Skip to end of metadata
Go to start of metadata

You are viewing an old version of this page. View the current version.

Compare with Current View Page History

« Previous Version 7 Next »

Sakai Conditional Release Developer's Guide

Motivation

Design

The first design attempt produced a plan for a system that would encapsulate into a Rule arbitrarily many queries to arbitrarily many Sakai services. In this way, if a particular resource were protected by a Rule, accessing the resource could cause a request to the GradebookService, another request to ContentHostingService, and so on. This design was deemed too chatty for reasonable performance in the Resources tool, which must conform to many of the performance characteristics of a file system.

In the file-system mold, we decided to create for each resource an access control list (ACL), which indicates for each site member whether they have access to the resource. Doing a table lookup in this way solves the performance problem, as compared with the service-query approach in the first design. The challenge for the new design is knowing when to update the ACL. The solution is to respond to events which are fired throughout Sakai.

For example, we want to be able to restrict access to a particular folder to just those students who score above an 80 on a recent quiz. When the instructor grades her students, each assignment grade goes out into Sakai as a gradebook.updateItemScore event. The folder in question has an event-listener registered in its name. When all the events go out, the event listener examines each one for the specified condition (grade must be above 80). When a particular event (student's grade) meets the condition, that student's ID is added to the ACL for the folder under protection, and thereafter the student will be able to access it.

Sakai has a built-in mechanism for responding to events, called the NotificationService. Clients of the service ask it to create objects of type Notification. The Notification extends Sakai Entity and specifies three additional key pieces of information:

  1. one or more strings containing event types to respond to. These are called functions.
  2. a string called a resource filter which indicates which resource may be affected by the event.
  3. an object of type NotificationAction which has the notify() method for taking action in response to the event.

The conditions module contains an implementation of NotificationAction called ResourceReleaseRule, which also implements Predicate from Apache commons-collections. The purpose of Predicate is to add a boolean evaluate(Object arg) method which answers the question "Has the condition been met?" The notify() method checks the condition and then takes the necessary action to update the state of the ContentHostingService.

Walkthrough

We now have all the pieces to describe exactly how conditional release works. Since the application of conditions is asynchronous, there can be an arbitrary space of time between steps 2 and 3:

  1. Instructor edits details on an item in the Resources tool and is presented with "Only if the following conditions is met" option. The list of available assignments and their conditions is passed from the ConditionService with no direct dependency from the Resources tool and the GradebookService.
  2. Instructor saves the details, and within ResourcesAction a Predicate is created from the condition parameters and is added to a fresh instance of ResourceReleaseRule which in turn is added to a fresh instance of Notification, along with the event type (function) to respond to and the resource filter (the resource's ID). The NotificationService takes care of persisting the Notification.
  3. As events propagate through Sakai, BaseNotificationService examines each one for a matching function in its stored list of Notifications. If a stored Notification matches both the function and resource filter of the inbound event, its embedded NotificationAction is dereferenced and its notify() method is called.
  4. In the case of a ResourceReleaseRule, an argument necessary for evaluating the embedded Predicate will be pulled from the event and the embedded Predicate will be evaluated. In response to the evaluation, the access rules of the underlying content resource will either be modified or left alone.

BooleanExpression

The Predicate within the ResourceReleaseRule is delegated to another class called BooleanExpression. The BooleanExpression is how we programmatically encapsulate such English language questions as "Is the assignment score above 80?" We have a conventional boolean expression here, with a left term, an operator, and a right term. What makes this interesting is that at the time the instructor sets up the condition, a necessary piece of information to answer the question is missing. In our example, the missing piece is the left term, the assignment score in question. This data only materializes later, at the time the instructor submits one or more assignment grades.

At the time a BooleanExpression is evaluated, a data object is passed to it as an argument. This data object contains the missing data to fulfill the expression. At construction time, the BooleanExpression is supplied with the fully qualified name of the class it can expect to receive as an argument when it is evaluated, the name of a method to call on that object to retrieve the missing left term, the operator to use in the evaluation, and the value of the right-hand term of the expression.

For our example, the condition that "assignment score must be above 80" is saved in ResourcesAction.saveCondition() by creating a new BooleanExpression with these arguments:

eventDataClassName: "org.sakaiproject.conditions.impl.AssignmentGrading"
missingTermMethodName: "getScore"
operator: "greater_than"
argument: new Double(80)

When the grading event occurs, the ResourceReleaseRule creates a simple object representation of the grading event in an AssignmentGrading object, which becomes the argument passed to the BooleanExpression's evaluate(Object arg) method. The BooleanExpression uses reflection to invoke the getScore() method on the AssignmentGrading object. It then has everything it needs to evaluate the expression: ((AssignmentGrading)arg).getScore() > 80

Note that we also have conditions that don't require an operator or a missing term, such as "Has the assignment due date passed?" In this case, the arguments to construct a BooleanExpression are as follows:

eventDataClassName: "org.sakaiproject.conditions.impl.AssignmentUpdate"
missingTermMethodName: "dueDateHasPassed"
operator: "no_operator"
argument: null

Invoking the specified method on the inbound data object argument provides the answer we need.

Known Issues

Conditional Release was written for Sakai 2.4.x. It is unknown at this time how cleanly it will port to 2.5.0 and above.

Events are useful for getting information as it becomes available, but do not help you if you want to evaluate the state of the system long after the event has fired. For example, when an instructor first sets up a condition for a particular grade in a particular assignment, some students might already have a grade at that point. If we are waiting for an event to evaluate that student's access to the resource, the event will never come. Our solution to this is to produce our own "new condition" event that must then pull data from the GradebookService. This works, but is a little awkward and depends on the service API to expose the information we want.

Not all the events you might want to respond to are in Sakai already. For our baseline of functionality, we needed to create a new event in the GradebookService for individual assignments as they are graded, and another event when grades are submitted from an external tool such as Samigo or Assignments.

The ConditionService does not yet fulfill its true purpose, which is to publish a list of conditions from any Sakai service that registers with it. We did not need this capability for the functionality of our first iteration, so the ConditionService is only sending a static Map of strings for the Resources tool to embed directly into its template markup. To do this the "right way" is to devise a tree-shaped object graph for conditions, where each node is a criterion, and each path from the root of the tree to a leaf represents one completely-defined condition. For an example of an implementation in another system, see Apple's NSRuleEditor, part of their Cocoa application framework.

The hard part of implementing the true ConditionService is that you must translate the object graph of conditions into a JavaScript object so that the HTML form in your user interface can be dynamic. Furthermore, we must be able to implement the form in the tool author's view technology of choice (Velocity, JSF, RSF, etc.).

Sakai's BaseNotificationService has its limitations, notably that it requires any implementations of NotificationAction to be physically inside its component WEB-INF directory. Also, it uses manual manipulation of DOM objects for persistence, which is error-prone and yields object storage in the database as a lump of XML.

  • No labels