Before we talk about how to create and destroy objects, we need to look at a more basic concept, namely that of object existence. What does it mean for an object to "exist" and, more fundamentally, what do we mean by the term "object"?
As mentioned in Section 2.2.2, an
Ice object is a conceptual entity, or abstraction that does not really exist. On the client side, the concrete representation of an Ice object is a proxy and, on the server side, the concrete representation of an Ice object is a servant. Proxies and servants are the concrete programming-language artifacts that represent Ice objects.
Because Ice objects are abstract, conceptual entities, they are invisible to the Ice run time and to the application code—only proxies and servants are real and visible. It follows that, to determine whether an
Ice object exists, any determination must rely on proxies and servants, because they are the only tangible entities in the system.
This may seem self-evident but, on closer examination, is a little more subtle than you might expect. In particular, Ice object existence has meaning only
within the context of a particular invocation. If that invocation raises
ObjectNotExistException, the object is known to not exist. Note that this says nothing about whether concurrent or future requests to that object will also raise
ObjectNotExistException—they may or may not, depending on the semantics that are implemented by the application.
Also note that, because all the Ice run time knows about are servants, an ObjectNotExistException really indicates that a
servant for the request could not be found at the time the request was made. This means that, ultimately, it is the application that attaches the meaning "the Ice object does not exist" to this exception.
In theory, the application can attach any meaning it likes to ObjectNotExistException and a server can throw this exception for whatever reason it sees fit; in practice, however, we recommend that you do not do this because it breaks with existing convention and is potentially confusing. You should reserve this exception for its intended meaning and not abuse it for other purposes.
The preceding definition does not say anything about object existence if something other than
ObjectNotExistException is returned in response to a particular request. So, here is the definitive statement of what it means for an Ice object to exist:
It is self-evident that an Ice object exists if a twoway invocation on it succeeds: obviously, the object received the invocation, processed it, and returned a result. However, note the qualification: this is true only for
twoway invocations; for oneway and datagram invocations (see Sections
32.14 and
32.15), nothing can be inferred about the existence of the corresponding Ice object by invoking an operation on it: because there is no reply from the server, the client-side Ice run time has no idea whether the request was dispatched successfully in the server or not. This includes user exceptions,
ObjectNotExistException,
FacetNotExistException, and
OperationNotExistException—these exceptions are never raised by oneway and datagram invocations, regardless of the actual state of the target object.
If a twoway invocation raises a user exception, the Ice object obviously exists: the Ice run time never raises user exceptions so, for an invocation to raise a user exception, the invocation was dispatched successfully in the server, and the operation implementation in the servant raised the exception.
If a twoway invocation raises FacetNotExistException, we do know that the corresponding Ice object indeed exists: the Ice run time raises
FacetNotExistException only if it can find the identity of the target object in the Active Servant Map (ASM), but cannot find the facet (see
Chapter 34) that was specified by the client.
1
These definitions simply capture the fact that a facet is a "sub-object" of an Ice object: if an invocation raises
ObjectNotExistException, we know that the facet does not exist either because, for a facet to exist, its Ice object must exist.
If an operation raises OperationNotExistException, we know that both the target Ice object and the target facet exist. However, the operation that the client attempted to invoke does not. (This is possible only if you use dynamic invocation (see
Chapter 36) or if you have mis-matched Slice definitions for client and server.)
The preceding definitions clearly state under what circumstances we can conclude that an Ice object (or its facet) does or does not exist. However, the preceding definitions are incomplete because operation invocations can have outcomes other than success or failure with
ObjectNotExistException,
FacetNotExistException, or
OperationNotExistException. For example, a client might receive a
MarshalException,
UnknownLocalException,
UnknownException, or
TimeoutException. In that case, the client cannot draw any conclusions about whether the Ice object on which it invoked a twoway operation exists or not—the exceptions simply indicate that something went wrong while the invocation was processed. So, to complete our definitions, we can state:
If a twoway invocation raises a run-time exception other than ObjectNotExistException, FacetNotExistException, or OperationNotExistException, nothing is known about the existence or non-existence of the Ice object that was the target of the invocation. Furthermore, it is impossible to determine the state of existence of an Ice object with a oneway or datagram invocation.
The preceding definitions capture the fact that, to make a determination of object existence or non-existence, the client-side Ice run time must be able to contact the server and, moreover, receive a reply from the server:
•
If the server can be contacted and returns an ObjectNotExistException (or
FacetNotExistException), the Ice object (or facet) does not exist. If the server returns an
OperationNotExistException, the Ice object (and its facet) exists, but does not provide the requested operation, which indicates a type mismatch due to client and server using out-of-sync Slice definitions or due to incorrect use of dynamic invocation.
Another way of looking at this is that a decision as to whether an object exists or not is
never made by the Ice run time and, instead, is
always made by the server-side
application code:
•
If an invocation returns ObjectNotExistException or
FacetNotExistException, the server-side application code was also involved:
This means that ObjectNotExistException and
FacetNotExistException are
authoritative: when you receive these exceptions, you can always believe what they tell you—the Ice run time never raises these exceptions without consulting your code, either implicitly (via an ASM lookup) or explicitly (by calling a servant locator’s
locate operation).
These semantics are motivated by the need to keep the Ice run time stateless with respect to object existence. For example, it would be nice to have stronger semantics, such as a promise that "once an Ice object has existed and been destroyed, all future requests to that Ice object also raise
ObjectNotExistException". However, to implement these semantics, the Ice run time would have to remember all object identities that were used in the past, and prevent their reuse for new Ice objects. Of course, this would be inherently non-scalable. In addition, it would prevent applications from controlling object identity; allowing such control for applications is important however, for example, to link the identity of an Ice object to its persistent state in a database (see
Chapter 40).
Note that, if the implementation of an operation calls another operation, dealing with
ObjectNotExistException may require some care. For example, suppose that the client holds a proxy to an object of type
Service and invokes an operation
provideService on it:
ServicePrx service = ...;
try {
service‑>provideService();
} catch (const ObjectNotExistException&) {
// Service does not exist.
}
Here is the implementation of provideService in the server, which makes a call on a helper object to implement the operation:
void
ServiceI::provideService(const Ice::Current&)
{
// ...
proxyToHelper‑>someOp();
// ...
}
If proxyToHelper happens to point at an object that was destroyed previously, the call to
someOp will throw
ObjectNotExistException. If the implementation of
provideService does not intercept this exception, the exception will propagate all the way back to the client, who will conclude that the service has been destroyed when, in fact, the service still exists but the helper object used to implement
provideService no longer exists.
Usually, this scenario is not a serious problem. Most often, the helper object cannot be destroyed while it is needed by
provideService due to the way the application is structured. In that case, no special action is necessary because
someOp will never throw
ObjectNotExistException. On the other hand, if it is possible for the helper object to be destroyed,
provideService can wrap a try-catch block for
ObjectNotExistException around the call to
someOp and throw an appropriate user exception from the exception handler (such as
ResourceUnavailable or similar).
