Native XML Support in DB2 Universal Database Matthias Nicola, Bert van der Linden IBM Silicon Valley Lab Presented by Mo Liu , Frate, Joseph and John Russo Some material in the talk is adapted from the slides of this paper’s conference talk. Agenda What is DB2 9 (Viper)? Native XML in the forthcoming version of DB2 Native XML Storage XML Schema Support XML indexes Querying XML data in DB2 Summery What is DB2 9 (Viper)? IBM DB2 9 is the next-generation hybrid data server with optimized management of both XML and relational data. IBM extended DB2 to include: • New storage techniques for efficient management of hierarchical structures inherent in XML documents. • New indexing technology New query language support (for XQuery), a new graphical query builder (for XQuery), and new query optimization techniques New support for validating XML data based on usersupplied schemas New administrative capabilities, including extensions to key database utilities Integration with popular application programming interfaces (APIs) XML Databases XML-enabled Databases The core data model is not XML (but e.g. relational) Mapping between XML data model and DB’s data model is required, or XML is stored as text E.g.: DB2 XML Extender v8 Native XML Databases Use the hierarchical XML data model to store and process XML internally No mapping, no storage as text Storage format = processing format E.g.: Forthcoming version of DB2 XML in Relational Databases – Today's Challenge Today’s Challenge: XML must be force fit into relational data model – 2 choices 1. Shredding or decomposing − Mapping from XML to relational often too complex − Loses hierarchical dependencies − Loses digital signature − Often requires dozens or hundreds of tables − Difficult to change original XML document 2. Large Object (BLOB, CLOB, Varchar) It allows for fast insert and retrieval of full documents but it needs XML parsing at query execution time. − SLOW performance − Search performance is slow (must parse at search time) − Retrieval of sub-documents is expensive − Update inside the document is slow − Indexing is inefficient (based on relative position) − Difficult to join with relational − Costs get worse as document size increases DB2 Hybrid XML Engine - Overview Integration of XML & Relational Capabilities in DB2 Native XML data type (not Varchar, not CLOB, not objectrelational) XML Capabilities in all DB2 components Applications combine XML & relational data Integrating XML and Relational in DB2 DB2 Hybrid XML Engine - Interfaces Data Definition create table dept(deptID int, deptdoc xml); Insert insert into dept(deptID, deptdoc) values (?,?) Index create index xmlindex1 on dept(deptdoc) generate key using xmlpattern ‘/dept/name’ as varchar(30); Retrieve select deptdoc from dept where deptID = ? SQL based Query select deptID, xmlquery('$d/dept/name' passing deptdoc as “d") from dept where deptID <> “PR27”; XQuery based Query for $book in db2-fn:xmlcolumn('BOOKS')/book for $entry in db2-fn:xmlcolumn('REVIEWS')/entry where $book/title = $entry/title return <review> {$entry/review/text()} </review>; Native XML Storage Efficient Document Tree Storage Information for Every Node Tag name, encoded as unique StringID A nodeID Node kind (e.g. element, attribute, etc.) Namespace / Namespace prefix Type annotation Pointer to parent Array of child pointers Hints to the kind & name of child nodes (for early-out navigation) For text/attribute nodes: the data itself XML Node Storage Layout XML Storage: “Regions Index” XML Indexes in DB2 Need index support to manage millions of XML documents Path-specific value indexes on XML columns to index frequently used elements and attributes XML-aware full-text indexing XML Value Indexes Table DEPT has two fields: “id” and “dept_doc” Field “dept_doc” is an XML document: <dept> <employee id=901> <name>John Doe</name> <phone>408 555 1212</phone> <office>344</office> </employee> </dept> CREATE INDEX idx1 ON DEPT(deptdoc) GENERATE KEY USING XMLPATTERN ‘/dept/employee/name’ AS SQL VARCHAR(35) Creates XML value index on employee name for all documents XML Value Indexes (continued) “xmlpattern” identifies the XML nodes to be indexed Subset of XPath language Wildcards, namespaces allowed XPath predicates such as /a/b[c=5] not supported “AS SQL” necessary to define data type, since DB2 does not require single XML schema for all documents in a table (so DB2 may not know data type to use for index) XML Value Indexes: Data Types Allowed data types for indexes: VARCHAR(n) VARCHAR HASHED, DOUBLE DATE TIMESTAMP DB2 index manager enhanced to handle special XML types (e.g., +0, -0, +INF, -INF, NaN) XML Value Indexes (continued) Node does not cast to the index type No error is raised No index entry created for that node Single document (e.g., XML field from single record) may contain 0, 1, or multiple index entries Different than relational index XML Value Indexes: unique indexes Unique indexes enforced within a document, and across all documents Example of unique index on employee id: CREATE UNIQUE INDEX idx2 ON DEPT(deptdoc) GENERATE KEY USING XMLPATTERN ‘/dept/employee/@id’ AS SQL DOUBLE XML Value Indexes: multiple elements or attributes Can create indexes on multiple elements or attributes Example: create index on all text nodes: CREATE INDEX idx3 ON DEPT(deptdoc) GENERATE KEY USING XMLPATTERN ‘//text()’ AS SQL VARCHAR(hashed) Example: create index on all attributes CREATE INDEX idx4 ON DEPT(deptdoc) GENERATE KEY USING XMLPATTERN ‘//@*’ AS SQL DOUBLE XML Value Indexes: namespaces Can index in a particular namespace XMLPATTERN can contain namespace declarations and prefixes Example: CREATE INDEX idx5 ON DEPT(deptdoc) GENERATE KEY USING XMLPATTERN ‘DECLARE NAMESPACE m=http://www.me.com/;/m:dept/m:employee/ m:name’ AS SQL VARCHAR(45) XML Value Indexes: internal For each XML document, each unique path mapped to an integer PathID (like StringID for tags) Each index entry includes: PathID to identify path of indexed node Value of the node cast to the index type RowID Identify rows containing the matching documents NodeID Identify matching nodes and regions within the documents XML Value Indexes: atomic vs. non-atomic Atomic Node: if it is an attribute, or if it is a text node, or if it is an element that has no child elements and exactly one text node child Indexes typically defined for atomic nodes Possible to define index on non-atomic nodes, e.g. index on ‘/dept/employee’ XML Value Indexes: atomic vs. non-atomic ‘/dept/employee’ non-atomic since it has child elements Single index entry for all of “employee” element, on all text nodes under “employee” (concatenation) Can be useful for mixed content in textoriented XML, e.g.: <title>The benefits of <bold>XML</bold></title> XML Full Text Indexes Allows full-text search of XML columns Can be fully indexed or partially indexed Example of full index: CREATE INDEX myIndex FOR TEXT ON DEPT(deptdoc) FORMAT XML CONNECT TO PERSONNELDB Example query: SELECT deptdoc FROM dept WHERE CONTAINS(deptdoc,’SECTIONS(“/dept/comment”) “Brazil” ‘) =1 Internal index structure System RX: One Part Relational, One Part XML Kevin Beyer, Roberta J Cochrane, Vanja Josifovski, Jim Kleewein, George Lapis, Guy Lohman, Bob Lyle, Fatma Özcan, Hamid Pirahesh, Normen Seemann, Tuong Truong, Bert Van der Linden, Brian Vickery, Chun Zhang Internal index structure XML index implemented with two B+ trees Path index Value Index Internal index structure: Path Index Path Index maps reverse path (revPath) to a generated path identifier (pathId) A “reverse path” is a list of node labels from leaf to root Compressed into vector of label identifiers Analogy to COLUMNS catalog from relational database Used for efficient processing of descendent queries Example: “//name” query Internal index structure: Value Index Value Index used to represent nodes Cconsists of the following key: PathId value nodeId rid Internal index structure: Value Index “value” is representation of the node’s data value when cast to the index’s data type “rid” identifies the row in the table (used for locking) “nodeId” identifies a node within the uses a Dewey node identifier can provide quick access to a node in the XML store “pathId” to retrieve specific path queries Internal index structure: Tradeoffs of Value Index key fields Order of keys is a tradeoff pathId first allows quick retrieval of specific queries e.g., index on //name might match many paths query on /book/author/name still has consecutive index entries but, query like //name=‘Maggie’ will need to examine every location in the index per matching path XML Schema Support Optional XML Schema validation Insert, Update, Query Limited support for DTDs an external entities Type annotation produced by validation persisted with document (query execution) Conforms to XML Query standard, XML Schema standard, XML standard XML Schema Support Register XML Schemas and DTDs in DB DB then stores type-annotated documents on disk, compiles execution plans with references to the XML Schemas Schemas stored in DB itself, for performance XML Schema Repository (XSR) XML Schema Support: XSR XSR consists of several new database catalog tables: Original XML schema documents for XML schema Binary representation of the schema for fast reference XML Schema Support: Registration Example: REGISTER XMLSCHEMA http://my.dept.com FROM dept.xsd AS departments.deptschema complete Schema URI is http://my.dept.com File with schema document is “dept.xsd” Schema identifier in DB is “deptschema” Belongs to relational DB schema “departments” XML Schema Support: Validation “XMLVALIDATE” function to validate documents in SQL statements Schema for validation is specified explicitly, or can be deduced from the schemaLocation hints in the instance documents Referenced by Schema URI or by identifier XML Schema Support: Validation Example (explicit by URI): INSERT INTO DEPT(detpdoc) VALUES xmlvalidate(?according to xmlschema uri ‘http://my.dept.com’) Example (explicit by ID): INSERT INTO DEPT(deptdoc) VALUES xmlvalidate(? according to xmlschema id departments.deptschema) XML Schema Support: Validation Example (implcit) DB2 tries to deduce schema from input document INSERT INTO dept(deptdoc) VALUES xmlvalidate(?) Try to find it in repository XML Schema Support: First repository design principle Repository will not require users to modify a schema before it is being registered require users to modify XML documents before they are inserted and validated Once document is validated in DB,it will never require updates to remain valid Considered infeasible to bulk-update all existing documents to become valid XML Schema Support: Second repository design principle Enable schema evolution Sequence of changes in an XML schema over its lifetime New or evolving business needs How to accomplish schema evolution is much-debated no standards business demands require it; so constrain problem XML Schema Support: Second repository design principle Flexibility of schema repository “paramount importance” DB2’s schema repository does not require namespace or the schema URI of each registered schema to be unique (user does not have control) Database-specific Schema identifier must be unique (user does have control) XML Schema Support: Second repository design principle Built-in support for one very simple type of schema evolution If new schema is backwards-cmpatible with old schema, then old schema can be replaced with new schema in the schema repository DB2 verifies all possible elements and attributes in old schema have same named types in the new schema Querying XML Data in DB2 Options Supported XQuery/XPath as a stand-alone language SQL embedded in XQuery XQuery/XPath embedded in SQL/XML Plain SQL for full-document retrieval DB2 treats SQL and XQuery as primary query languages. Both will operate independently on their data models Can also be integrated Sample Tables create table ship ( shipNo capacity class purchDate maintenance ) varchar(5) primary key not null, decimal(7,2), int, Notice the date, xml datatype xml create table captain ( captID varchar(5) primary key not null, lname varchar(20), fname varchar(20), DOB date, contact xml ) Sample XML Data Ship.maintenance <mrecord> <log> <mntid>2353</mntid> <shipno>39</shipno> <vendorid>2345</vendorid> <captid>9875</captid> <maintdate>01/10/2007</maintdate> <service>Removed rust on hull </service> <resolution>complete</resolution> <cost>13450.96</cost> <nextservice>01/10/2008</nextservice> </log> <log> <mntid>1254</mntid> <shipno>39</shipno> <vendorid>1253</vendorid> <captid>9234</captid> <maintdate>09/20/2005</maintdate> <service>Replace rudder</service> <resolution>complete</resolution> <cost>34532.21</cost> <nextservice>NA</nextservice> </log> </mrecord> Sample XML Data Captain.contactinfo <contactinfo> <Address> <street>234 Rolling Lane</street> <city>Rockport</city> <state>MA</state> <zipcode>01210</zipcode> </Address> <phone> <work>9783412321</work> <home>9722342134</home> <cell>9782452343</cell> <satellite>2023051243</satellite> </phone> <email>love2fish@finmail.com</email> </contactinfo> Standalone XQuery in DB2 for $s in db2-fn:xmlcolumn(‘ship.maintenance’) Db2-fn:xmlcolumn returns let $ml:= $s//log sequence of all documents where $ml/cost = > 10000 in the XML column order by $ml/shipno return <MaintenanceLog> {$ml/shipno,$ml} </MaintenanceLog> SQL Embedded in XQuery for $m in db2-fn:sqlquery(‘select maintenance from ship where class = 1’) let $ml := $m//log order by $ml/shipno return <maintenanceLog> {$ml} </mantenanceLog> This will return the documents for all class one ships. Select Statement using XML Column Select shipno,class,maintenance from ship where class = 1 This will produce the maintenance document for each ship that is class 1. We can also create views this way SQL/XML Queries Restricting results using XML element values select captid,lname,fname from captain where xmlexists(‘$c/contactinfo/Address[state=“MA”]’ passing captain.contact as “c” • This will return the captid, lname and fname of all captains who live in Massachusetts SQL/XML Queries Projecting XML element values Two functions: XMLQuery and XMLTable XMLQuery retrieves value for 1 element XMLTable retrieves value for multiple elements XMLQuery example: select xmlquery(‘$c/contactinfo/email’ passing contact as “c”) from captain where state = ‘MA’ This will return email addresses for all captains in Massachusetts SQL/XML Queries XMLQuery (Continued) We could also look for only first email for each captain by changing the first line: select xmlquery(‘$c/contactinfo/email[1]’ … Similarly, we could use xmlexists select xmlquery(‘$c/contactinfo/email’ passing contact as “c”) from captain where state = ‘MA’ and xmlexists(‘$c/contactinfo/email’ passing contact as “c”) to qualify: SQL/XML Queries XMLTable XMLTable retrieves XML elements Elements are mapped into result set columns Maps XML data as relational data SQL/XML Queries XMLTable Example select s.shipNo,sm.mid,sm.vid,sm.md,sm.cost from ship s, xmltable(‘$c/mrecord/log’ passing s.maintenance as “c” columns varchar(4) mid path ‘mntid’, varchar(4) vid path ‘vendorid’, date md path ‘maintdate’, decimal(7,2) cost path ‘cost’) as sm This will produce a list of maintenance logs for all ships Joining XML and Relational Data select c.captid,c.lname,c.fname from captain, ship where xmlexists(‘$s/mrec/log[captid=$c]’ passing ship.maintenance as “s”, captain.captid as “c”) If the captain was the captain of any ship when it underwent maintenance, he or she will be listed Using FLWR Expressions in SQL/XML select captid, xmlquery(‘for $c in $cn/contactinfo let $x := $c//city return $x’ passing contact as “cn”) from captain where class = 1 Returns captid as well as city information XMLElement XML Element allows you to publish relational data as XML select xmlelement(name “captain”, xmlelement(name “captid”, captid), xmlelement(name “lname”,lname), xmlelement(name “fname”,fname), xmlelement(name “class”,class)) from captain where class <= 2 XMLElement Output from previous command <captain> <captid>3563</captid> <lname>Smith</lname> <fname>John</fname> <class>2</class> </captain> … Aggregating and Grouping Data select xmlelement(name “captainlist”, xmlagg(xmlelement(name “captain”, xmlforest(cid as “captid”,lname as “lname”,fname as “fname”,class as “class”)) order by cid)) from captain group by class This query produces three captainlist elements each with a number of captains. Updating and Deleting XML Data Updates Use XMLParse command. You must specify the entire XML column to update. If you specify only 1 element to update, the rest of the data will be lost. Deletion Same as standard SQL Can also use xmlexists to use XML as qualifier Query Execution Plans •Separate parsers for SQL and XQuery statements •Integrated query compiler for both languages •QGMX is an internal query graph model •Query execution plans contain special operators for navigation (XSCAN), XML index access (XISCAN) and joins over XML indexes (XANDOR) Source: [2] Query Run-time Evaluation 3 major components added for processing queries over XML: XML Navigation XML Index Runtime XQuery Function Library Summary Problems with CLOB and Shredded XML storage Native XML support in DB2 offers: Hierarchical and parsed representation Path-specific XML indexing New XML join and query methods Integration of SQL and XQuery References [1] Nicola, M. and van der Linden, B. 2005. Native XML support in DB2 universal database. In Proceedings of the 31st international Conference on Very Large Data Bases (Trondheim, Norway, August 30 - September 02, 2005). Very Large Data Bases. VLDB Endowment, 1164-1174. [2] Beyer, K., Cochrane, R. J., Josifovski, V., Kleewein, J., Lapis, G., Lohman, G., Lyle, B., Özcan, F., Pirahesh, H., Seemann, N., Truong, T., Van der Linden, B., Vickery, B., and Zhang, C. 2005. System RX: one part relational, one part XML. In Proceedings of the 2005 ACM SIGMOD international Conference on Management of Data (Baltimore, Maryland, June 14 - 16, 2005). SIGMOD '05. ACM Press, New York, NY, 347-358. [3] http://www-128.ibm.com/developerworks/db2/library/techarticle/dm-0603saracco2/