org.apache.hadoop.hive.ql.optimizer
Class MapJoinProcessor

java.lang.Object
  org.apache.hadoop.hive.ql.optimizer.MapJoinProcessor

All Implemented Interfaces:

Transform

public class MapJoinProcessor
extends Object
implements Transform

Implementation of one of the rule-based map join optimization. User passes hints to specify map-joins and during this optimization, all user specified map joins are converted to MapJoins - the reduce sink operator above the join are converted to map sink operators. In future, once statistics are implemented, this transformation can also be done based on costs.

Nested Class Summary

static class	MapJoinProcessor.CurrentMapJoin CurrentMapJoin.
static class	MapJoinProcessor.Default Default.
static class	MapJoinProcessor.MapJoinDefault MapJoinDefault.
static class	MapJoinProcessor.MapJoinFS MapJoinFS.
static class	MapJoinProcessor.MapJoinWalkerCtx MapJoinWalkerCtx.

Constructor Summary

MapJoinProcessor()
empty constructor.

Method Summary

static void	checkMapJoin(int mapJoinPos, JoinCondDesc[] condns)
static MapJoinOperator	convertMapJoin(LinkedHashMap<Operator<? extends Serializable>,OpParseContext> opParseCtxMap, JoinOperator op, QBJoinTree joinTree, int mapJoinPos, boolean noCheckOuterJoin) convert a regular join to a a map-side join.
MapJoinOperator	generateMapJoinOperator(ParseContext pctx, JoinOperator op, QBJoinTree joinTree, int mapJoinPos)
static String	genMapJoinOpAndLocalWork(MapredWork newWork, JoinOperator op, int mapJoinPos)
static HashSet<Integer>	getBigTableCandidates(JoinCondDesc[] condns) Get a list of big table candidates.
static NodeProcessor	getCurrentMapJoin()
static NodeProcessor	getDefault()
static NodeProcessor	getMapJoinDefault()
static NodeProcessor	getMapJoinFS()
ParseContext	transform(ParseContext pactx) Transform the query tree.

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Detail

MapJoinProcessor

public MapJoinProcessor()

empty constructor.

Method Detail

genMapJoinOpAndLocalWork

public static String genMapJoinOpAndLocalWork(MapredWork newWork,
                                              JoinOperator op,
                                              int mapJoinPos)
                                       throws SemanticException

Throws:

SemanticException

convertMapJoin

public static MapJoinOperator convertMapJoin(LinkedHashMap<Operator<? extends Serializable>,OpParseContext> opParseCtxMap,
                                             JoinOperator op,
                                             QBJoinTree joinTree,
                                             int mapJoinPos,
                                             boolean noCheckOuterJoin)
                                      throws SemanticException

convert a regular join to a a map-side join.

Parameters:

opParseCtxMap -

op - join operator

joinTree - qb join tree

mapJoinPos - position of the source to be read as part of map-reduce framework. All other sources are cached in memory

noCheckOuterJoin -

Throws:

SemanticException

generateMapJoinOperator

public MapJoinOperator generateMapJoinOperator(ParseContext pctx,
                                               JoinOperator op,
                                               QBJoinTree joinTree,
                                               int mapJoinPos)
                                        throws SemanticException

Throws:

SemanticException

getBigTableCandidates

public static HashSet<Integer> getBigTableCandidates(JoinCondDesc[] condns)

Get a list of big table candidates. Only the tables in the returned set can be used as big table in the join operation. The logic here is to scan the join condition array from left to right. If see a inner join, and the bigTableCandidates is empty or the outer join that we last saw is a right outer join, add both side of this inner join to big table candidates only if they are not in bad position. If see a left outer join, set lastSeenRightOuterJoin to false, and the bigTableCandidates is empty, add the left side to it, and if the bigTableCandidates is not empty, do nothing (which means the bigTableCandidates is from left side). If see a right outer join, set lastSeenRightOuterJoin to true, clear the bigTableCandidates, and add right side to the bigTableCandidates, it means the right side of a right outer join always win. If see a full outer join, return null immediately (no one can be the big table, can not do a mapjoin).

Parameters:

condns -

Returns:

list of big table candidates

checkMapJoin

public static void checkMapJoin(int mapJoinPos,
                                JoinCondDesc[] condns)
                         throws SemanticException

Throws:

SemanticException

transform

public ParseContext transform(ParseContext pactx)
                       throws SemanticException

Transform the query tree. For each join, check if it is a map-side join (user specified). If yes, convert it to a map-side join.

Specified by:

transform in interface Transform

Parameters:

pactx - current parse context

Returns:

ParseContext

Throws:

SemanticException

getMapJoinFS

public static NodeProcessor getMapJoinFS()

getMapJoinDefault

public static NodeProcessor getMapJoinDefault()

getDefault

public static NodeProcessor getDefault()

getCurrentMapJoin

public static NodeProcessor getCurrentMapJoin()