Tuesday, 17 May 2011

JDev 11g, Task Flows & ADF BC – one root Application Module to rule them all?


In my previous blog post I discussed the power of the ADF task flow functionality, and the devil in the detail for uninitiated developers using the transaction and data control scope options. This post will extend the discussion on task flows and the ADF Controller's interaction with ADF Business Components, in order to show another scenario where programmers must understand the underlying behaviour of the framework.

Developers who have worked with the ADF framework for some time, especially from the JDeveloper 10g edition and earlier, will likely have stumbled across the concepts of root and nested Application Modules (AM) in the ADF Business Component (ADF BC) layer. The Fusion Guide has the following to say on the two types of AMs:
Application modules support the ability to create software components that mimic the modularity of your use cases, for which your higher-level functions might reuse a "subfunction" that is common to several business work flows. You can implement this modularity by defining composite application modules that you assemble using instances of other application modules. This task is referred to as "application module nesting". That is, an application module can contain (logically) one or more other application modules, as well as view objects. The outermost containing application module is referred to as the "root application module".
At runtime, your application works with a "main" — or what's known as a "root "— application module. Any application module can be used as a root application module; however, in practice the application modules that are used as root application modules are the ones that map to more complex end-user use cases, assuming you're not just building a straightforward CRUD application. When a root application module contains other nested application modules, they all participate in the root application module's transaction and share the same database connection and a single set of entity caches. This sharing is handled for you automatically by the root application module and its "Transaction" object.
The inference is that if for a single user you want to support more than one transaction at a time, you must create 2 or more root AMs. However the implication of this is if you do so, as a transaction and database connection have a 1 to 1 relationship, a user exercising more than one root AM in the ADF BC layer during their session will take out the same amount of connections with the database. This implication further effects scalability, as the more connections a user takes out from the database, and more connections in our midtier pool will be tied up by each user too.

Task Flow transaction options

The transaction and data control scope behavioural options available to bounded task flows provide a sophisticated set of functionality for spawning and managing one or more transactions during an ADF user's session, an extension of the facilities provided by the ADF BC Application Module. Straight from the Fusion Developer's Guide the task flow transaction options are:

• <No Controller Transaction>: The called bounded task flow does not participate in any transaction management.

• Always Use Existing Transaction: When called, the bounded task flow participates in an existing transaction already in progress.

• Use Existing Transaction If Possible: When called, the bounded task flow either participates in an existing transaction if one exists, or starts a new transaction upon entry of the bounded task flow if one doesn't exist.

• Always Begin New Transaction: A new transaction starts when the bounded task flow is entered, regardless of whether or not a transaction is in progress. The new transaction completes when the bounded task flow exits.

Ignoring the "No Controller Transaction" option which defaults back to the letting the ADF BC layer manage its own transactions, the other task flow transaction options allow the developer to create and reuse transactions at a higher level of abstraction for relating web pages, page fragments and other task flow activities. As such the root AM doesn't constrain when the application establishes a new transaction and connection to the database.

Yet if now we've the option for spawning more transactions in our application, what's the implication on scalability and the number of connections that are taken out in the midtier and database for each user? Thanks to a previous OTN forum post and the kind assistance of Steve Muench this post will demonstrate how the ADF framework as a whole attempts to minimize this issue.

The "No Controller Transaction" option

Before investigating how the ADF framework addresses the scalability issue, it's useful to run through an example of using the "No Controller Transaction" bounded task flow option. This will demonstrate how the framework establishes database connections when defaulting back to the ADF BC layer's own transaction and database connection functionality.

As explained by Frank Nimphius in the following OTN Forum post the No Controller Transaction option infers that task flows are taking no control of the transactions established during the user's session when the task flow is called, all such functionality is delegated back to the underlying business services. In context of this blog post the business service layer is ADF Business Components.

As seen in the following picture overall the ADF BC layer for our example is application is a fairly simple one:

As can be seen the application is comprised of two ADF BC Application Modules (AM), named Root1AppModule & Root2AppModule. Both expose the same View Object (VO) OrganisationsView as two separate usages. In the following picture you see Root1AppModule exposes OrganisationsView as OrganisationsView1:

...and in the following picture Root2AppModule exposes OrganisationsView2 off the same OrganisationsView VO:

For what it's worth to the discussion, OrganisationsView is based on the same Organisations EO:

However as there are 2 root AMs at play in our example application, each AM will instantiate its own OrganisationsView (namely OrganisationsView1 and OrganisationsView2 respectively) and Organisation EO cache, implying the record sets in the midtier are distinctly different even though we're using the same underlying design time constructs.

Like that explained in the previous blog post, how do we know when an AM actually creates a connection? Without knowing this, in our trials with the transaction options supported by Bounded Task Flows, unless the ADFc explicitly throws an error, we'll have trouble discerning what the ADF BC layer is actually doing underneath the task flow transaction options.

While external tools like the Fusion Middleware Control will give you a good insight into this, the easiest mechanism is to extend each root Application Module's ApplicationModuleImpl's class with our own AppModuleImpl and override the create() and prepareSession() methods. The following code shows an example for the Root1AppModuleImpl:
public class Root1AppModuleImpl extends ApplicationModuleImpl {
// Other generated methods

protected void create() {
if (isRoot())
System.out.println("########Root1AppModuleImpl.create() called. AM isRoot() = true");
System.out.println("########Root1AppModuleImpl.create() called. AM isRoot() = false");

protected void prepareSession(Session session) {
if (isRoot())
System.out.println("########Root1AppModuleImpl.prepareSession() called. AM isRoot() = true");
System.out.println("########Root1AppModuleImpl.prepareSession() called. AM isRoot() = false");
Pretty much the code for the Root2AppModuleImpl is the same, except the name of the classes change in the System.out.println calls.

Overriding the create() method allows us to see when the Application Module is not just instantiated, but ready to be used. This doesn't tell us when a transaction and connection is established with the database, but, is useful in identifying situations where the framework creates a root or nested AM.

The prepareSession() method is a chokepoint method the framework uses to set database session state when a connection is established with the database. As such overriding this method allows us to see when the AM does establish a new connection and transaction.

Once we've setup our ADF BC Model project, we'll now create two Bounded Task Flows (BTFs) to allow the user to work with the underlying AMs and VOs.

Both our task flows are comprised of only 2 activities, a page to view the underlying the Organisations data and an Exit Task Flow Activity. Root1AppModuleTaskFlow contains the following activities:

...and Root2AppModuleTaskFlow is virtually identical:

The OrganisationsTable1.jspx and OrganisationsTable2.jspx pages in each task flow show an editable table of Organisations data, visually there's no difference between them:

While the pages don't differ visually, underneath their pageDef files are completely different as they source their data from different root Application Modules and View Objects. In the following picture we can see the OrganisationsTable1PageDef.xml file makes use of the ADF BC OrganisationsView1 VO mapping to the Root1AppModuleDataControl:

And in the following picture we can see the OrganisationsTable2PageDef.xml file uses the ADF BC OrganisationsView2 VO mapping to the Root2AppModuleDataControl:

Finally we can see the No Controller Transaction option set for the first task flow:

..and the second:

Note I've deliberately set the data control scope to Shared to highlight a point in the future.

At this point we'll include a Start.jspx page in our Unbounded Task Flow (UTF) adfc-config.xml file with navigation rules to call each BTF, and navigation rules to return on the Exist Task Flow Return activity being called inside of each BTF:

On running our application starting with the Start.jspx page we see:

At this point inspecting the console output for the application in the JDeveloper log window, we don't see the System.out.println messages:

Returning to the Start page and selecting the Go Root1AppModuleTaskFlow button we then see:

Note in the above picture I've deliberately selected the 5th recorded and changed the Organisation's name to uppercase. Checking the log window we now see:

As such we can see that the Root1AppModule in the ADF BC layer has been instantiated, and has established a connection with the database, also establishing a transaction.

If we now return via the Exit button to the Start.jspx page in the UTF, then select the Go Root2AppModuleTaskFlow button we see:

Note that the 5th record hasn't been updated to capitals, and indeed the 1st record is the default selected record, not the 5th, implying that we're not seeing the same cache of data from the midtier, essentially 2 separate transactions with 2 separate Organisations EO caches, and in turn two separate VOs with their own current row record indicators. This last point shows that the Shared data control scope we set in the task flows has no use when the No Controller Transaction option is used.

Returning to the JDev log window we see that the second Root2AppModule was instantiated and has separately established its own connection and therefore transaction:

For completeness if we select the Exit button and return to the Start.jspx page, then return to the first task flow we can see the 5th record is still selected and updated:

The key point from this example is the fact that each Application Module is instantiated as a root AM and also creates its own connection.

A chained "No Controller Transaction" task flow example

With the previous example in hand, and building upto our final example, in this section I'd like to rearrange our example application such that rather than calling the two task flows separately, we'll instead chain them together, the 1st calling the 2nd.

The substantial change to the setup is that of the Root1AppModuleTaskFlow, where we now include a Task Flow Call to the Root2AppModuleTaskFlow, and a small modification to the OrganisationsTable1.jspx page to include a button to navigate to the Root2AppModuleTaskFlow:

In running the application, we first land in the Start.jspx page:

At this stage the log window shows no root AMs have been created:

On selecting the Go Root1AppModuleTaskFlow button we arrive at the OrganisationsTable1.jspx page:

The log window shows the Root1AppModule has been instantiated and a new connection/transaction created:

This time we'll select and change the 6th record to all caps:

Now we'll select the Go Root2AppModuleTaskFlow button that's located in the OrganisationsTable1.jspx page. On navigating to the new page we see:

As can be seen, the current row indicator lies on the first row and the 6th row hasn't been modified, exactly the same as the last example. In addition in the JDev log window we can see the establishment of the new root AM and connection/transaction:

The conclusion at this point of the blog post is chaining task flows when the No Controller Transaction options are used has no effect on the root AMs and the establishment of connections/transaction.

Always Begin New Transaction and Always Use Existing transaction example

With the previous 2 examples, we can see with the No Controller Transaction option in use for our bounded task flows, the design of our root Application Modules definitely has an effect on number of transactions and therefore connections taken out from the database.

With the following third and final example, we'll introduce BTF transaction options that relegate the transactional behaviour of the underlying ADF BC business services to the task flows themselves.

Taking our original Root1AppModuleTaskFlow, we'll now change its transaction options to use Always Begin New Transaction and an Isolated data control scope:

For our second task flow Root2AppModuleTaskFlow, which we intend to call from Root1AppModuleTaskFlow, we'll change its transaction options to use Always Use Existing Transaction. The IDE in this case enforces a Shared data control scope:

From the task flow transaction options described at the beginning of this post via the Fusion Dev Guide, the options we've selected here should mean the Root1AppModuleTaskFlow establishes the transaction and therefore connection, and the second Root2AppModuleTaskFlow should borrow/attach itself to the same transaction and connection. However from our previous examples with the No Controller Transaction option, surely the root Application Modules will enforce they both create their own transaction/connections?

Running this application we first hit out Start.jspx page:

And our log window shows no AMs have been created or established connections at this stage:

On navigating to Root1AppModuleTaskFlow via the associated button in the Start.jspx page we see very similar to previously:

Yet our log window shows something interesting:

We can see that the Root1AppModule has been created as a root AM, then it's established a connection via the prepareSession() method. Oddly however we can see a further Root1AppModule has been created? Even more oddly this secondary AM is not a root AM, but a nested AM? By deduction as there are no other log entries, this second instance of Root1AppModule must be nested under the root Root1AppModule? Interesting, but let's continue with the example.

Now that we've entered the Root1AppModuleTaskFlow, let's modify the 7th record:

Followed by selecting the button to navigate to the Root2AppModuleTaskFlow, we then see on visiting the OrganisationsTable2.jspx page of the second task flow:

Note the 7th record's data, and where the current row indicator is located! Before drawing conclusions, let's look at the log window's output:

From the log window we can see a 3rd AM has been instantiated, this time a Root2AppModule as a nested AM.

What's going on?

Without a key bit of information from Steve Muench's reply to my previous OTN Forum's post, it may be hard to determine the BTF behaviour in the last scenario. In context of ADF BC and BTFs given the right combination of transaction and data control scope options, the framework will automatically nest your AMs regardless if they're a root AM.

So when we're running our application, on calling the first task flow, we see:

a) A single root AM created based on Root1AppModule, as the framework needs at least a signle root AM to drive the transactions and connections. Let's refer to this as Root1AppModule1.

b) A second instance of Root1AppModule as a nested AM under its own root instance. Let's refer to this as Root1AppModule2. This is a little odd, but as the framework automatically nests the AMs of the called BTF under a root AM instance, it's using the first AM encountered for dual purposes, essentially nesting Root1AppModule2 under Root1AppModule1.

c) By the time we hit our second BTF, only one instance of Root2AppModule is created, nested under Root1AppModule1. Let's refer to this as Root2AppModule1.

To summarize at this point, our AM structure at runtime is:

(Root) Root1AppModule1
(Nested Level 1) Root1AppModule2
(Nested Level 1) Root2AppModule1

Given that we now know nesting is occurring, it does explain why we're seeing the changes to the 7th record in both BTF OrganisationTable pages. Only if separate root AMs were used would there be separate EO caches for the Organisations table. However as the BTFs have nested the AMs, they're essentially sharing the same EO cache. So a change to data in one BTF will be reflected in the second, as long as they share transactions and are ultimately based on the same EO.

The final mystery is why if we can see the data changes are reflected in each BTF, how come the current row indicators of the underlying VOs are out of sync? This is simply explained by the fact that with the two different VOs defined in each AM, namely OrganisationsView1 & OrganisationsView2, at runtime essentially they are both instantiated separately, even though ultimately they exist under the same Root1AppModule1 at runtime and share the same EO cache. As such both VOs maintain their own separate row currency, and other state carried by the ADF BC View Objects.

Changing AM connection configurations

At this point we may start to ask how the framework is deciding to nest the AMs under the one parent AM? Is the underlying framework being smart enough to look at the database connections for the separately defined root AMs, and in determining they're one and the same, nesting the AMs as it doesn't matter when they both connect to the same database?

From my testing it appears to make no difference if you change the connections. It's essentially the first AM whose connection details are used, and it's just assumed the second AM uses the same connection details. The further inference of this is it's just assumed that all the database objects required by the second root AM database connection are available to the first.

This has two knock on effects:

1) If both AMs did connect to the same database schema, if the first connection is missing privileges on objects required by the second AM's objects, you'll hit database privilege errors during runtime when those second AM objects interact with the database (e.q. at query and DML time).

2) If you actually use your root AMs to connect to entirely different data sources, such as different databases, this automatic nesting will cause your application to fail.

The first here is easily solved with some database privilege fixes. For the second, this implies you should move back to the No Controller Transaction options to return to an application where the root AMs use their own connections.

Some caveats

If the automatic nesting functionality proves useful to you, be warned of one caveat. From the OTN Forum replies from Steve Muench, even though he describes the fact the framework will automatically nest your AMs for the purposes of connections/transactions, don't assume this means other AM features are joined or always will display the same behaviour. To quote:
In fact, in a future release we will likely be changing that implementation detail so that the AMs are always used from their own AM pool, however they will share a transaction/connection.
A second caveat is around the mixing of BTF transaction options. Notably the No Controller Transaction option because of its inherit different use of root AMs to that of the normal transaction options would imply that mixing No Controller Transaction with other BTFs not using this option could leave to some disturbing and unpredictable behaviour around transactions and connections. Either use No Controller Transaction exclusively, or not at all.

A third and final caveat is this and the previous post that describe the transaction behaviours is in context of the interaction with ADF Business Components. Readers should not assume that the same transaction behaviour will be exhibited by different underlying business services such as EJBs, POJOs or Web Services. As example Web Services don't have the concept of transactions, so we can probably guess that there's no point using anything but the No Controller Transaction option .... however again you need to experiment with these alternatives yourself, don't base your conclusions on this post.

The implication for ADF Libraries

In Andrejus Baranovski's blog post he shows a technique for mashing the Model projects of several ADF Libraries containing BTFs and their own AMs into a single Model project, such that a single root AM is defined across all BTFs at design time, to be reused by all ADF Libraries. With this blog post we see we while Andrejus's technique is fine for those relying on the No Controller Transaction options of BTFs, such an extended build of ADF Libraries and their AM Model projects is not essential to minimize the number of connections of our application.


This blog post and the previous have taken a lot of research and testing to arrive at it's conclusions. If future readers find anything wrong, or find alternative scenarios where what's written here doesn't pan out, it would be greatly appreciated if you could leave a comment to that effect, such that this post doesn't mislead future readers.

No comments: