Difference between revisions of "Scope Management"

Revision as of 17:26, 18 August 2008

What's in Scope?

In gCube, a resource may only be shared within one or more scopes. Outside these, the resource can neither be discovered nor used.

A Running Instance, in particular, may operate in different scopes at different times. For this to happen, clients must specify a scope when contacting the instance. Similarly, any state which the instance might create must retain the scope of the call which triggered the creation.

It is so that resources, WS-Resources, and calls are all scoped in gCube.

Modelling Scope

In gCube, there are three levels of scope: the Infrastructure, the Virtual Organization (VO), and the Virtual Research Environment (VRE).

The levels are hierarchically related: the VRE groups resources which satisfy the requirements of a particular application, the VO groups resources which support the operation of one or more VREs, and the infrastructure groups resources which support the operation of one or more VOs. In particular, a VRE exists only in the context of a VO which in turn exists in the context of some infrastructure. Informally, we say that the VRE is below the VO and that the infrastructure is above it.

We use expressions of the form /A/B/C to denote a VRE C below a VO B below an infrastructure A. Informally, /A/B/C may be thought of as the fully qualified name of C. More formally, we use the following grammar for scope expressions:

SCOPE : = = I | VO | VRE
I :== /NAME
VO :== I/NAME
VRE :== VO/NAME
NAME :== <a conventional label>

In gCF, scope expressions can be conveniently thought as serialisations of GCUBEScope objects. to be continued...

Scope Rules

Scoping in gCube is rooted on a first, basic rule:

R1) a resource in the scope of a VO is also in the scope of the VREs below.

So, a resource in /A/B is also in /A/B/C but not necessarily in /A or in /A/D.

R1 captures the intuition that resources shared by a Virtual Organization are immediately available to all its applications. As noted above, these resources are most often dedicated to the management of the VREs: gCube Hosting Nodes and Running Instances of Core services have typical example of VO-level resources.

Note: The converse of R1 does not hold: a resource shared in a VRE may be dedicated to it. That is, it may not be shared in the VO above (and, for R1), in the other VREs of that same VO).

Note: R1 does not extend to infrastructure scope: a resource in that scope may not be automatically shared in the VOs below. This captures the intuition that most resources with infrastructure scope manage VOs in a way which is transparent to them, 'in the background'.

We now consider additional rules which restrict, widen, or simply complement R1 in reflection of the semantics and pragmatics of individual resources types.

Scoping Rules for gCube Hosting Nodes

Sharing in gCube starts with GHNs, the main fabric of the infrastructure. GHNs are first pooled at the infrastructure scope and, in accordance with R1), must be explicitly joined to one ore more VOs. The rules which govern the sharing of GHNs are constrained by the following rules:

GHN1) all the scopes of a GHN must be rooted in the same infrastructure.

GHN2) a GHN in a given scope is also in the scopes above.

GHN1 rules assert that GHNs cannot be shared across infrastructures (though the machine on which they run can of course). GHN2 states that the converse of' R1 holds for GHNs: a GHN cannot be dedicated to VREs or VOs.

Note: gCore allows site administrators to configure their GHN to start in zero ore more VOs. gCF enforces GHN1 by making sure that the start VOs are all rooted in the same infrastructure. It then uses GHN2 to optimise scope management (keeps rack only of VO scopes).

Scoping Rules for gCube Services

Let SV be a gCube Service.

SV1) if SV is shared in a scope it is also shared below it: SV in S => for each S' in S. R in S'

Note: This extends R1) to infrastructure-level scope make scoping simplest for services.

gCore implications: none special, scopes are immutable for services (i.e. for the ones already deployed).

Scoping Rules for Running Instances (RIs) of gCube services

Running instances are a first example of resource type for which scoping rules are further constrained by the scope of related resources. In fact, running instances relate to their services and the hosting nodes on which they are deployed.

Let SV(RI) and GHN(RI) denote, respectively, the service of a RI and the GHN on which the RI is deployed.

RI1) RI can only be shared where its GHN and service can: RI in S => GHN(RI) in S and SV(RI) in S 

Note: The opposite of RI1) does not hold: a RI is not necessarily shared wherever its GHN and its service are.

gCore implications:

• on add, enforce RI1) (ServiceContext)
• on startup,
  • generic service: subscribe consumer for GHN remove scope events only (ServiceContext) => remove all RI scopes contained in those removed from the GHN (consumer).
  • local service: subscribe consumer for GHN remove and add scope events (ServiceContext) => add new GHN scopes to RI (consumer).
• on bootstrap,
  • allow JNDI configuration of zero or more start scopes (RIBuilder)
  • static only : add all GHN scopes if no start scopes are available (RIBuilder). Prevent in dynamic bootstrap.

Scoping Rules for WS-Resource generated by Running Instances of gCube services

WS-Resources are anther example of resource type for which scoping rules are further constrained by the scope of related resources, in this case the running instance which generate them. Let WSR denote a WS-Resource and RI(WSR) the RI through which has generated WSR.

WSR1) WSR can only be shared where its RI can: WSR in S => RI(WSR) in S 

Note: The opposite of RI1) does not hold: a RI is not necessarily shared wherever its GHN and its service are.

gCore implications:

• on creation, enforce WSR1) (GCUBEHandler)
• on remove, remove from all WS-Resource which have been created by some stateful-port-type of the RI below the removed RI scope;
  • due to persistence, cannot be easily performed locally, use instead queries to the IS

Configuring scope (Static scope)

Static scope applies only for those resources that can be manually created by humans. For such gCube resources, a scope S can be statically (i.e. manually) configured and changed through their appropriate configuration files.

Static scope for gCube Hosting Nodes

A gHN is always manually deployed, therefore a static scope configuration is always needed. Such a configuration is expressed with two parameters in the $GLOBUS_LOCATION/config/GHNConfig.xml file.

• infrastructure reports the GI the gHN joins (and it can be one and only one, as for GHN1). In the following example, the gHN is configured to join the /gcube infrastructure:

 <environment
        name="infrastructure"
        value="gcube"
        type="java.lang.String"
        override="false" />

• startScope is a comma-separated list of VOs the gHN joins. These VOs has to be below the infrastructure (as for GHN2)), otherwise they are ignored. The list of VOs can be an empty list. In the following example, the gHN is configured to join both the /gcube/devsec and the /gcube/testing VO:

<environment
        name="startScopes"
        value="devsec,testing"
        type="java.lang.String"
        override="false" />

Static scope for gCube Services

A gCube service has always only a static scope, specified at service registration time in the Software Repository service.

Static scope for Running Instances (RIs) of gCube services

Static scope for RIs has a mean only for those that are manually deployed. A RI can be configured to join one or more VOs and/or VREs with a comma-separated list. Due to RI1), the VOs has to be a subset of the VOs to which the GHN(RI) is joined to and the VREs has to be below the VOs to which the GHN(RI) is joined to.

In order to specify the static scope of a RI, insert a startScopes parameter in the service context section of the JNDI's RI with the fully qualified name of the scope as follows:

<service name="gcube/common/vremanagement/Deployer">
	
	<environment 
		name="configDir" 
	 	value="@config.dir@" 
	 	type="java.lang.String"
	 	override="false" />
		 		 	
	<environment 
		name="securityManagerClass" 
	 	value="org.gcube.common.core.security.GCUBESimpleServiceSecurityManager" 
	 	type="java.lang.String"
	 	override="false" />
	 	
	<environment 
		name="startScopes" 
	 	value="/gcube/devsec/EM,/gcube/testing/test1" 
	 	type="java.lang.String"
	 	override="false" />

</service>

Note: The startScopes parameter is optional: if it is not used, the RI is automatically configured by the gCore to join the VO(s) expressed in the startScopes parameter of the GHNConfig.xml.

gCube Calls & the gCube Handler

[coming soon]

Scope Managers

[coming soon]

The Client Perspective:Scopes & Stubs