simple PersonInfo representation. One needs to get OpenJPA auto-generating the
metamodel classes for the tuple selections working and you can see how it can be done here, this tip helped me a lot. 

Also note that PersonInfoStorage has the injected bean properties passed to SearchContext. Why this may have to be done is explained here but in short it lets to get the property names used in query language completely decoupled from the corresponding properties of the capturing bean, for example, in the demo case, the client can type "builder.is('childName').equalTo('Fred')" and have it working with the 'childName'  correctly mapped to "children.name", where "children" points to a Person collection of children. This makes it simpler and easier to work with for the client, while keeping the internals of how the properties are actually linked to each other completely opaque. That is cool, I can hear you saying :-)

Finally have a look at how JPA2 EntityManager is initialized and injected. I started with arguably a simpler approach, I basically used Spring ORM module to get the entity manager wired in, see this beans.xml.  This actually back-fired on me when I tried to make the demo working in Karaf - Spring ORM needs to see persistence.xml and is not capable on its own of inspecting OSGI Meta-Persistence property set up in the common module's pom.xml so I just ended up duplicating persistence.xml in both common and service modules.

On the TODO list is to follow an excellent tutorial from Christian and make the demo working with OSGi JPA Service. Perhaps someone from the community will be interested in doing a related pull request ?
  
So this is it and hope you'll find something interesting in this demo, enjoy !