[wsrp-interfaces] [wsia][wsrp][wsrp-wsia joint interfaces]: Analternative

[Michael Freedman <Michael.Freedman@oracle.com>]

The attached is an alternative proposal to draft 0.1.2.  Its a more
simplified interface based on a (web) service model (not OO).  It
assumes the consumer provides the (transient) entity key to the
producer.  For simplicity I haven't defined these as bulk operations.
My personal preference is if we want to support bulk operations then
such operations would be added vs. replace these.  Also I have taken a
little different tack in defining each methods incoming parameters.  I
have separated what I thought were the logical base parameters to the
method from the contextual parameters -- those that identify the
consumer, the user etc.  This later group is packaged into a single data
element as a Call Context.  As the context varies a little for each
method, each method has its own call context defined.

Also, I have left out the Action method pending my understanding if/why
we need it.  Finally, so I can get this out today, I leave describe()
and setProperties() as an exercise for the reader.

I hope you will find this API not only a little simplier/clearer but
also more manageable.  I think it creates far fewer "dynamic" references
that hopefully lead to more performant and scalable systems.  Please let
me know what you think.
     -Mike-




The following is a counter-proposal to draft 0.1.2.  Its a more simplified
interface based on a (web) service model (not OO).  It assumes the
consumer provides the (transient) entity key to the producer.  For
simplicity I haven't defined these as bulk operations.  My personal
preference is if we want to support bulk operations then such operations
would be added vs. replace these.  Also I have taken a little different
tack in defining each methods incoming parameters.  I have separated
what I thought were the logical base parameters to the method from the
contextual parameters -- those that identify the consumer, the user etc. 
This later group is packaged into a single data element as a Call Context. 
As the context varies a little for each method, each method has its own
call context defined.

Also, I have left out the Action method pending my understanding if/why we need it.  Finally, so I can get this out today, I leave describe() and setProperties() as an exercise for the reader.

I hope you will find this API not only a little simplier/clearer but also more managable.  It creates far fewer "dynamic" references that hopefully lead to more performant and scalable systems.  Please let me know what you think.
     -Mike-

References:
A reference is something that allows a consumer to identify an element that is co-managed between the producer and the consumer.   Currently there are 4 references: ConsumerRef, SessionRef, PersistenceRef, and EntityRef.  each is represented as a String.  The ConsumerRef identifies who the consumer is.  The SessionRef identifies a conversation between a consumer and a producer.  A PersistenceRef identifies a particular entities persistent data.  An EntityRef identifies a converstation between a consumer and an entity within the producer.

The major model change here is that there is no longer a concept of producer entity (instances).  Rather we just exist in a world where we reference elements that need to be identified between the producer and the consumer.  For producers managing a heterogeneous set of portlets, each (type of) portlet is identified by an EntityType.  Each portlet (EntityType) is a single object ala a servlet.  References are passed to the portlet to allow it to establish its (data) context.  A portlet may maintain 2 types of data context, transient and persistent.  All transient data is scoped to a "session" that is managed/referred to via the SessionRef.  As sessions are (generally) shared across the entire producer, a second reference, the entityRef is provided that allows a portlet to safely partition a portion of the session data so that it is isolated/private from other "entities".  An "entity" (and its reference) in this case is defined by the consumer and refers to runtime reference to an instance of a particular type. Note:  as sessions are scoped to a particular consumer and entityRefs only exist to allow a portlet to namespace the session entityRefs need only be unique within a given consumer session.

Ref extends String;

ConsumerRef extends Ref;
SessionRef extends Ref;
PersistenceRef extends Ref;
EntityRef extends Ref;

A single method is provided to tell the producer a reference has gone out
of scope.  The method takes the reference itself and a reference type. 
The reference type is used to figure out the type of the reference being
released (as the reference is just an ID/String).  The ReleaseRefCtx
contains the contextual data this method (may) need to operate.  In
this case I think it would just contain the ConsumerRef and UserIdentify.
 
> void releaseRef(ReleaseRefCtx ctx, Ref ref, RefType type);

ConsumerIdentity
A consumer is identified either by a reference generated by the producer in response to a registerConsumer call.  This id may be as simple as the consumer name passed into the call.  The ConsumerProperties are consistent with those defined in the original draft and include the consumer name, vendor, etc.  registerConsumer must return a non-null value unless an exception is generated. The updateRegistration call is used to update the registration of an existing consumer.  We don't allow a new reference to be generated.  The Context parameter contains the SessionRef in case information is cached in transient data and needs updating as well.  I.e. the philosophy is that sessions may be created by the consumer at the first time the consumer talks to the producer once registered -- as all subsequent calls (can) occur within the conversational period they all receive a sessionRef.
 

> ConsumerRef registerConsumer(ConsumerProperties consumer);
> void updateRegistration(UpdateRegistrationCtx ctx, ConsumerRef
> ref, ConsumerProperties consumer);
>  
Sessions
Sessions represent a conversational period between a consumer and a
producer.  The SessionRef allows the producer to maintain transient
state scoped to this conversation.  Sessions only come into existence
via explicit request by the consumer.  This means a consumer must
be informed by the producer that a session is desired.  This can either
be done via meta data in the describe call or on demand.  I.e. 
all calls that can run receive a sessionRef can through a InvalidSession
exception.  Consumers are expected to catch this exception, create
a (new) session and retry the operation.

The createSession call takes a context, a reference to the consumer and the user identity.

> SessionRef createSession(CreateSessionContext ctx);
>  
> CreateSessionContext
>   {
>     private ConsumerRef consumer;
>     private UserIdentify user;
>   }
>  
Persistence
Many producers manage entities that require persistent data. 
Some producers rely on having the consumer manage this persisitent data. 
In this case, the producer/entity publishes a description of its data and
expects the consumer to pass the relevant values on request.  Other
entities manage the data itself.  The consumer defines when new persistent
instances come into existence.  In the case the producer/entity manages
the data, the consumer requests the producer to create a new persistent
record and to return a reference to it.  For convience, new records
can be created from old ones (by reference).  The consumer is expected
to supply this reference back to the producer in subsequent operational
requests.

A new persistence reference is creaetd either from an existing one or from scratch by identifying the type of persistent data.  The type is the entityType (aka the portlet ID).  The values aer passed as a single parameter (PersistenceType) that acts as a union.  Either the ref has a value or the entityType has a value.  The typeIsRef field allows the code to determine which element is valid in the union.

Cloning is similar to creating except is cerates siblings rather then children ( see the draft).

public PersistenceRef createPersistenceRef(CreatePersistenceRefCtx ctx,
  PersistenceType type, PersistentProperty[] values);

  CreatePersistenceRefCtx
  {
    private ConsumerRef consumer;
    private SessionRef session;
    private UserIdentify user;
  }

  PersistenceType
  {
    private boolean typeIsRef;
    private PersistenceRef ref;
    private String entityType;
  }

public PersistenceRef clonePersistenceRef(ClonePersistenceRefCtx ctx,
  PersistenceRef ref);

  ClonePersistenceRefCtx
  {
    private ConsumerRef consumer;
    private SessionRef session;
    private UserIdentify user;
  }

Operations:
As I don't yet understand how Actions differ semantically from getFragment
(from a protocol perspective) the only operation I model here is getFragment. 
getFragment takes a call context and request data.

The call context contains all the references that identify the context of this call.  This includes both an entityRef and entityType.  entityType is the portlet id; it identifies which portlet/entity is the target of the request.  entityRef is a consumer generated ID that identifies a particular consumer instance of this entity.  It can be used by the producer to namespace/cerate a private space in the session for storing entity specific transient data.

The request data is much like in the original draft -- its the data that is particular to this request including parameters, attributes (headers), modes/states, etc.

getFragment returns a FragmentResponse.  This includes responseProperties, the fragment itself, and persistentProperties.  The responseProperties are attribute/values returned to the consumer that either identify information about the fragment (char set etc) or define attribuets to be passed on subsequent requests to the same entityRef (aka headers).  The fragment is the content itself.  The persistentProperties is a property list of persistent values used to pass back nwe/updated values when a consumer is maintaining the persisitent data on behalf of the producer/portlet.
 

public FragmentResponse getFragment(GetFragmentCtx ctx, FragmentRequestData
  parameters);

  public class FragmentResponse
  {
    Property[] responseProperties;
    String fragment;
    PersistentProperty[] values;
  }

 
  public class GetFragmentCtx
  {
    private ConsumerRef consumerId;
    private SessionRef sessionId;
    private UserIdentify user;
    private Ref entityRef;
    private String entityType;
    private PersistenceRef persistenceId; // passed if portlet maintains data
    private PersistenceProperties persistenceData; // passed if consumer does
    private PreambleData updateData;  // update the included session or
                   // persistent data before generating the fragment;
  }

  public class FragmentRequestData
  {
    private String charSet;
    private String markupType;
    private String locale;
    private Boolean secureClientCommunications;
    private Property[] headers;
    private Property[] parameters;
    private DeviceInfo clientDevice;
    private Mode producerMode;
  }