Friday, July 27, 2012

Jettison 1.3.2 is out

Jettison 1.3.2 has been released this week, please check the Download page.

Those who try to customize the way Jettison works should find it easier to override various Jettison classes, for example, in CXF I've been able to remove about 50 lines of code I had to copy earlier on to get large Jettison sequences optionally restricted.

Jettison will no longer require a namespace map set up for the serialization to work, in cases when it is not configured to ignore namespaces, as I moved a fix provided by Benson from CXF to Jettison.

A number of minor performance enhancements have also been implemented thanks to the proposals from Fabian Lange.

During the next  few interactions we will continue minimizing a number of outstanding issues, however I'd also like to encourage the community to provide patches - they will get an immediate attention during the release, as it's been the case during the last few releases. I haven't resolved one issue with the available patch to do with treating XML attributes as elements, in CXF we can manage it by providing a custom XML writer, so I needed more time on investigating if that issue can be resolved at the current Jettison implementation level effectively or not.

At some point in the future an optional streaming support will have to be introduced, either on read or write sides, even if certain restrictions will have to be there.

In meantime, please keep stressing Jettison and provide the patches :-) 

Thursday, July 19, 2012

CXF Log Browser Demo

A Log Browser demo has been available in the CXF distributions  for more than a year now. This demo is based on the brilliant contribution from Thomasz Opanovicz done as part of his GSOC project.

What I would like to do is explain what exactly the CXF Log Browser can do right now, and suggest some ideas on how it can be enhanced.

At the moment, the browser can be used to poll the Atom-enabled management endpoints and display the available log entries, this can be done per every CXF service, JAX-WS or JAX-RS one. The Atom endpoints can be set up as described in this section and shown in the demo. Have a look at this image on how the browser can show the log entries, I think it is quite cool.  Note the entries are paged, example, one can go to the next, previous, first or last pages, from the current page.

At the moment the browser polls every time the "refresh" button is pressed - this is one minor thing that can be customized further, example, a browser can be asked to poll every 60 seconds.

The browser offers an option to restrict which log entries are displayed by letting users to set FIQL search queries.

Typically, the Atom endpoint (which handles the browser queries) captures the log events, transforms them to Atom feeds and keeps them in memory or overflows to a registered ReadWriteLogStorage if needed but what is interesting it can also be configured from the start  to read the log entries from the existing log files.

For example, it can be configured to check the directory where Karaf gets the log files created. CXF offers a file-based ReadableLogStorage interface implementation  which can check the existing log files matching a given naming pattern, figure out which log file is newer or older and get the log entries in whatever format is used to format them using the simple configuration involving no regular expressions at all. For example, have a look at this test which works with the log files available here as well as this configuration file (check the bean with id "storage3").

All of it needs to be more extensively tested but Log Browser is already offering a fairly involved support for working with the available log entries.

However, quite a few enhancements can be done which can make it more useful and eventually turn the browser into a management console of its own.

One obvious thing which is missing is the ability to capture events immediately, with a browser acting as a receiver. For example, it would be good to get Atom push endpoints sending the events to it too. Implementing an enhancement like this one would very likely lead to Web Sockets supported in CXF. The browser would be configured to support either push or pull style on the startup or perhaps it would keep the pull  style by default and get the real-time events displayed in the running line at the bottom of the window.

The browser can also offer an option to get the log files from the selected location downloaded to the local disk. 

We can also have at least two more tabs available. One would show the JMX statistics related to CXF, another one would show the message exchange details.

I'd like to encourage users to try the demo, and consider helping us with enhancing the browser further.  

Advanced queries involving multiple entities

As I've mentioned a number of times, FIQL can help with expressing the advanced search conditions in a compact and easy to understand syntax.

The queries like "find all the books published before a given date" are very easy to type in FIQL and extending this query with a restriction like "and having the page count between 80 and 100 pages or less than 20" is quite straightforward too, manually, or with the help of the client FIQL builder.

However, what if one would like to do a query involving multiple entities, for example, "find the library in a preferred location where all the books published before a given date and written by more than 2 authors are available" ? Or imagine or more interesting query if you'd like.

This bit, "where all the books published before a given date and written by more than 2 authors" is an easy one to support with the CXF search extension. The question is, how to structure the advanced query such that Book entities matching the arbitrary complex query are selected first and then a Library entity is checked per every Book on whether it matches its own selection criteria or not.

Another question is how to express the requirement that it is actually "the list of books",  "published before a given date and written by more than 2 authors and available in one of the libraries in a preferred location", that has to be returned to a user.

As you can see, there are at least two main requirements to deal with. First one is how to let users choose which entity (among a number of entities being matched) is actually returned which affects the response representation. Second one is how to do an advanced query involving multiple entities at the level of the JAX-RS application handling such a query.

The good news is that the first requirement can be easily managed with the help of URI path and/or query components. For example, an expression like "/books[date=le=2012-02-11]/library" can be used to find all the libraries having the matching books, and the one like "/library(dist=lt=10;dist=gt=5)/books(date=le=2012-02-11)" can get all the books matching a given search criteria available in a library within a 10 km range from some well-known location (such as the city center) but not closer than 5 (given that the parking is not free within a 5 km range).

It is really up to the designer of the service how to get the actual expressions captured, which characters to use to mark them, example "[" and "]", "(" and ")" or ";" and how to support the selective retrieval of the entities involved in the multi-entity search. See this section for more examples.

Yet another good news is that a complex multi-entity query can be managed pretty easily at the JAX-RS resource level, see this section for a number of options on how it might be done. Note how multiple FIQL expressions can be handled.

What is mentioned there is that it can be more optimal to get say JPA to execute a JOIN like query at a database level in one go as opposed to getting it to find a list of entities matching the first FIQL expression, then doing the follow-up in-memory match against the selected entities.

Right now CXF can not automate the process of creating a JPA TypedQuery or SQL expression which can 'span' the multiple entities - this is the next task, and in meantime please experiment with the demonstrated approach and see how it fares against the one involving the composite JOIN-like queries. Alternatively, introspect the captured search conditions as shown in this section and build a composite query in the code.

Please also have a look at the OData4J project and check how the advanced queries can be managed there. 

Finally, I'd like to answer on the following question: why one would worry about offering such a search interface to users, would it not be simpler to offer a Google-like search interface where one enters a few words and gets the list of matching data pages ? I think that when one creates a service for exposing the data with the known properties and relationships, the opportunity is there to offer an optimized search engine for users to get the customized search experience.

Tuesday, July 10, 2012

JMS Transport support for CXF JAX-RS clients

I blogged about the support for JMS by CXF JAX-RS endpoints two years ago.

The main reason behind making the JAX-RS frontend (associated by most users with supporting HTTP-based communications) JMS-aware was to do with getting the most from the 'investment' made into implementing the RESTful services on top of CXF JAX-RS.

If one has the resource code relying on the JAX-RS runtime to make the inbound data delivered to the right method and easily consumable in the form of a given JAXB bean instance, then the possibility is that this code can work equally well when the data comes to this resource handler either via HTTP or JMS or indeed some other transport, example, CXF Local Transport.    

CXF is super-flexible in the way it can support multiple transports, and it does make sense to get JAX-RS based endpoints optionally supporting non-HTTP transports.

And it is exactly for the same reason that CXF JAX-RS proxies (and I have to admit, WebClient :-) - simply because proxies and WebClient are using the same base code) also optionally support JMS transport now, thanks to the contribution from Willem.

So why would one want to use CXF JAX-RS client code with JMS ? As noted above, it is primarily about making the same code, client one in this case, re-usable in different contexts. I could've noted again that REST principles can probably be applied even to RS-232 endpoints if really really needed :-) but I'd rather focus on the re-usability aspect.

For example, consider a newly introduced HTTPSPStateManager implementation for making it easy for users to quickly set up a centralized distributed SSO cache endpoint. You've tested it and seen it working with HTTP. Now the time has come up for you to move to a more sophisticated distributed cache implementation for it to scale really well in the production. But then you realize that by simply getting HTTPSPStateManager proxies and the endpoint addresses use  a JMS scheme you can get an entirely different cache support, still using the same configuration, so you are going to give it a go and see what happens.

Or perhaps you have a Camel JAX-RS endpoint, supported by Camel transport and jaxrs:endpoint or CXFRS bean or endpoint, supporting both HTTP and JMS. It would be cool to get the same Camel CXFRS client to be able to talk to the endpoints using either HTTP or JMS.

I'm hoping users can come up with more interesting cases which will be possible thanks to the optional end-to-end support for JMS in the JAX-RS frontend.


How to test CXF JAX-RS endpoints

Users have been asking during the last couple of years how to test CXF JAX-RS endpoints. One of the users from the CXF community would always point to either a blog entry or paste a code example showing how the endpoints can be tested easily enough.

The problem has been all the time that there was no any documentation on the CXF JAX-RS wiki specifically describing the steps required to get the tests set-up and running.

A new wiki page has been added recently and it documents how the endpoints and indeed the consumers can be tested easily by using either the embedded Jetty or the newly added support for CXF Local Transport. The latter option is quite cool is that it does let to test the complete end-to-end invocation without spending the time on setting up HTTP mocks, by only having an address value to include a "local:" URI scheme.

This wiki page will be expanded to show how the tests can be run in the different test environments. Let us know please if you have some material ready to be contributed to the wiki :-)