On Jun 20, 2007, at 12:30 PM, Manu George wrote:
I will download 1.0 and have a go at porting this, looks pretty cool
and we definitely should have it in 3.0 :).
Cool. Here is some info to arm you in how it worked. Knowledge
transfer commencing in 3.. 2.. 1....
Plain old stateless beans were used as the "servlets". To make a
bean that would show up in the Webadmin Console you simply had to
implement the HttpBean interface (i think it's now called
HttpListener) and give your bean a deploymentId following this format
"webadmin/{section}/{page}". Anyone could add to the Webadmin
console by doing both of these things, which is really cool as people
developing EJB apps can also deploy beans for administering those
apps right in the exact same jar. This is not only easy for
packaging but means new sections can be added/removed on the fly.
Using the described "webadmin/{section}/{page}" deploymentId format,
things end up automagically grouped in the JNDI tree. There's a
'webadmin' context we grab which will contain any number of "section"
contexts ("ClientTools", "EJBGenerator", etc.). Each of those
section subcontexts will contain several beans which we will use to
make the pages. Making the menu is pretty easy as we just iterate
over the webadmin section of the global jndi tree.
When an http request came in we just took the path part of the GET or
POST request, prepended "webadmin/" and then went looking for a bean
with that deployment id and invoked it via it's HttpBean (now called
HttpListener) interface passing in a HttpRequest and HttpResponse
objects which are trimmed down versions of similar servlet classes.
There'll be some changes to this as now we support our plain ejb
protocol over our http implementation, so the two will have to find a
way to share the URL space. See the openejb-http module.
To implement session state, we had a stateful session bean
implementing an HttpSession interface (again, similar to the servlet
equivalent) and simply wrote the internal ID of the bean instance
into a Cookie sent to the browser. For some reason we would write
the javax.ejb.Handle of the stateful bean's EJBObject to disk and
read it back out on subsequent requests then used it to get a
reference to the EJBObject again. I'm not sure why we didn't just
keep a static hashmap and put the EJBObject right in it using an ID
we could just make up like UUID (that would have been way simpler).
We had a standard superclass we favored for beans that implemented
the HttpBean (HttpListener) interface that did templating and the
aforementioned menu construction. The templating was perhaps too
tricky as we used a non-printable character to determine where to
insert data. Now we could just use swizzle-stream for some pretty
decent templating ability or even velocity. I'd be hesitant to
endorse velocity as we already have a dep on swizzle-stream and
wouldn't want to see us add another meg to our distro size if we can
avoid it -- we have like 3 times as many deps as 1.0 did and we
should probably start tightening the belt.
To serve static things like images, we had a "default" HttpBean which
searched for the items in the classpath and wrote them into the
response setting the mime type, etc. correctly. One thing that we
never did which still needs to happen is that the bean didn't have
the logic to set the modification information into the response so
the "If modified since" header which would allow the browser to rely
on it's cache instead of requesting the same images over and over again.
That pretty much covers the way it was put together.
As for the individual pages, I have to give a shout out to the people
who worked on them.
- The Jndi Viewer, Class Viewer, Ejb Viewer, and Object Invoker
were written by yours truly
- The EJB Generator was written by Jeremy Whitlock
- Everything else was written by Tim Urberg. Tim was "WebAdmin
guy" for a good long while. Before Tim came along the webadmin was
just some experimental code I had in a branch, he did more than he
realizes by putting his energy into it -- active people attract/
create more active people. Maybe we can convince him to come back
and work on it ;)
And of course I have to mention our own Paulo Lopes who wrote a
really cool project out in SF.net (http://beangen.sourceforge.net/)
which was the first plugin for the OpenEJB Webadmin. He wrote it
before we even had shipped a release containing the Webadmin or had
any docs at all on it, which in my mind shows just how neat the idea
of using ejb's and simple conventions to do the console really is.
Some notes going forward is that we have a truck load of meta-data
now available via SystemInstance.get().get
(OpenEjbConfiguration.class). Certainly JSR-77 is one option, but we
could do far better with plain old java code. That tree is the
primary source of meta-data for OpenEJB, it's what was used to
construct every container, bean, resource adapter, database connector
and *everything* in the system (well, sans the protocols). Someone
new to the project can look at it and understand it without having to
read any abstract specs. Something to consider. The tree is read
only in it's function, however it is possible to copy then edit and
make new containers, etc. based on existing definitions.
Additionally, using this same data structure it's possible to show
the potential services available via the service-jar.xml files in the
classpath that detail containers, resource adapters, database
connectors, etc. which can be configured/created at runtime. So we
could also display a sort of catalogue of components (aka. services)
that someone could click and deploy via the console.
I wrote a big long email about the service-jar.xml some time ago. I
think Karan knows which one I'm referring to as he had mentioned a
couple times he'd like to add it to the website -- assuming my memory
hasn't gone bad :)
-David
