Difference between revisions of "Content Manager"
(→gDoc Trees) |
(→Tree Predicates) |
||
Line 421: | Line 421: | ||
=== Tree Predicates === | === Tree Predicates === | ||
+ | |||
+ | The <code>gDoc</code> model is untyped, in that neither the topology of trees nor the values of their attributes or leaves are subject to constraints (beside those dictated by the XML serialisation). | ||
+ | Types are reintroduced later, under the view that they are ''projected'' on <code>gDoc</code> trees at the point of consumption. Type projections serve two main purposes: | ||
+ | |||
+ | * to validate the content of <code>gDoc</code> trees. | ||
+ | : The main use case for validation is at the point of content ingestion through the write operations of the Content Manager. In particular, a plugin may project a type on incoming <code>gDoc</code> trees, with a view to rejecting those that fail the projection. | ||
+ | * the identify the data of interest within <code>gDoc</code> trees. | ||
+ | : The main use case for content identification is at the point of content retrieval through the read operations of the Content Manager. Through the service, in particular, a client may ask plugins to return only the portion of the data that succeeds the projection, and to discard the rest. Content ''pruning'' results in minimal bandwidth consumption and delivers content to client in forms which are optimal for their processing or bindings. | ||
+ | |||
+ | Accordingly, support for type projections requires: | ||
+ | |||
+ | * a language of ''tree types'' with which clients and plugins can capture the required shape and content of <code>gDoc</code> trees. | ||
+ | * the ability to project such types over <code>gDoc</code> trees with both validation and pruning semantics. | ||
+ | |||
+ | XML schema languages are natural candidates for the choice of tree types. However, the also introduce complexity - both conceptually and in terms of tooling - which is not required when working with the subset of XML that corresponds to the <code>gDoc</code> model. As importantly, schema languages are strongly associated with validation and there are no implementations that use them towards document pruning (or indeed content extraction). | ||
+ | |||
+ | Accordingly, the tree API includes a native language of tree types, or ''tree predicates'', and support for projecting them over content for validation and pruning purposes. Tree predicates, in particular, can be used to constrain: | ||
+ | |||
+ | * the topology of <code>gDoc</code> trees, including the labels and cardinality of edges (e.g. the existence of at least one edge with a given label). | ||
+ | * the values of leaves, so that they conform to the textual literal of a range of atomic types (e.g. numbers or boolean values) or simply verify some type-specific predicate. | ||
+ | |||
+ | |||
+ | '''note''': Support for predicates on attributes and wildcard expressions on edge labels is forthcoming. | ||
+ | |||
+ | |||
+ | Tree predicates are defined in the packages <code>org.gcube.contentmanagement.contentmanager.stubs.model.predicates</code> and <code>org.gcube.contentmanagement.contentmanager.stubs.model.constraints</code> | ||
== Service Interface == | == Service Interface == |
Revision as of 16:14, 7 September 2010
The Content Manager service provides its clients with uniform access to content served by a variety of back-ends, both inside and outside the system. It is the central component of the gCube subsystem that deals with the organisation of content and related data.
Contents
Service Design
The Content Manager is designed as an OCMA service. In OCMA terms, it classifies as a multi-type, 1-N adapter service:
- it is a multi-type service because it supports two front types for, respectively, reading and writing content modelled as labelled trees.
- Collectively, the front types and the tree content model form the
gDoc
access type of the service.
- it is an adapter service because it adapts the
gDoc
access type to multiple back types, where each back type corresponds to the access type of a whole class of remote repositories.
- For this, the service employes an open architecture of type-specific plugins to which it delegates the creation and operation of its collection managers.
- Plugins are dynamically deployed within single instances of the services, and different instances may host different plugins. In addition, some plugins may support both service front types, i.e. grant read and write access to the corresponding repository. Others may instead support read-only access or, less commonly, write-only access.
The figure below overviews the design and use of the service in the context of one its running instances. The instance exposes three stateful port-types:
- the
ReadManager
serves as the interface of collection managers that offer read-only operations over the content of the bound collection.
- The interface defines the
gDocRead
front type of the service. - The front type and the identifier of the bound collection are published as Resource Properties of the manager, in accordance with OCMA patterns for publication and discovery of service state. A third Resource Property is the name of the bound plugin, i.e. the plugin to which the manager delegates the resolution of its requests.
- the
WriteManager
serves as the interface of collection managers that offer write-only operations over the content of the bound collection.
- The interface defines the
gDocWrite
front type of the service. - Again, the type, the identifier of the bound collection, and the name of the bound plugin are published as Resource Properties of the manager.
- the
Factory
serves as the front-end of a single WS Resource that createsReadManager
andWriteManager
resources .
- The resource is created at the activation of the service instance in the gCube Hosting Node.
- During its lifetime, it publishes creation requests as activation records. Conversely, it subscribes for the activation records that are published by other instances of the service, in line with OCMA patterns for replication of service state.
- The resource also publishes as a Resource Property the list of summary descriptions of the plugins that are hosted at the service instance.
Service plugins logically extend factory and collection manager resources with corresponding resource delegates. In particular:
- the factory delegate extends the
Factory
resource at plugin deployment time in order to handle requests that are specifically addressed to the plugin; - at each such request, the factory delegate processes plugin-specific parameters to create one ore more read delegates and/or write delegates, which the service instance uses to create and extend corresponding collection managers;
- future requests to the managers are then handled by their delegates, which translate the requests against the back-end repository that exposes the collection bound to the managers.
Finally, note that factory and collection managers are persistent resources and may thus be re-activated across restarts of the gCube Hosting Node:
- the factory persists the history of its activations, i.e. the activation records that it published and/or processed.
- the collection managers persist the name and state of their delegates.
Content Model
Architectural considerations aside, the most distinguished element in the design of the Content Manager is its content model. Rather than settle for a fixed set of document structures, the service adopts a generic structure that can act as a 'carrier' for an arbitrary number of concrete document models. In particular, the service deals with edge-labelled and node-attributed trees, the gDoc
trees.
The expectation here is that producers (service plugins) and consumers (service clients) will convene on concrete document models and exchange gDoc
trees with an agreed shape. The agreement may be bilateral or involve any number of parties, and it may apply to the entire document or to distinguished parts of it (e.g. document metadata, annotations, raw content packaging, etc). For maximum decoupling between consumers and producers, the agreement may reflect system-wide conventions and result in canonical tree forms.
gDoc Trees
A gDoc
tree has the following properties:
- its nodes may have an identifier and a number of uniquely named attributes;
- its edges have a label;
- its leaf nodes may have a value;
- its root may identify the collection of the corresponding document.
In particular:
- identifiers, attributes, and leaves have text values;
- attribute names and labels may be qualified with a namespace.
The figure below uses a graphical representation to show an example of a gDoc
tree.
gDoc
trees serialise to XML documents for exchange over the network. In particular:
- nodes serialise to elements and attributes serialise to element attributes
- elements are named like the edges that enter the corresponding nodes
- the document element is arbitrarily named
- the elements of inner nodes contain the elements of their children
- the elements of leaves contain their value
- node identifiers serialise to attributes called
http://gcube-system.org/namespaces/contentmanagement/gdoc:id
- node states serialise to attributes called
http://gcube-system.org/namespaces/contentmanagement/gdoc:state
- collection identifiers serialise to attributes called
http://gcube-system.org/namespaces/contentmanagement/gdoc:collID
For example, the gDoc
tree above may serialise as:
<g:gdoc xmlns:g="http://gcube-system.org/namespaces/contentmanagement/gdoc" g:id="1" x="..." y="..." g:collID="..."> <a g:id="2"> <b g:id="5"/> </a> <a g:id="3"> <c> <d >...</d> <d w="..">...</d> </c> </a> <b g:id="4" w="..." /> </g:gdoc>
Note that gDoc
trees inherit constraints from their XML serialisation. In particular, the names of edges, the names of attributes, the values of attributes, and the values of leaves are regulated by the definition of the format.
gDoc API
The XML serialisation of gDoc
trees is 'natural', in that it does not employ dedicated element structures for the representations nodes, edges, attributes, etc. This streamlines its manipulation with standard XMl technologies (e.g. XPath, XSLT, XQuery, DOM, SAX, etc.) and does not inhibit object binding technologies (e.g. JAXB, XStream, etc). As a native option, however, the service defines a bespoke object model and API for gDoc
trees which offer:
- dedicated support for tree processing requirements associated with the use of the service;
- transparencies and optimisations for tree storage, construction, deconstruction, and input/output.
While the model is available to service clients, it also forms the basis of the interface between the service and its plugins. For this reason, its main features are overviewed here while its client-oriented features are discussed later on.
As the figure below illustrates, the model is defined in org.gcube.contentmanagement.contentmanager.stubs.model.trees
in terms of the following components:
-
Node
: an abstract base for nodes with an identifier, a state, and a map ofQName
-ed attributes. -
State
: an inner enumeration ofNode
for node states. -
Edge
: AQName
-ed edge to a targetNode
. -
InnerNode
: aNode
with a list of outgoingEdge
s. -
Leaf
: aNode
with a value. -
gDoc
: anInnerNode
with a collection identifier. -
Nodes
: a collection of static utilities to generateNode
s andEdge
s. -
Bindings
: a collection of static utilities to serialise and deserialiseNodes
s to and from DOM trees and/or character streams. -
NodeView
: a base class for JAXB bindings toNode
s. -
GDocView
: aNodeViewM
for JAXB bindings toGDoc
nodes.
The model API is illustrated by example in the rest of this Section. The full list of methods and their signatures can be found in the code documentation.
Building Trees
The first and obvious way to create gDoc
trees is with the constructors of the concrete node classes (GDoc
, InnerNode
, Leaf
).
As a first example, the following code illustrates the creation of a tree with an attributed root and two leaf nodes:
GDoc doc = new GDoc("someid"); doc.setAttribute(new QName("x"), "1"); doc.setAttribute(new QName("someNS","y"), new Date().toString()); doc.collectionID("..."); Leaf leaf1 = new Leaf(null,"2"); //no identifier Leaf leaf2 = new Leaf(null,"true"); Edge e1 = new Edge(new QName("a"),leaf1); Edge e2 = new Edge(new QName("someNS","b"),leaf2); doc.add(e1,e2);
While already more convenient than cross-language and format-oriented tree APIs (e.g. DOM), step-by-step construction is verbose, even in the case of small trees.
For a first degree of improvement, the node classes offer rich suites of constructors and setter overloads that allow for more 'in-lined' tree constructions and absorb the creation of QName
s:
GDoc doc2 = new GDoc("someid", new Edge("a", new Leaf(null,"2")), new Edge("someNS","b", new Leaf(null,"true"))); doc2.setAttribute("x", "1"); doc2.setAttribute("someNS","y", new Date().toString()); doc2.collectionID("somecollID");
For additional convenience, the Nodes
class defines a large number of generators, i.e. factory methods that can be statically imported and then composed into a pseudo literal syntax for gDoc
trees:
import static org.gcube.contentmanagement.contentmanager.stubs.model.trees.Nodes.*; ... GDoc doc3 = attr( gdoc("somecollID","someid", e("a",2), e("someNS","b",true)), a("x",1),a("someNS","y",new Date()));>
Here, gdoc
, attr
, e
, a
are examples of node, attribute, and edge generators. Besides allowing fully in-lined tree expressions without the use of the new
operator, the generators offer QName
creation transparencies and object-to-string conversion transparencies (cf. the int
, boolean
, and Date
example above). The transparency of date conversions is particularly important here, as it ensures adherence to XML serialisation standards that are not natively adopted in Java (e.g. in the implementation of toString
). See the code documentation for the full list of available generators, as well as for the additional examples that follow:
doc = gdoc(); doc = gdoc("someid"); doc = gdoc("collectionid","someid"); doc = gdoc("1", e("a", n("2", //n() => inner node generator e("b",l("3",0))))); doc = gdoc( //no identifier e("a", attr( n("2", e("b",l("3",0)), //l()= explicit leaf generator for identity assignment e("a",l("4",0))), a("foo","0")))); doc = attr(gdoc("1", e("a",l("2",5)), e("b",attr( n("3",e("c",4)), a("foo",0))), e("c",5)), a("x",0)); doc = attr(gdoc("1", e("a", n("2", e("b",n("$2")))), e("a",n("a1", e("c",n( e("d","..."), e("d",attr( //l()= explicit leaf generator for attribute assignments l("<xml>..</xml>"), a("w",".."))))))), e("b",attr( n("1:/2"), a("w","...")))), a("x","http://org.acme:8080"),a("y","<a>...</a>"));
The literal construction of trees is particularly convenient in during testing, though it composes well with the programmatic construction in the development of production code:
Edge edge = .... InnerNode node = ....; attr( node.add(e("before","..."), edge, e("after","...")) ), a("newattr","...");
note: the node classes override equals
for equivalence-based comparisons, and hashCode
for correct use as keys within hash-based data structures, and toString
for convenience of debugging.
Serialising and Deserialising Trees
The Bindings
class offers static facilities to transform native models of gDoc
trees into XML-based models. Two representations are supported natively, based on which other XML-based representation can be produced using standard platform facilities (e.g. TRAX):
-
Bindings.toElement(GDoc)
converts native models ofgDoc
trees into equivalent DOM models. -
Bindings.fromElement(Element)
converts DOM models ofgDoc
trees into equivalent native models. -
Bindings.toXML(GDoc, Writer, boolean?)
converts the native model into XML document streams, optionally excluding document declarations. -
Bindings.fromXML(Reader)
converts XML document streams intogDoc
trees.
note: DOM conversions of native models are implemented directly, as they are most commonly required for interactions with the Content Manager service. Stream conversions are instead derived from DOM conversions via TRAX, at an additional processing cost.
note: conversions from native models to XML-based models assign the conventional name http://gcube-system.org/namespaces/contentmanagement/gdoc:gdoc
(cf. Bindings.GDOC_NS
, and Bindings.GDOC_NAME
constants) to the document element. Vice versa, conversion from XML-based representations to native models discard the name of the document element.
Here is a usage example, which shows that equivalence of native models is preserved under round-trip conversions.
import static org.gcube.contentmanagement.contentmanager.stubs.model.trees.Bindings.*; ... GDoc doc = .... //DOM conversion GDOc doc2 = fromElement(toElement(doc)); assert doc.equals(doc2); //true! //stream conversion StringWriter w = new StringWriter(); toXML(doc,w); GDOC doc3 = fromXML(w.toString()); assert doc.equals(doc3); //true!
note: due to the treatment of root element names, equivalence of XML-based representations is not necessarily preserved after round-trip conversion. It is preserved only if the XML-based representations have been previously produced with the conversion routines.
note: in all the conversions above, null
values in attribute and leaf values are serialised using a special constant (exposed programmatically as Node.NULL
).
note: the conversions are also available at arbitrary inner nodes, not only roots (cf. Bindings.nodeToElement(Node, QName?)
, Bindings.nodeFromElement(Element)
,Bindings.nodeToXML(Node, Writer, QName)
, and Bindings.nodeFromXML(Reader)
.
Consuming Trees
The gDoc
API offers simple means of procedural tree navigation. For declarative queries, clients can convert the model into an XML-based representation and leverage platform standards and popular offerings (e.g. XPath, XQuery, or XSLT implementations). If required, the gDoc
API can then be reasserted on query outputs.
The Node
class defines methods to expose the state common to all nodes of a gDoc
tree:
-
id()
: returns the identifier. -
state()
: returns the state. -
parent()
: returns the parent. -
ancestors()
: returns the list of all nodes from the parent to the root. -
ancestorsAndSelf()
: behavesancestors
but the returned list includes and starts with the recipient node. -
attributes()
: returns a copy of the attributes, indexed by name. -
attribute(QName)
: returns the value of an attribute with the given name (or fails). -
hasAttribute(QName)
: checks for the existence of an attribute with a given name.
note: identifiers can only be set at node creation time. State and attributes can be set and attributes removed at any point (cf. Node.state(Node.State)
, setAttribute(QName,String)
, removeAttribute(QName)
.
note: for convenience, all methods that take QName
s are overloaded to accept local names as well as (namespace,local name) pairs.
note: invoking methods that take node types is simplified by statically importing the class constants defined in the Nodes
class (cf. Nodes.N
for InnerNode.class
and Nodes.L
for Leaf.class
).
The Leaf
class adds methods to read and set the value (cf. value()
, value(String)
).
The InnerNode
class adds methods to navigate along edges and or identifiers:
-
children()
: returns the list of children. -
children(QName)
: returns the list of children under edges with the given label. -
<T extends Node> children(class<T>)
: returns the list of children of a given node type. -
<T extends Node> children(class<T>, QName)
: returns the list of children of a given node type under edges with a given label. -
child(QName)
: returns the child under an edge with a given label (or fails if there are zero o more such children). -
<T extends Node> child(Class<T>, QName)
:returns the child of a given node type under an edge with a given label (or fails if there are zero o more such children). -
descendants(QName*)
: returns the list of descendants that can be reached following edges with given labels. -
<T extends Node> descendants(Class<T>,QName*)
: returns the list of descendants of a given type that can be reached following edges with given labels. -
edges()
: returns the list of all the edges. -
edges(QName)
: returns the list of edges with a given label. -
edge(QName)
: returns the list of edges with a given label (o fails if there are zero or more such edges). -
hasEdge(QName)
: checks for the existence of an edge with a given label. -
labels()
: returns the list of labels.
note: edges can be added or more removed at any time (cf. add(Edge*)
, removeEdge(Edge*)
, removeEdge(QName)
).
note: as above, methods that take QName
s have overloads that accepts local names and, where appropriate, overloads that accept (namespace,local name) pairs.
note: as above, invoking methods that take node types is simplified by statically importing the corresponding constants in Nodes
(cf. Nodes.N
, Nodes.L
).
The GDoc
class adds a method to read and set the collection identifier (cf. collectionID
, collectionID(String)
).
Finally, the Edge
class exposes its label and target (cf. label()
, target()
).
The following example illustrates some of the supported idioms, do check the code documentation for detailed information about method signatures:
import static org.gcube.contentmanagement.contentmanager.stubs.model.trees.Nodes.*; GDoc doc = attr(gdoc("1", e("a",l("2",5)), e("b",attr( n("3",e("c",4)), a("foo",0))), e("c",5)), a("x",0)); //typed children String val = doc.child(L,"a").value(); //typed descendant String val2 = doc.descendant(N,"3").child(L,"c").value(); for (InnerNode node : doc.children(N)) for (QName l : node.labels()) //process label for (Node d : doc.descendants("b","e")) for (Edge siblingEdge : d.parent().edges()) if (siblingEdge.target()!=d) //process sibling of descendant
Binding Trees
Clients that expect gDoc
trees of a given form may wish to bind them to objects. The API offers two classes to streamline JAXB object bindings. In particular, it includes two base classes for node and document bindings to XML serialisations of gDoc
trees:
-
NodeView
is a base class for node bindings. The view binds and exposes the identifier of the node as well as the URL of the node, if one exists (cf.getID()
,getURL()
). Node URLs are discussed later.
-
GDocView
extendsNodeView
as a base class for document bindings. The view binds and exposes the collection identifier of the root node (cf.getCollID()
), in addition to what already bound and exposed via its superclass.
Clients can extend these classes and the corresponding bindings. The following example illustrates:
@XmlRootElement(name=Bindings.GDOC_NAME,namespace=Bindings.GDOC_NS) class MyDocView extends GDocView { @XmlElement(namespace="http://acme.org") int i; @XmlElement(namespace="http://acme.org") MyDocComponent c; class MyDocComponent extends MyNodeView { @XmlElement Date date; }
MyDocView
and MyDocComponent
are toy examples of user-defined views over gDoc
trees and tree nodes, and they should be familiar to JAXB users.
MyDocView
extends GDocView
and uses JAXB annotations to specify the qualified name of the document elements to which it will be bound. Here we have chosen a name that aligns with the serialisations produced by the Bindings
class, as shown above, but different names may be specified if the binding target serialisations produced through different means.MyDocView
then includes two fields in its own namespace, an integer field and a MyDocComponent
field, both of which are bound to XML elements. MyDocComponent
extends NodeView
, specifies a single Date
, and uses JAXB annotations to bind it to an XML element. In both classes, we have chosen simple JAXB annotations. For example, we have assumed that the gDoc
trees that will come to binding have labels that match the field names. The full range of JAXB facilities is of course available to customise bindings to less aligned trees.
Suppose now MyDocView
is to be bound to the gDoc
tree below. Wee use the generators of the gDoc API to denote the tree, but this is just for convenience of exposition; the tree may have been generated through any suitable means.
GDoc doc = gdoc("collID","123", e(NS,"i",3), e(NS,"c", n("789", e("d",l("1",new Date())), e("b",l("2",15)))), e(NS,"d",new Date()), e(NS,"b",n("456")));
Clearly, the tree contains a subset that matches the binding expectations of the classes above. As with all JAXB clients, the binding would require steps similar to the following:
JAXBContext context = JAXBContext.newInstance(MyDocView.class); ... //assuming a DOM binding to the tree has already occurred (other JAXB inputs could have been used instead, e.g. character streams) Element docElement = .... //bind MyDocView mv = (MyDocView) context.createUnmarshaller().unmarshal(docElement); ...mv.id()... ...mv.collID()... ...mv.url()... ...mv.i... ...mv.c... ...mv.id()... ...mv.url()... ...mv.c.d... //serialiase (again to DOM) Document dom = ....; Marshaller m = context.createMarshaller(); m.marshal(mv,dom);
Tree Predicates
The gDoc
model is untyped, in that neither the topology of trees nor the values of their attributes or leaves are subject to constraints (beside those dictated by the XML serialisation).
Types are reintroduced later, under the view that they are projected on gDoc
trees at the point of consumption. Type projections serve two main purposes:
- to validate the content of
gDoc
trees.
- The main use case for validation is at the point of content ingestion through the write operations of the Content Manager. In particular, a plugin may project a type on incoming
gDoc
trees, with a view to rejecting those that fail the projection.
- the identify the data of interest within
gDoc
trees.
- The main use case for content identification is at the point of content retrieval through the read operations of the Content Manager. Through the service, in particular, a client may ask plugins to return only the portion of the data that succeeds the projection, and to discard the rest. Content pruning results in minimal bandwidth consumption and delivers content to client in forms which are optimal for their processing or bindings.
Accordingly, support for type projections requires:
- a language of tree types with which clients and plugins can capture the required shape and content of
gDoc
trees. - the ability to project such types over
gDoc
trees with both validation and pruning semantics.
XML schema languages are natural candidates for the choice of tree types. However, the also introduce complexity - both conceptually and in terms of tooling - which is not required when working with the subset of XML that corresponds to the gDoc
model. As importantly, schema languages are strongly associated with validation and there are no implementations that use them towards document pruning (or indeed content extraction).
Accordingly, the tree API includes a native language of tree types, or tree predicates, and support for projecting them over content for validation and pruning purposes. Tree predicates, in particular, can be used to constrain:
- the topology of
gDoc
trees, including the labels and cardinality of edges (e.g. the existence of at least one edge with a given label). - the values of leaves, so that they conform to the textual literal of a range of atomic types (e.g. numbers or boolean values) or simply verify some type-specific predicate.
note: Support for predicates on attributes and wildcard expressions on edge labels is forthcoming.
Tree predicates are defined in the packages org.gcube.contentmanagement.contentmanager.stubs.model.predicates
and org.gcube.contentmanagement.contentmanager.stubs.model.constraints