The VisiBroker Location Service provides enhanced object discovery that enables you to find object instances based on particular attributes. Working with VisiBroker Smart Agents, the Location Service notifies you of what objects are presently accessible on the network, and where they reside. The Location Service is a VisiBroker extension to the CORBA specification and is only useful for finding objects implemented with VisiBroker. For more information on the Smart Agent (osagent), see “Using the Smart Agent”.
The Location Service is an extension to the CORBA specification that provides general-purpose facilities for locating object instances. The Location Service communicates directly with one Smart Agent which maintains a catalog, which contains the list of the instances it knows about. When queried by the Location Service, a Smart Agent forwards the query to the other Smart Agents, and aggregates their replies in the result it returns to the Location Service.
The Location Service knows about all object instances that are registered on a POA with the BY_INSTANCE Policy and objects that are registered as persistent on a BOA. The server containing these objects may be started manually or automatically by the OAD. For more information, see “Using POAs”, “Using the BOA with VisiBroker”, and “Using the Object Activation Daemon (OAD)”.
The Location Service can make use of the information the Smart Agent keeps about each object instance. For each object instance, the Location Service maintains information encapsulated in the structure ObjLocation::Desc shown below.
|
•
|
The object reference, ref, is a handle for invoking the object.
|
|
•
|
The iiop_locator interface provides access to the host name and the port of the instance's server. This information is only meaningful if the object is connected with IIOP, which is the only supported protocol. Host names are returned as strings in the instance description.
|
|
•
|
The repository_ id, which is the interface designation for the object instance that can be looked up in the Interface and Implementation Repositories. If an instance satisfies multiple interfaces, the catalog contains an entry for each interface, as if there were an instance for each interface.
|
|
•
|
The instance_name, which is the name given to the object by its server.
|
|
•
|
The activable flag, which differentiates between instances that can be activated by an OAD and instances that are started manually.
|
|
•
|
The agent_hostname, the name of the Smart Agent with which the instance is registered.
|
The Location Service is useful for purposes such as load balancing and monitoring. Suppose that replicas of an object are located on several hosts. You could deploy a bind interceptor that maintains a cache of the host names that offer a replica and each host's recent load average. The interceptor updates its cache by asking the Location Service for the hosts currently offering instances of the object, and then queries the hosts to obtain their load averages. The interceptor then returns an object reference for the replica on the host with the lightest load. For more information about writing interceptors, see “Using Portable Interceptors” and “Using VisiBroker Interceptors”.
The Location Service is accessible through the Agent interface. Methods for the Agent interface can be divided into two groups: those that query a Smart Agent for data describing instances and those that register and unregister triggers. Triggers provide a mechanism by which clients of the Location Service can be notified of changes to the availability of instances.
The Location Service agent is a collection of methods that enable you to discover objects on a network of Smart Agents. You can query based on the interface's repository ID, or based on a combination of the interface's repository ID and the instance name. Results of a query can be returned as either object references or more complete instance descriptions. An object reference is simply a handle to a specific instance of the object located by a Smart Agent. Instance descriptions contain the object reference, as well as the instance's interface name, instance name, host name and port number, and information about its state (for example, whether it is running or can be activated).
The locserv executable no longer exists since the service is now part of the core VisiBroker ORB.
The following sections explain how the methods provided by the Agent class can be used to query VisiBroker Smart Agents for information. Each of the query methods can raise the Fail exception, which provides a reason for the failure.
Using the HostnameSeq method, you can find out which servers are hosting VisiBroker Smart Agents. In the example shown in the figure below, this method would return the addresses (such as, IP address string) of two servers: Athena and Zeus.
You can query the VisiBroker Edition Smart Agents on a network to find out about all accessible interfaces. To do so, you can use the RepositoryIDSeq method. In the example shown in the following figure, this method would return the repository IDs of two interfaces: Car and Sedan.
In the example shown in the figure above, a call to either method with the request IDL:Automobile/Car:1.0 would return three instances of the Car interface: Tom's Car on Athena, Tom's Car on Zeus, and Keri's Car. The Tom's Car instance is returned twice because there are occurrences of it with two different Smart Agents.
In the example shown in the previous figure, a call to either method specifying the repository ID IDL:Automobile/Sedan:1.0 and instance name Tom's Car would return two instances because there are occurrences of it with two different Smart Agents.
A trigger is essentially a callback mechanism that lets you determine changes to the availability of a specified instance. It is an asynchronous alternative to polling an Agent, and is typically used to recover after the connection to an object has been lost. Whereas queries can be employed in many ways, triggers are special-purpose.
The trigger methods in the Agent class are described in the following tables:
Both of the Agent trigger methods can raise the Fail exception, which provides a reason for the failure.
The TriggerHandler interface consists of the methods described in the following tables:
|
This method is called by the Location Service when an instance matching the desc becomes accessible.
|
|
A TriggerHandler is a callback object. You implement a TriggerHandler by deriving from theTriggerHandlerPOA class (or the TriggerHandlerImpl class with BOA), and implementing its impl_is_ready() and impl_is_down() methods. To register a trigger with the Location Service, you use the reg_trigger() method in the Agent interface. This method requires that you provide a description of the instance you want to monitor, and the TriggerHandler object you want invoked when the availability of the instance changes. The instance description (TriggerDesc) can contain combinations of the following instance information: repository ID, instance name, and host name. The more instance information you provide, the more particular your specification of the instance.
If a field in the TriggerDesc is set to the empty string (“”), it is ignored. The default for each field value is the empty string.
For example, a TriggerDesc containing only a repository ID matches any instance of the interface. Looking back to our example in the figure above, a trigger for any instance of IDL:Automobile/Car:1.0 would occur when one of the following instances becomes available or unavailable: Tom's Car on Athena, Tom's Car on Zeus, or Keri's Car. Adding an instance name of “Tom's Car” to the TriggerDesc tightens the specification so that the trigger only occurs when the availability of one of the two “Tom's Car” instances changes. Finally, adding a host name of Athena refines the trigger further so that it only occurs when the instance Tom's Car on the Athena server becomes available or unavailable.
Triggers are “sticky.” A TriggerHandler is invoked every time an object satisfying the trigger description becomes accessible. You may only be interested in learning when the first instance becomes accessible. If this is the case, invoke the Agent's unreg_trigger() method to unregister the trigger after the first occurrence is found.
This section contains two examples of using the Location Service to find instances of an interface. The first example uses the Account interface shown in the following IDL excerpt:
The following code sample uses the all_instances() method to locate all instances of the Account interface. Notice that the Smart Agents are queried by passing “LocationService” to the ORB::resolve_initial_references() method, then narrowing the object returned by that method to an ObjLocation::Agent. Notice, as well, the format of the Account repository id: IDL:Bank/Account:1.0. To find all instances satisfying the AccountManager interface:
The following code sample shows how to find everything known to Smart Agents. It does this by invoking the all_repository_ids() method to obtain all known interfaces. Then it invokes the all_instances_descs() method for each interface to obtain the instance descriptions.
The following code sample implements and registers a TriggerHandler. The TriggerHandlerImpl's impl_is_ready() and impl_is_down() methods display the description of the instance that caused the trigger to be invoked, and optionally unregister itself.
If it unregisters itself, the method calls the CORBA::ORB::shutdown() method which directs the BOA to exit the main program's impl_is_ready() method so the program can terminate.
Notice that the TriggerHandlerImpl class keeps a copy of the desc and Agent parameters with which it was created. The unreg_trigger() method requires the desc parameter. The Agent parameter is duplicated in case the reference from the main program is released.