Issues for Mailing list of the Deployment Revision Task Force

List of issues (green=resolved, yellow=pending Board vote, red=unresolved)

List options: All ; Open Issues only; or Closed Issues only

Issue 7557: No type safety when external packages are referenced
Issue 7667: Process Collocation Deployment Directive Capability Lacking
Issue 7877: UML 2.0 port issue
Issue 7880: Target Manager Resource Commitment Granularity
Issue 7940: Resource Deployment Description Clarification
Issue 7941: Resource Deployment Description Insufficient
Issue 7945: Requirement Satisfier "name" Attribute is not Optional
Issue 7954: Separation of components during deployment (opposite of collocation).
Issue 7971: Resource Commitment Sequence Diagram Clarification
Issue 8944: Enumerated SatisfierProperties
Issue 8979: Distinguish Node vs. Component Exec Parameters in Deployment Plan
Issue 8983: Deployment Requirements in Deployment Plan
Issue 8984: Information about Port Delegations Lost In Planning
Issue 9239: IDL name clash due to enumeration values defined twice

Issue 7557: No type safety when external packages are referenced (deployment-rtf)

The ComponentPackageReference element (6.4.8) allows to reference
an external package (to be resolved during planning by searching
repositories), for example to instantiate a subcomponent instance
from, in an assembly.


However, there is no type safety here; an assembly design tool has
no way of validating that the ports (which are referenced by
connections within the assembly) actually exist in the component
interface of the referenced component.


Actually, there is little point in ComponentPackageReference having
an optional "requiredType"; there should always be a mandatory type
that the referenced package must support.


To an assembly design tool, just having the "requiredType" repository
id is insufficient for validation, the full
ComponentInterfaceDescription
is necessary.


Proposed resolution:


In section 6.4.8.2, ComponentPackageReference Attributes, delete the
"requiredType" attribute.


Add a "requiredType" composition association to ComponentInterface-
Description, with multiplicity 1..1.


In section 6.4.8.3, ComponentPackageReference Constraints, delete the
constraints, and replace with "No constraints." (Rationale: when type
conformance is mandatory, the existing constraing will always be met.)


In section 6.4.8.4, ComponentPackageReference Semantics, replace the
existing text


  The planner will search one or more repositories for package
  configurations that satisfy all requirements.


with


  The planner will search one or more repositories for package
  configurations that support the requiredType, and that match the
  requiredUUID and requiredName attributes, if present.

Resolution:
Make the type information for references to external packages mandatory by pointing to the respective ComponentInterfaceDescription

Problem: The CCM DTD in the assembly descriptor has the
capability to direct deployment of components on the same
processor and the PIM and PSM for SWRadio Components infrastructure
supports the deployment concept of process collocation. The capability is
needed for deployment considerations to direct
which components are to be collocated within the same process. Flexibility
should be given to express how components are to be deployed.
One reason to do this is performance considerations. Another use case is
the components could be an assembly which represents an application service
that is deployed in one process space.
Recommended solution: Is to add an optional zero to many Process
Collocation
element to the ComponentAssemblyDescription.

Resolution:
Introduce an additional class called Locality to the ComponentAssemblyDescription that allows for specifying locality constraints on the deployment process. 
Add a new class called PlanLocality to the DeploymentPlan that allows for specifying special process locality constraints for the deployment engine executing the deployment plan.

UML 2.0 allows for ports to have a set of required interfaces and a set
of provided interfaces.


The definition of port interfaces in the PIM should allow for this, even though
the PSM for CCM must constrain this.


Note that change must also be consistently applied to the external reference endpoints
(all other endpoints are for component ports with an associated port interface
definition in the ComponentPortDescription class).

See issue #6602 for disposition  Resolution:
As part of the resolution to issue 7953, a constraint has been added that states each port stereotyped as SWRadioPort has a set of 1 required and 1 provided interface at most.  Since SWRadioPort is the parent for all Port stereotypes defined in the PIM,  by definition, this takes care of the problem addressed in issue 7877.  Hence, issue 7877 should be closed with no change required to the specification, per the resolution for issue 7953.
Revised Text:
N/A
Disposition:	Closed, no change

At the moment, resources are committed and later released according to
a Deployment Plan -- the Execution Manager just sends over the full
plan, and the Target Manager has to pick it apart. We considered this
a good idea -- after all, the plan contains all information about
resources that will be used by an application. In retrospect, however,
it looks like a design break. The plan contains a lot of information
that the Target Manager is not interested in. So the Target Manager
has to dissect information that it should not need to see in the first
place.


Also, and maybe more importantly, the only way of releasing resources
is by sending the very same plan to the Target Manager all over again,
as the Target Manager does not keep track of allocations.


This does not allow for the early release of resources, in case the
application does not need all that were reserved, or in case part of
the application finishes early.


Therefore I propose to replace the current "commitResources" and
"releaseResources" operations with a per-application "resource
commitment" object that allows to commit and release resources in a
fine-grained manner, and that keeps track of the current commitment,
so that the remaining resources that are in use by an application
can be released by simply destroying the object.


By not sending the full Deployment Plan, that would also dramatically
cut communication between the Execution Manager and the Target Manager.

Resolution:
As indicated by the issue, the "commitResources" and "releaseResources" are replaced by a per-application "resource commitment" object. This solution aims to minimize the communication between the Execution Manager and the Target Manager, both in terms of the amount of data and the required roundtrips.

First, a data structure "ResourceAllocation" is introduced into the Target Management Model. It identifies a resource by name, and an amount that is being allocated. An application's total set of allocated resources can be expressed as a sequence of this structure.

Sections 6.8.9, 6.8.10 and 6.8.11 describe the ResourceDeployment-
Description, InstanceResourceDeploymentDescription and Connection-
ResourceDeploymentDescription elements, which identify the resources
that were chosen to satisfy a requirement.


There are three minor editorial issues here:


(1)


In 6.8.11.2 (ConnectionResourceDeploymentDescription Attributes),
the targetName attribute always identifies an Interconnect or a Bridge,
but never a Node.


Proposed resolution:


Update the description for the targetName attribute; change the text
in parentheses from


  (i.e., the name of a Node, Interconnect or Bridge)


to


  (i.e., the name of an Interconnect or Bridge)



(2)


Also in 6.8.11.2 (ConnectionResourceDeploymentDescription Attributes),
the targetName provides a scope for the resource, not for a requirement.


Proposed resolution:


Update the description for the targetName attribute; change the
second half of the first sentence (after the comma) from


  [...], to provide scope for the requirementName.


to


  [...], to provide scope for the resourceName.



(3)


Finally, I'd like to clarify in section 6.8.10 (InstanceResource-
DeploymentDescription) that the resourceName is within the scope of
the node that this instance is deployed on.


Proposed resolution:


In section 6.8.10.5 (InstanceResourceDeploymentDescription semantics),
add the following text:


  An InstanceResourceDeploymentDescription is always used in the context
  of an InstanceDeploymentDescription, which identifies the node that a
  component instance is deployed on. This node provides scope for the
  resourceName.

Resolution:
Change the text as proposed.

In section 6.8.9, the ResourceDeploymentDescription, part of a
DeploymentPlan, contains information about the resources that were
chosen to satisfy a requirement.


It has a "resourceValue" association to an Any value.


However, resources and requirements have not just one value, but a
set of properties; see 6.10.4 (RequirementSatisfier) and 6.10.7
(Requirement).


Therefore, the "resourceValue" must be replaced with a set of
"Properties" elements.


Proposed resolution:


In section 6.8.9.3 (ResourceDeploymentDescription associations),
replace the current association


  resourceValue: Any [1]
        The aspect of the resource actually allocated, if any, of
        the appropriate type of the resource's SatisfierPropertyKind
        attribute. For Quantity, it is the ordinal allocated. For
        Allocation, it is the allocated capacity, for Selection, it
        is the matched string. For others, it is the value of the
        matched property.


with


  property: Property [*]
        The aspects of the resource actually allocated, if any. The
        property values are of the appropriate types according to
        their matched SatisfierProperty's SatisfierPropertyKind
        attribute. For Quantity, it is the ordinal allocated. For
        Allocation, it is the allocated capacity, for Selection, it
        is the matched string. For others, it is the value of the
        matched property

Resolution:
Change the text as proposed

In section 6.10.4.2 (RequirementSatisfier attribute), the
description for the "name" attribute reads, "An optional name
for the requirement satisfier."


This is in violation of our convention for "name" attributes
(see section 6.3), and in fact, other parts of the specification
make use of the fact that the name uniquely identifies a
RequirementSatisfier within its parent, e.g., the Resource
Deployment Description element; its resourceName identifies a
resource by the requirement satisfier's name.


Proposed resolution:


In section 6.10.4.2, change the description for the "name"
attribute from


  name: String    An optional name for the requirement satisfier.


to


  name: String    The requirement satisfier's name, which uniquely
                  identifies this requirement satisfier within its
                  container.

Resolution:
Change the text as proposed

The current specification does not support the separation of components during depoyment. It should be possible to specify a such a requirement as input for the deployment planning process. Use cases for this are for instance:
- to request the instantiation of components in different processes for single threaded component implementations
- to request the instantiation of components in different processes or on different hosts for fault tolerance reasons
- to request the instantiation of components on different hosts for performance reasons (with respect to CPU load)

This issue should be resolved together with issue #7666 (host collocation) and issue #7667 (process collocation).

See issue #7667 for disposition

In the latest document, with the resolution to issue 6039 applied,
figure 8-13 on page 123, contains a note attached to the
"commitResource"
operation between the DomainApplicationManager and the TargetManager,
indicating that:


    This call could come from either the DomainApplicationManager
    or the DomainApplication, as an implementation choice. 


It was intended that implementations can also delegate resource
commitment
can also be delegated to nodes.


Proposed resolution:


Change the note identified above to read


    Can be delegated to DomainApplication or NodeApplication.

Resolution:
Change the text as proposed above

Section 9.3.5, "SatisfierProperty," defines the concrete
data types that a SatisfierProperty's value can assume,
depending on the SatisfierPropertyKind.


For the "Attribute" kind, it allows the value to be of a
(user-defined) enumeration type. It then says, "if the value
of the SatisfierProperty is of enumeration type, the value
of the Property is of type string, containing the enumeration
value that must compare equal to the SatisfierProperty value.


This is straightforward enough. E.g., it allows a
DomainAdministrator to enumerate the kinds of processors
in a domain,


  enum Processors { ix86, ppc, sparc };


and then to tag each node with a resource containing a
property of this type, its value specifying the actual
processor type of the node.


It seemed like a good idea. Yet this feature does not buy
much. The DomainAdministrator could just as well use an
attribute of type string to the same effect. The only
advantage is that the enumerated property allows the
DomainAdministrator to constrain the values that a
SatisfierProperty may have. But it's the DomainAdmininstrator
constraining himself, so it's not much of a constraint.


However, this feature does add a dependency on DynamicAny
for both the TargetManager and the Planner, which do not
know about any user-defined enumeration types. All the other
SatisfierPropertyKind options use basic IDL data types (long,
unsigned long, double or string).


So, because of the added complexity, and the low mileage, I
propose to remove enumerated properties from the spec.


Proposed resolution:


In section 9.3.5, "SatisfierProperty," change the fourth
and fifth bullet from:


    For the Attribute kind, the value of the SatisfierProperty
    is of type long, double, string, or an enumeration type.
    In the case of long, double or string, the value of the
    Property must be of the same type. If the value of the
    SatisfierProperty is of enumeration type, the value of
    the Property is of type string, containing the enumeration
    value that must compare equal to the SatisfierProperty
    value.


    For the Selection kind, the value of the SatisfierProperty
    is a sequence of type long, double, string, or an
    enumeration type. The same rules as for the Attribute kind
    apply.


To:


    For the Attribute kind, the value of the SatisfierProperty
    is of type long, double or string. The value of the
    Property must be of the same type and compare equal to
    the SatisfierProperty value.


    For the Selection kind, the value of the SatisfierProperty
    is a sequence of type long, double or string. The same
    rules as for the Attribute kind apply.

Resolution:
Because of the added complexity, and the low mileage, I propose to remove enumerated properties from the spec. 

The resolution for issue 6435 split the former "execution
parameters" into "node execution parameters" and "component
execution parameters" within the Component Data Model, by
extending the MonolithicImplementationDescription element
to have separate "nodeExecParameter" and "componentExec-
Parameter" property lists.


However, the resolution neglected to make a similar change
to the Execution Data Model -- the MonolithicDeployment-
Description element in the Deployment Plan only has a single
list of "execParameter" properties.


This should be changed to be in sync with the Component
Data Model.


Proposed resolution:


In section 6.8.3.3 (MonoliticDeploymentDescription
associations), remove the "execParameter" association.


Instead, add the following two associations:


  nodeExecParameter: Property[*]
    Execution parameters that are passed to the target
    environment. Copied from the MonoliticImplementation-
    Description.
  componentExecParameter: Property[*]
    Execution parameters that are passed to the primary
    artifact. Copied from the MonoliticImplementation-
    Description.

Resolution:
Split the "execution parameters" in the Execution Data Model into two sets of parameters to be in sync with the Component Data Model.

The Deployment Plan, in its Artifact Deployment Description
and Monolithic Deployment Description element, copies the
deployment requirements from the component data model, i.e.,
from the Implementation Artifact Description and Monolithic
Implementation Description, respectively.


However, while the deployment requirements for a monolithic
implementation are represented using the Implementation
Requirement element, Monolithic Deployment Description only
allows deployment requirements to be of type Requirement,
and is thus not able to completely capture the original
requirement.


This is an effect of the resolution of issue 6392, which
extended deployment requirements for monolithic
implementations in the component data model from
"Requirement" to "ImplementationRequirement", but neglected
to make a similar change to the Monolitic Deployment
Description.


This information may not be necessary in the deployment plan
in the first place, as an Instance Deployment Description
now, also as a result of this resolution, captures the
resources that were chosen to satisfy the monolithic
implementation's requirements.


As noted in the resolution, an issue to remove the copies
of deployment requirements should be raised separately. Here
it is.


Proposed resolution:


Remove the copies of deployment requirements, in both
Artifact Deployment Description and Monolithic Deployment
Description.


In section 6.8.2.1, "ArtifactDeploymentDescription
Description", change the last sentence from


  Execution parameters and deployment requirements are
  copied from the ImplementationArtifactDescription.


to


  Execution parameters are copied from the
  ImplementationArtifactDescription.


In section 6.8.2.3, "ArtifactDeploymentDescription
Associations", remove the "deployRequirement"
association.


In section 6.8.3.1, "MonolithicDeploymentDescription
Description", change the last sentence from


  The execution parameters and deployment requirements
  are copied from the MonolithicImplementationDescription.


to


  The execution parameters are copied from the
  MonolithicImplementationDescription.


In section 6.8.3.3, "MonolithicDeploymentDescription
Associations", remove the "deployRequirement"
association.

Remove the copies of deployment requirements, in both ArtifactDeploymentDescription and MonolithicDeploymentDescription.

The resolution for issue 6392 introduced the ability
for a monolithic implementation to delegate some of its
ports to a resource.


To this effect, the ImplementationRequirement element
has three attributes, "resourceUsage", "resourcePort"
and "componentPort" to identify how the component
instance uses the resource.


In the deployment plan, the InstanceResourceDeployment-
Description element captures which resource was chosen
to satisfy the requirement. It has a "resorceUsage",
but neither "resourcePort" and "componentPort"
attributes.


This information is necessary in the deployment plan.


Proposed resolution:


Add "resourcePort" and "componentPort" attributes to
the InstanceResourceDeploymentDescription element, to
carry the information from the original implementation
requirement.


In section 6.8.12.2, add two attributes:


  resourcePort [0..1]: The resource port used by the
      component instance. Copied from the original
      ImplementationRequirement.


  componentPort [0..1]: The component port delegated
      to or used by the resource. Copied from the
      original ImplementationRequirement.

Resolution:
Add "resourcePort" and "componentPort" attributes to the InstanceResourceDeploymentDescription element, to carry the information from the original implementation requirement.

In section 6.4.11.5 (Semantics of LocalityKind), different values including "DifferentProcess" and "NoConstraint" are defined for the LocalityKind enumeration of the Component Data Model. In section 6.8.6.5 (Semantics of PlanLocalityKind), the constants for the PlanLocalityKind enumeration are defined for the Execution Data Model, also containg these two names mentioned above. The problem is that enumeration names are introduced into their parent scope. Thus, the definition of the PlanLocalityKind reusing the same names as used in the LocalityKind enumeration causes a name clash in the IDL.


Proposed resolution:

Add a prefix "Plan" to all the constants of the PlanLocalityKind enumeration.

In section 6.8.6.5, change the bullet list from


SameProcess: This value specifies that the component instances that the PlanLocality element referers to shall be started in the same process by the deployment engine.

DifferentProcess: This value specifies that the component instances that the PlanLocality element referers to shall be started in different processes by the deployment engine.

NoConstraint: This value specifies that there is no locality constraint for the component instances that the PlanLocality element refers to. The purpose of this special value is to enable to switch locality constraints temporarily off without loosing the structure of the DeploymentPlan, i.e. the PlanLocality class instance with its associations to InstanceDeploymentDescription can be kept for later reuse but has currently no impact on the execution of the DeploymentPlan. 
to


PlanSameProcess: This value specifies that the component instances that the PlanLocality element referers to shall be started in the same process by the deployment engine.

PlanDifferentProcess: This value specifies that the component instances that the PlanLocality element referers to shall be started in different processes by the deployment engine.

PlanNoConstraint: This value specifies that there is no locality constraint for the component instances that the PlanLocality element refers to. The purpose of this special value is to enable to switch locality constraints temporarily off without loosing the structure of the DeploymentPlan, i.e. the PlanLocality class instance with its associations to InstanceDeploymentDescription can be kept for later reuse but has currently no impact on the execution of the DeploymentPlan.