1 / 40

Query Compilation

Query Compilation. Parsing Logical Query Plan. Source: our textbook, slides by Hector Garcia-Molina. SQL query. parse. parse tree. convert. answer. logical query plan. execute. apply laws. statistics. Pi. “improved” l.q.p. pick best. estimate result sizes.

tambre
Download Presentation

Query Compilation

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Query Compilation Parsing Logical Query Plan Source: our textbook, slides by Hector Garcia-Molina

  2. SQL query parse parse tree convert answer logical query plan execute apply laws statistics Pi “improved” l.q.p pick best estimate result sizes {(P1,C1),(P2,C2)...} l.q.p. +sizes estimate costs consider physical plans {P1,P2,…..}

  3. Outline • Convert SQL query to a parse tree • Semantic checking: attributes, relation names, types • Convert to a logical query plan (relational algebra expression) • deal with subqueries • Improve the logical query plan • use algebraic transformations • group together certain operators • evaluate logical plan based on estimated size of relations • Convert to a physical query plan • search the space of physical plans • choose order of operations • complete the physical query plan

  4. Parsing • Goal is to convert a text string containing a query into a parse tree data structure: • leaves form the text string (broken into lexical elements) • internal nodes are syntactic categories • Uses standard algorithmic techniques from compilers • given a grammar for the language (e.g., SQL), process the string and build the tree

  5. Example: SQL query SELECT title FROM StarsIn WHERE starName IN ( SELECT name FROM MovieStar WHERE birthdate LIKE ‘%1960’ ); (Find the movies with stars born in 1960) Assume we have a simplified grammar for SQL.

  6. SELECT <SelList> FROM <FromList> WHERE <Condition> <Attribute> <RelName> <Attribute> LIKE <Pattern> nameMovieStar birthDate‘%1960’ Example: Parse Tree <Query> <SFW> SELECT <SelList> FROM <FromList> WHERE <Condition> <Attribute> <RelName> <Tuple> IN <Query> titleStarsIn <Attribute> ( <Query> ) starName <SFW>

  7. The Preprocessor • replaces each reference to a view with a parse (sub)-tree that describes the view (i.e., a query) • does semantic checking: • are relations and views mentioned in the schema? • are attributes mentioned in the current scope? • are attribute types correct?

  8. Outline • Convert SQL query to a parse tree • Semantic checking: attributes, relation names, types • Convert to a logical query plan (relational algebra expression) • deal with subqueries • Improve the logical query plan • use algebraic transformations • group together certain operators • evaluate logical plan based on estimated size of relations • Convert to a physical query plan • search the space of physical plans • choose order of operations • complete the physical query plan

  9. Convert Parse Tree to Relational Algebra • Complete algorithm depends on specific grammar, which determines forms of the parse trees • Here give a flavor of the approach

  10. Conversion • Suppose there are no subqueries. • SELECT att-list FROM rel-list WHERE cond is converted into PROJatt-list(SELECTcond(PRODUCT(rel-list))), or att-list(cond( X (rel-list)))

  11. <Query> <SFW> SELECT movieTitle FROM StarsIn, MovieStar WHERE starName = name AND birthdate LIKE '%1960'; SELECT <SelList> FROM <FromList> WHERE <Condition> <Attribute> <RelName> , <FromList> AND <Condition> movieTitleStarsIn <RelName> <Attribute> LIKE <Pattern> MovieStarbirthdate'%1960' <Condition> <Attribute> = <Attribute> starName name

  12. Equivalent Algebraic Expression Tree movieTitle  starname = name AND birthdate LIKE '%1960' X StarsIn MovieStar

  13. Handling Subqueries • Recall the (equivalent) query: SELECT title FROM StarsIn WHERE starName IN ( SELECT name FROM MovieStar WHERE birthdate LIKE ‘%1960’ ); • Use an intermediate format called two-argument selection

  14. Example: Two-Argument Selection title  StarsIn <condition> <tuple> IN name <attribute> birthdate LIKE ‘%1960’ starName MovieStar

  15. Converting Two-Argument Selection • To continue the conversion, we need rules for replacing two-argument selection with a relational algebra expression • Different rules depending on the nature of the subquery • Here show example for IN operator and uncorrelated query (subquery computes a relation independent of the tuple being tested)

  16. Rules for IN   C R <Condition> X R  t IN S S C is the condition that equates attributes in t with corresponding attributes in S

  17. Example: Logical Query Plan title starName=name   StarsIn name birthdate LIKE ‘%1960’ MovieStar

  18. What if Subquery is Correlated? • Example is when subquery refers to the current tuple of the outer scope that is being tested • More complicated to deal with, since subquery cannot be translated in isolation • Need to incorporate external attributes in the translation • Some details are in textbook

  19. Outline • Convert SQL query to a parse tree • Semantic checking: attributes, relation names, types • Convert to a logical query plan (relational algebra expression) • deal with subqueries • Improve the logical query plan • use algebraic transformations • group together certain operators • evaluate logical plan based on estimated size of relations • Convert to a physical query plan • search the space of physical plans • choose order of operations • complete the physical query plan

  20. Improving the Logical Query Plan • There are numerous algebraic laws concerning relational algebra operations • By applying them to a logical query plan judiciously, we can get an equivalent query plan that can be executed more efficiently • Next we'll survey some of these laws

  21. Associative and Commutative Operations • product • natural join • set and bag union • set and bag intersection • associative: (A op B) op C = A op (B op C) • commutative: A op B = B op A

  22. Laws Involving Selection • Selections usually reduce the size of the relation • Usually good to do selections early, i.e., "push them down the tree" • Also can be helpful to break up a complex selection into parts

  23. Selection Splitting • C1 AND C2 (R) = C1 ( C2 (R)) • C1 OR C2 (R) = (C1 (R)) Uset (C2 (R)) if R is a set • C1 ( C2 (R)) = C2 ( C1 (R))

  24. Selection and Binary Operators • Must push selection to both arguments: • C (R U S) = C (R) U C (S) • Must push to first arg, optional for 2nd: • C (R - S) = C (R) - S • C (R - S) = C (R) - C (S) • Push to at least one arg with all attributes mentioned in C: • product, natural join, theta join, intersection • e.g., C (R X S) = C (R) X S, if R has all the atts in C

  25. Pushing Selection Up the Tree • Suppose we have relations • StarsIn(title,year,starName) • Movie(title,year,len,inColor,studioName) • and a view • CREATE VIEW MoviesOf1996 AS SELECT * FROM Movie WHERE year = 1996; • and the query • SELECT starName, studioName FROM MoviesOf1996 NATURAL JOIN StarsIn;

  26. Remember the rule C(R S) = C(R) S ? The Straightforward Tree starName,studioName year=1996 StarsIn Movie

  27. starName,studioName starName,studioName starName,studioName year=1996 year=1996 year=1996 year=1996 StarsIn StarsIn Movie StarsIn Movie push selection up tree push selection down tree Movie The Improved Logical Query Plan

  28. Laws Involving Projections • Consider adding in additional projections • Adding a projection lower in the tree can improve performance, since often tuple size is reduced • Usually not as helpful as pushing selections down • If a projection is inserted in the tree, then none of the eliminated attributes can appear above this point in the tree • Ex: L(R X S) = L(M(R) X N(S)), where M (resp. N) is all attributes of R (resp. S) that are used in L • Another example: • L(R Ubag S) = L(R) UbagL(S) But watch out for set union!

  29. Projection and Union • Example: Suppose • R(a,b) = {(1,2)}, • S(a,b) = {(1,3)}, and • bag projection • a(R Ubag S) = a({(1,2), (1,3)}) = {1, 1} = a({(1,2)} Ubaga({(1,3)}) =a(R) Ubaga(S) • a(R Uset S) = a({(1,2), (1,3)}) = {1, 1} ≠ {1} = {1} Uset {1} = a({(1,2)}) Useta({(1,3)}) =a(R) Useta(S)

  30. Push Projection Below Selection? • Rule: L(C(R)) = L(C(M(R))) where M is all attributes used by L or C • But is it a good idea? SELECT starName FROM StarsIn WHERE movieYear = 1996; starName starName movieYear=1996 extra work to scan through StarsIn twice movieYear=1996 starName,movieYear StarsIn StarsIn

  31. Joins and Products • Recall from the definitions of relational algebra: • R C S = C(R X S) (theta join) • R S = L(C(R X S)) (natural join) where C equates same-name attributes in R and S, and L includes all attributes of R and S dropping duplicates • To improve a logical query plan, replace a product followed by a selection with a join • Join algorithms are usually faster than doing product followed by selection (on large result of the product)

  32. Duplicate Elimination • Moving  down the tree is potentially beneficial as it can reduce the size of intermediate relations • Can be eliminated if argument has no duplicates • a relation with a primary key • a relation resulting from a grouping operator • Legal to push  through product, join, selection, and bag intersection • Ex: (R X S) = (R) X (S) • Cannot push  through bag union, bag difference or projection

  33. Duplicate Elimination Pitfalls • Example: Suppose • R has two copies of tuple t • S has one copy of t • T contains only (1,2) and (1,3) • bag union: • (R Ubag S) has one copy of t • (R) Ubag(S) has two copies of t • bag difference: • (R  S) has one copy of t • (R) (S) has no copies of t • bag projection: • ((T)) = {1} • ((T)) = {1,1}

  34. Grouping and Aggregation • Since  produces no duplicates: • (L(R)) = L(R) • Get rid of useless attributes: • L(R) = L(M(R)) where M contains all attributes in L • If L contains only MIN and MAX: • L(R) = L((R))

  35. year,MAX(birthdate) name=starName X MovieStar StarsIn Example • Suppose we have the relations MovieStar(name,addr,gender,birthdate) StarsIn(title,year,starName) • and we want to find the youngest star to appear in a movie for each year: SELECT year, MAX(birthdate) FROM MovieStar,StarsIn WHERE name = starName GROUP BY year;

  36. year,MAX(birthdate) year,birthdate year,MAX(birthdate) name=starName  name=starName  name=starName MovieStar StarsIn X MovieStar StarsIn MovieStar StarsIn Example cont'd year,MAX(birthdate) year,birthdate  • birthdate, name • year, starName

  37. Summary of LQP Improvements • Selections: • push down tree as far as possible • if condition is an AND, split and push separately • sometimes need to push up before pushing down • Projections: • can be pushed down • new ones can be added (but be careful) • Duplicate elimination: • sometimes can be removed • Selection/product combinations: • can sometimes be replaced with join

  38. Outline • Convert SQL query to a parse tree • Semantic checking: attributes, relation names, types • Convert to a logical query plan (relational algebra expression) • deal with subqueries • Improve the logical query plan • use algebraic transformations • group together certain operators • evaluate logical plan based on estimated size of relations • Convert to a physical query plan • search the space of physical plans • choose order of operations • complete the physical query plan

  39. Grouping Assoc/Comm Operators • Group together adjacent joins, adjacent unions, and adjacent intersections as siblings in the tree • Sets up the logical QP for future optimization when physical QP is constructed: determine best order for doing a sequence of joins (or unions or intersections) U D E F U D E F U A B C A B C

  40. Evaluating Logical Query Plans • The transformations discussed so far intuitively seem like good ideas • But how can we evaluate them more scientifically? • Estimate size of relations, also helpful in evaluating physical query plans • Coming up next…

More Related