Outline. Depth-first Binary Tree Traversal. Gerênciade Dados daweb -DCC922 - XML Query Processing. Motivation 24/03/2014

Transcription

1 Outline Gerênciade Dados daweb -DCC922 - XML Query Processing ( Apresentação basedaem material do livro-texto [Abiteboul et al., 2012]) 2014 Motivation Deep-first Tree Traversal Naïve Page-based Storage Processing Simple Paths Fragmenting XML Documents on Disk XML Node Identifiers (XML Query Processing Techniques) Motivation PTIME algorithms for evaluating XPath queries: Simple tree navigation; Translation into logic; Tree-automata techniques. These techniques assume XML data in memory. For verylargexml documentsthenumberofdisk accessesiscritical thegoalisnotreducingthe complexity of the algorithms, but designing methodsthatreducedisk accesses. Depth-first Binary Tree Traversal Preorder perform the following operations recursively, starting from the root: 1. Visit the root 2. Traverse the left subtree 3. Traverse the right subtree Postorder perform the following operations recursively, starting from the root: 1. Traverse the left tree 2. Traverse the right tree 3. Visit the root 1

2 Naïve Page-based Storage Storage strategy: A preorder traversal of the document from the root groups as many nodes as possible within a page; When the page is full, a new page is used to store the nodes found next, and soon. Processing Simple Paths /auctions/item requires the traversal of two disk pages; /auctions/item/description and /auctions/open_auctions/auction/initial both require traversing three disk pages; //initial requires traversing all the pages of the document. Remedies Smart fragmentation: Group nodes that are often accessed simultaneously, so that the number of pages that need to be accessed is globally reduced. Rich node identifiers: Use sophisticated node identifiers for reducing the cost of join operations. Fragmenting XML Documents Basic strategies can be conceptualized by a set of relations. XML queries are processed in two steps: 1. Translating the XML query into a relational query; 2. Evaluating the relational query on the relations representing the XML document content. The simplest approach considers a document as a set of edges stored in a single relation Edge(pid,cid,clabel) where cidis the ID of some node, pidis the ID of its parent and clabelis its label. 2

3 XML Document with Node Identifiers Partial Instance of the Edge Relation Processing XPath Queries //initial - Direct lookup on index Edge.clabel. /auctions/item - A join is needed. //auction//bid-unbounded number of joins that leads to an infinite union of XPath queries! //auction/bid //auction/*/bid //auction/*/*/bid wherea = σ clabel=actions (Edge), B= σ clabel=bid (Edge) Alternative Strategies Tag partitioning: The Edgerelationis partitioned into as many as tagbased relations as there are different tags in the original document reduces disk I/O needed to retrieve the IDs of elements having the same tag. Path partitioning: The idea is to encode a structural summary of the XML document as a set of relations, including one relation for each distinct parent-child path and an additional relation Paths containing all unique paths. 3

4 Tag-partitioned Edge Relations Path-partitioned Storage Reduces the disk I/O needed to retrieve the identifiers of elements having a given tag. Partitioning of queries with // steps in nonleading position remains as difficult as before. Query Processing using Path Partitioning //item//bid can be evaluated in two steps: 1. Scan the Pathsrelation and identify all parent-child paths matching the given linear XPath query; 2. For each of the paths thus obtained, scan the corresponding path-partitioned table. Many branches still require joins across the relations; however, the evaluation of // steps is simplified. For very structured data, the Pathsrelation is typically much smaller than the data set itself. Thus, the cost of the first processing step is likely negligible, while the performance benefits of avoiding numerous joins are quite important. XML Node Identifiers Within a persistent XML store, each node must be assigned a unique identifier (ID). An ID plays the role of a primary key to the node. We want to encapsulate structure information into these identifiers to facilitate query evaluation. Basic strategies: Region-based identifiers; Dewey-based identifiers. 4

5 Region-based Identifiers The so-called region-based identifier scheme simply assigns to each XML node n, the pair composed of the offset of its begin tag and the offset of its end tag. We denote this pair by (n.begin, n.end): the region-based identifier of the <a> element is the pair (0, 90); the region-based identifier of the <b> element is the pair (30, 50). Using Region-based Identifiers Comparing the region-based identifiers of two nodes n 1 and n 2 allows deciding whether n 1 is an ancestor of n 2. Observe that this is the case if and only if: n 1.start < n 2.start, and n 2.end < n 1.end. No need to use byte offset: (begintag, endtag). Count only opening and closing tags (as one unit each) and assign the resulting counter values to each element: for the <a><b> </b></a> doc the IDs are {(1,4),(2,3)}; (pre, post). Preorder and postorderids (n 1 is an ancestor of n 2 iff n 1.pre <= n 2.pre and n 2.post <= n 1.post ); (pre, post, depth). The depth can be used to decide whether a node is a parent of another one (n 1 is a parent of n 2 iffn 1 is an ancestor of n 2 and n 1.depth=n 2.depth-1. Region-based identifiers are quite compact, as their size only grows logarithmically with the number of nodes in a document. (pre, post, depth) Node Identifiers Dewey Node Identifiers 5

6 Using Dewey Identifiers Ancestor decided by looking if an ID is a prefix of another one (largest non-identical prefix for parent). Possible to decide preceding-sibling, followingsibling, preceding, following. Given two Dewey IDs n 1 and n 2, one can find the ID of the lowest common ancestor (LCA) of the corresponding nodes. The ID of the LCA is the longest common prefix of n 1 and n 2. Useful for keyword search. Main drawback: length (large, variable) of the identifiers. Updating Identifiers Consider a node with Dewey ID Suppose we insert a new first child to node 1.2. Then the ID of node becomes In general: Offset-based identifiers need to be updated as soon as a character is inserted or removed in the document; (start, end), (pre, post) and Dewey IDs need to be updated when the elements of the documents change. Possible to avoid relabeling on deletions, but gaps will appear in the labeling scheme. Relabeling operation quite costly. XML Query Processing Techniques Structural Joins Nested loop join Hash join Stack-based join Hollistic Twig Joins PathStack TwigStack References Abiteboul, S.; Manolescu, I.; Rigaux, P.; Rousset, M.-C.; Senellart, P. Web Data Management. Cambridge, 2012 (Cap. 4). Su-Cheng Haw, Chien-Sing Lee: Data storage practicesandqueryprocessingin XML databases: A survey. Knowledge-Based Sysems. 24(8): , Abiteboul, S.; Manolescu, I.; Rigaux, P.; Rousset, M.-C.; Senellart, P. Web Data Management. Cambridge, 2012 (Sect 4.3). 6