The Silk Route system Ø The system goals

The Silk. Route system Ø The system goals Ø Scenario, examples Ø View Forests ØView forest and query composition ØView forest efficient execution 2005 rel-xml-ii 1

ØThe system goals • Publishing of relational data in xml form • Efficient execution of XQuery queries on published views Assumptions: • The only real data is the relational data • The public XML view is virtual • XQuery queries on published views are composed with view def. • Execution is always by SQL queries, since relational engines are efficient 2005 rel-xml-ii 2

ØScenario, examples The scenario: e-commerce • Each supplier has a private relational store • This store can be represented in a canonical XML view • The format/contents of both is private • There is a public XML schema def, that all participants in the scenario know about and use – the public view • Resellers, merchants pose queries on the public view, as per their current interests 2005 rel-xml-ii 3

Supplier private data: rel and XML (fig. 2, p. 5, fig. 4, p. 8) 2005 rel-xml-ii 4

The generation of the canonical view from the relational data is standard, can be completely automated But, the db is not converted to the XML view – doing this will throw away the availability of SQL for queries The canonical view is virtual And, it is just a basis for further work; this data is considered private For the public, there is an XML schema agreed upon by the community – the public view 2005 rel-xml-ii 5

The public view schema: (fig. 1 , p. 5) back 10 back 14 XQuery type syntax – an internal format (more human readable) of XML, used by the XQuery type checker Differences: one root, mainly products, change of names of some fields, reports inside product 2005 rel-xml-ii 6

The db admin can now write queries on the canonical view (p. 8, p. 9) $Canonical. View is a pre-defined variable, bound to the canonical view 2005 rel-xml-ii 7

A more complex query, with fusion (fig. 5, p. 10) 2005 rel-xml-ii 8

Note: XQuery is compositional: Given Q 1 on a a source S, Q 2 on results of Q 1, there exists a query Q 3 on S, such that Q 3(S) = Q 2(Q 1(S)) Simple approach: define Q 1 in a let, then use it as in Q 2 (see prev. page for let) Here (next page) is a query defining our supplier’s public view (ignore boxes) (fig. 6, p. 11) 2005 rel-xml-ii 9

public view 2005 rel-xml-ii 10

Users (e. g. , resellers) may pose queries on the public view e. g. somebody wants only the deeply discounted items (fig. 7) This can be composed with the def. of the public view An efficient composition, that does not generate intermediate results, and retrieves only what is needed, is more difficult 2005 rel-xml-ii 11

Here is an efficient composition (as might be produced by an optimizer) (fig. 8, p. 13) 2005 rel-xml-ii 12

ØView Forests But, the canonical view is also virtual – all data is in a relational db How do we represent XML virtual views over relational, so that composition can be performed? The representation in Silk. Route: View forest (typically a tree, but XQuery allows XML sequences, hence forest) • A tree structure – the structure of the XML view • Nodes are labeled XML labels, and with SQL fragments – these are used to generate the data 2005 rel-xml-ii 13

public view back 17 2005 rel-xml-ii 14

Explanations: • Each node is assigned a (Dewey style) id, no explicit tree is maintained • The XML label of an internal node is an element/attribute • The XML label of a leaf is an atomic type (string, float, …) • Each node is labeled with an SQL fragment, with FROM, WHERE, SELECT components (may be empty, not always shown) – A leaf must have a SELECT – Every variable used in a node, must be defined (in FROM) in the node or in an ancestor 2005 rel-xml-ii 15

The SQL query for a node: C(n) The union of the fragments for the nodes on the path from the node to the root Examples: N 1. 1. 1: N 1. 2: SELECT * (select an empty result tuple) FROM clothing c WHERE c. category = “outerwear” N 1. 2. 1. 1: N 1. 2. 5. 1: Note: if n is parent of n’, then C(n’) is an extension of C(n) (possibly the same, possibly a new var, or a new condition…) 2005 rel-xml-ii 16

The semantics of a view forest: Given a (relational) instance I, view forest V, V(I) is: • Collection of nodes (n, r) where – n is a node of V – r is a row in the answer of C(n) on I • (n, r) is parent of (n’, r’) if – n is parent of n’ in V – r, r’ are obtained from variables assignments that agree on all the variables of C(n) Look on the view (p. 14) and nodes N 1. 1. 1, N. 1. 2. 1. 1, N 1. 2. 5. 1 2005 rel-xml-ii 17

The semantics a naïve, inefficient method for generating the XML represented by a view V from instance I More efficient method: combine many SQL queries into one Will be discussed later (2 nd half of paper) 2005 rel-xml-ii 18

Here is a view forest for a fragment of the canonical view Can be generated automatically from db schema 2005 rel-xml-ii 19