/**
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements.See the NOTICE file
* distributed with this work for additional information
* regarding copyright ownership.The ASF licenses this file
* to you under the Apache License, Version 2.0 (the
* "License"); you may not use this file except in compliance
* with the License.You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package org.apache.hadoop.hive.ql.optimizer;
import java.util.ArrayList;
import java.util.Arrays;
import java.util.HashMap;
import java.util.List;
import java.util.Map;
import java.util.Set;
import java.util.Stack;
import org.apache.hadoop.hive.common.JavaUtils;
import org.apache.hadoop.hive.conf.HiveConf;
import org.apache.hadoop.hive.metastore.api.Order;
import org.apache.hadoop.hive.ql.ErrorMsg;
import org.apache.hadoop.hive.ql.exec.DummyStoreOperator;
import org.apache.hadoop.hive.ql.exec.JoinOperator;
import org.apache.hadoop.hive.ql.exec.MapJoinOperator;
import org.apache.hadoop.hive.ql.exec.Operator;
import org.apache.hadoop.hive.ql.exec.ReduceSinkOperator;
import org.apache.hadoop.hive.ql.exec.SMBMapJoinOperator;
import org.apache.hadoop.hive.ql.exec.TableScanOperator;
import org.apache.hadoop.hive.ql.lib.Node;
import org.apache.hadoop.hive.ql.lib.NodeProcessor;
import org.apache.hadoop.hive.ql.lib.NodeProcessorCtx;
import org.apache.hadoop.hive.ql.metadata.Partition;
import org.apache.hadoop.hive.ql.metadata.Table;
import org.apache.hadoop.hive.ql.parse.ParseContext;
import org.apache.hadoop.hive.ql.parse.PrunedPartitionList;
import org.apache.hadoop.hive.ql.parse.QB;
import org.apache.hadoop.hive.ql.parse.SemanticException;
import org.apache.hadoop.hive.ql.parse.TableAccessAnalyzer;
import org.apache.hadoop.hive.ql.plan.ExprNodeDesc;
import org.apache.hadoop.hive.ql.plan.JoinCondDesc;
import org.apache.hadoop.hive.ql.plan.JoinDesc;
import org.apache.hadoop.hive.ql.plan.OperatorDesc;
import org.apache.hadoop.hive.ql.plan.ReduceSinkDesc;
import org.apache.hadoop.hive.ql.plan.SMBJoinDesc;
import org.apache.hadoop.util.ReflectionUtils;
//try to replace a bucket map join with a sorted merge map join
abstract public class AbstractSMBJoinProc extends AbstractBucketJoinProc implements NodeProcessor {
public AbstractSMBJoinProc(ParseContext pctx) {
super(pctx);
}
public AbstractSMBJoinProc() {
super();
}
@Override
abstract public Object process(Node nd, Stack<Node> stack, NodeProcessorCtx procCtx,
Object... nodeOutputs) throws SemanticException;
/*
* Return true or false based on whether a bucketed mapjoin can be converted successfully to
* a sort-merge map join operator. The following checks are performed:
* a. The mapjoin under consideration is a bucketed mapjoin.
* b. All the tables are sorted in same order, such that join columns is equal to or a prefix
* of the sort columns.
*/
protected boolean canConvertBucketMapJoinToSMBJoin(MapJoinOperator mapJoinOp,
Stack<Node> stack,
SortBucketJoinProcCtx smbJoinContext,
Object... nodeOutputs) throws SemanticException {
// Check whether the mapjoin is a bucketed mapjoin.
// The above can be ascertained by checking the big table bucket -> small table buckets
// mapping in the mapjoin descriptor.
// First check if this map-join operator is already a BucketMapJoin or not. If not give up
// we are only trying to convert a BucketMapJoin to sort-BucketMapJoin.
if (mapJoinOp.getConf().getAliasBucketFileNameMapping() == null
|| mapJoinOp.getConf().getAliasBucketFileNameMapping().size() == 0) {
return false;
}
if (!this.pGraphContext.getMapJoinOps().contains(mapJoinOp)) {
return false;
}
String[] srcs = mapJoinOp.getConf().getBaseSrc();
for (int srcPos = 0; srcPos < srcs.length; srcPos++) {
srcs[srcPos] = QB.getAppendedAliasFromId(mapJoinOp.getConf().getId(), srcs[srcPos]);
}
boolean tableEligibleForBucketedSortMergeJoin = true;
// All the tables/partitions columns should be sorted in the same order
// For example, if tables A and B are being joined on columns c1, c2 and c3
// which are the sorted and bucketed columns. The join would work, as long
// c1, c2 and c3 are sorted in the same order.
List<Order> sortColumnsFirstTable = new ArrayList<Order>();
for (int pos = 0; pos < srcs.length; pos++) {
tableEligibleForBucketedSortMergeJoin = tableEligibleForBucketedSortMergeJoin
&& isEligibleForBucketSortMergeJoin(smbJoinContext,
mapJoinOp.getConf().getKeys().get((byte) pos),
mapJoinOp.getConf().getAliasToOpInfo(),
srcs,
pos,
sortColumnsFirstTable);
}
if (!tableEligibleForBucketedSortMergeJoin) {
// this is a mapjoin but not suited for a sort merge bucket map join. check outer joins
if (MapJoinProcessor.checkMapJoin(mapJoinOp.getConf().getPosBigTable(),
mapJoinOp.getConf().getConds()) < 0) {
throw new SemanticException(
ErrorMsg.INVALID_BIGTABLE_MAPJOIN.format(mapJoinOp.getConf().getBigTableAlias()));
}
return false;
}
smbJoinContext.setSrcs(srcs);
return true;
}
// Convert the bucket map-join operator to a sort-merge map join operator
protected SMBMapJoinOperator convertBucketMapJoinToSMBJoin(MapJoinOperator mapJoinOp,
SortBucketJoinProcCtx smbJoinContext) {
String[] srcs = smbJoinContext.getSrcs();
SMBMapJoinOperator smbJop = new SMBMapJoinOperator(mapJoinOp);
SMBJoinDesc smbJoinDesc = new SMBJoinDesc(mapJoinOp.getConf());
smbJop.setConf(smbJoinDesc);
HashMap<Byte, String> tagToAlias = new HashMap<Byte, String>();
for (int i = 0; i < srcs.length; i++) {
tagToAlias.put((byte) i, srcs[i]);
}
smbJoinDesc.setTagToAlias(tagToAlias);
int indexInListMapJoinNoReducer =
this.pGraphContext.getListMapJoinOpsNoReducer().indexOf(mapJoinOp);
if (indexInListMapJoinNoReducer >= 0 ) {
this.pGraphContext.getListMapJoinOpsNoReducer().remove(indexInListMapJoinNoReducer);
this.pGraphContext.getListMapJoinOpsNoReducer().add(indexInListMapJoinNoReducer, smbJop);
}
Map<String, DummyStoreOperator> aliasToSink =
new HashMap<String, DummyStoreOperator>();
// For all parents (other than the big table), insert a dummy store operator
/* Consider a query like:
*
* select * from
* (subq1 --> has a filter)
* join
* (subq2 --> has a filter)
* on some key
*
* Let us assume that subq1 is the small table (either specified by the user or inferred
* automatically). The following operator tree will be created:
*
* TableScan (subq1) --> Select --> Filter --> DummyStore
* \
* \ SMBJoin
* /
* /
* TableScan (subq2) --> Select --> Filter
*/
List<Operator<? extends OperatorDesc>> parentOperators = mapJoinOp.getParentOperators();
for (int i = 0; i < parentOperators.size(); i++) {
Operator<? extends OperatorDesc> par = parentOperators.get(i);
int index = par.getChildOperators().indexOf(mapJoinOp);
par.getChildOperators().remove(index);
if (i == smbJoinDesc.getPosBigTable()) {
par.getChildOperators().add(index, smbJop);
}
else {
DummyStoreOperator dummyStoreOp = new DummyStoreOperator(par.getCompilationOpContext());
par.getChildOperators().add(index, dummyStoreOp);
List<Operator<? extends OperatorDesc>> childrenOps =
new ArrayList<Operator<? extends OperatorDesc>>();
childrenOps.add(smbJop);
dummyStoreOp.setChildOperators(childrenOps);
List<Operator<? extends OperatorDesc>> parentOps =
new ArrayList<Operator<? extends OperatorDesc>>();
parentOps.add(par);
dummyStoreOp.setParentOperators(parentOps);
aliasToSink.put(srcs[i], dummyStoreOp);
smbJop.getParentOperators().remove(i);
smbJop.getParentOperators().add(i, dummyStoreOp);
}
}
smbJoinDesc.setAliasToSink(aliasToSink);
List<Operator<? extends OperatorDesc>> childOps = mapJoinOp.getChildOperators();
for (int i = 0; i < childOps.size(); i++) {
Operator<? extends OperatorDesc> child = childOps.get(i);
int index = child.getParentOperators().indexOf(mapJoinOp);
child.getParentOperators().remove(index);
child.getParentOperators().add(index, smbJop);
}
// Data structures coming from QBJoinTree
smbJop.getConf().setQBJoinTreeProps(mapJoinOp.getConf());
//
pGraphContext.getSmbMapJoinOps().add(smbJop);
pGraphContext.getMapJoinOps().remove(mapJoinOp);
return smbJop;
}
/**
* Whether this table is eligible for a sort-merge join.
*
* @param pctx parse context
* @param op map join operator being considered
* @param joinTree join tree being considered
* @param alias table alias in the join tree being checked
* @param pos position of the table
* @param sortColumnsFirstTable The names and order of the sorted columns for the first table.
* It is not initialized when pos = 0.
* @return
* @throws SemanticException
*/
private boolean isEligibleForBucketSortMergeJoin(
SortBucketJoinProcCtx smbJoinContext,
List<ExprNodeDesc> keys,
Map<String, Operator<? extends OperatorDesc>> aliasToOpInfo,
String[] aliases,
int pos,
List<Order> sortColumnsFirstTable) throws SemanticException {
String alias = aliases[pos];
/*
* Consider a query like:
*
* select -- mapjoin(subq1) -- * from
* (select a.key, a.value from tbl1 a) subq1
* join
* (select a.key, a.value from tbl2 a) subq2
* on subq1.key = subq2.key;
*
* aliasToOpInfo contains the SelectOperator for subq1 and subq2.
* We need to traverse the tree (using TableAccessAnalyzer) to get to the base
* table. If the object being map-joined is a base table, then aliasToOpInfo
* contains the TableScanOperator, and TableAccessAnalyzer is a no-op.
*/
Operator<? extends OperatorDesc> topOp = aliasToOpInfo.get(alias);
if (topOp == null) {
return false;
}
// get all join columns from join keys
List<String> joinCols = toColumns(keys);
if (joinCols == null || joinCols.isEmpty()) {
return false;
}
TableScanOperator tso = TableAccessAnalyzer.genRootTableScan(topOp, joinCols);
if (tso == null) {
return false;
}
// For nested sub-queries, the alias mapping is not maintained in QB currently.
/*
* Consider a query like:
*
* select count(*) from
* (
* select key, count(*) from
* (
* select --mapjoin(a)-- a.key as key, a.value as val1, b.value as val2
* from tbl1 a join tbl2 b on a.key = b.key
* ) subq1
* group by key
* ) subq2;
*
* The table alias should be subq2:subq1:a which needs to be fetched from topOps.
*/
if (pGraphContext.getTopOps().containsValue(tso)) {
for (Map.Entry<String, TableScanOperator> topOpEntry :
this.pGraphContext.getTopOps().entrySet()) {
if (topOpEntry.getValue() == tso) {
alias = topOpEntry.getKey();
aliases[pos] = alias;
break;
}
}
}
else {
// Ideally, this should never happen, and this should be an assert.
return false;
}
Table tbl = tso.getConf().getTableMetadata();
if (tbl.isPartitioned()) {
PrunedPartitionList prunedParts = pGraphContext.getPrunedPartitions(alias, tso);
List<Partition> partitions = prunedParts.getNotDeniedPartns();
// Populate the names and order of columns for the first partition of the
// first table
if ((pos == 0) && (partitions != null) && (!partitions.isEmpty())) {
Partition firstPartition = partitions.get(0);
sortColumnsFirstTable.addAll(firstPartition.getSortCols());
}
for (Partition partition : prunedParts.getNotDeniedPartns()) {
if (!checkSortColsAndJoinCols(partition.getSortCols(),
joinCols,
sortColumnsFirstTable)) {
return false;
}
}
return true;
}
// Populate the names and order of columns for the first table
if (pos == 0) {
sortColumnsFirstTable.addAll(tbl.getSortCols());
}
return checkSortColsAndJoinCols(tbl.getSortCols(),
joinCols,
sortColumnsFirstTable);
}
private boolean checkSortColsAndJoinCols(List<Order> sortCols,
List<String> joinCols,
List<Order> sortColumnsFirstPartition) {
if (sortCols == null || sortCols.size() < joinCols.size()) {
return false;
}
// A join is eligible for a sort-merge join, only if it is eligible for
// a bucketized map join. So, we dont need to check for bucketized map
// join here. We are guaranteed that the join keys contain all the
// bucketized keys (note that the order need not be the same).
List<String> sortColNames = new ArrayList<String>();
// The join columns should contain all the sort columns
// The sort columns of all the tables should be in the same order
// compare the column names and the order with the first table/partition.
for (int pos = 0; pos < sortCols.size(); pos++) {
Order o = sortCols.get(pos);
if (pos < sortColumnsFirstPartition.size()) {
if (o.getOrder() != sortColumnsFirstPartition.get(pos).getOrder()) {
return false;
}
}
sortColNames.add(o.getCol());
}
// The column names and order (ascending/descending) matched
// The first 'n' sorted columns should be the same as the joinCols, where
// 'n' is the size of join columns.
// For eg: if the table is sorted by (a,b,c), it is OK to convert if the join is
// on (a), (a,b), or any combination of (a,b,c):
// (a,b,c), (a,c,b), (c,a,b), (c,b,a), (b,c,a), (b,a,c)
// but it is not OK to convert if the join is on (a,c)
return sortColNames.subList(0, joinCols.size()).containsAll(joinCols);
}
// Can the join operator be converted to a sort-merge join operator ?
// It is already verified that the join can be converted to a bucket map join
protected boolean checkConvertJoinToSMBJoin(
JoinOperator joinOperator,
SortBucketJoinProcCtx smbJoinContext) throws SemanticException {
if (!this.pGraphContext.getJoinOps().contains(joinOperator)) {
return false;
}
String[] srcs = joinOperator.getConf().getBaseSrc();
// All the tables/partitions columns should be sorted in the same order
// For example, if tables A and B are being joined on columns c1, c2 and c3
// which are the sorted and bucketed columns. The join would work, as long
// c1, c2 and c3 are sorted in the same order.
List<Order> sortColumnsFirstTable = new ArrayList<Order>();
for (int pos = 0; pos < srcs.length; pos++) {
if (!isEligibleForBucketSortMergeJoin(smbJoinContext,
smbJoinContext.getKeyExprMap().get((byte) pos),
joinOperator.getConf().getAliasToOpInfo(),
srcs,
pos,
sortColumnsFirstTable)) {
return false;
}
}
smbJoinContext.setSrcs(srcs);
return true;
}
// Can the join operator be converted to a sort-merge join operator ?
protected boolean canConvertJoinToSMBJoin(
JoinOperator joinOperator,
SortBucketJoinProcCtx smbJoinContext) throws SemanticException {
boolean canConvert =
canConvertJoinToBucketMapJoin(
joinOperator,
smbJoinContext
);
if (!canConvert) {
return false;
}
return checkConvertJoinToSMBJoin(joinOperator, smbJoinContext);
}
// Can the join operator be converted to a bucket map-merge join operator ?
@SuppressWarnings("unchecked")
protected boolean canConvertJoinToBucketMapJoin(
JoinOperator joinOp,
SortBucketJoinProcCtx context) throws SemanticException {
// This has already been inspected and rejected
if (context.getRejectedJoinOps().contains(joinOp)) {
return false;
}
if (!this.pGraphContext.getJoinOps().contains(joinOp)) {
return false;
}
Class<? extends BigTableSelectorForAutoSMJ> bigTableMatcherClass = null;
try {
String selector = HiveConf.getVar(pGraphContext.getConf(),
HiveConf.ConfVars.HIVE_AUTO_SORTMERGE_JOIN_BIGTABLE_SELECTOR);
bigTableMatcherClass =
JavaUtils.loadClass(selector);
} catch (ClassNotFoundException e) {
throw new SemanticException(e.getMessage());
}
BigTableSelectorForAutoSMJ bigTableMatcher =
ReflectionUtils.newInstance(bigTableMatcherClass, null);
JoinDesc joinDesc = joinOp.getConf();
JoinCondDesc[] joinCondns = joinDesc.getConds();
Set<Integer> joinCandidates = MapJoinProcessor.getBigTableCandidates(joinCondns);
if (joinCandidates.isEmpty()) {
// This is a full outer join. This can never be a map-join
// of any type. So return false.
return false;
}
int bigTablePosition =
bigTableMatcher.getBigTablePosition(pGraphContext, joinOp, joinCandidates);
if (bigTablePosition < 0) {
// contains aliases from sub-query
return false;
}
context.setBigTablePosition(bigTablePosition);
String joinAlias =
bigTablePosition == 0 ?
joinOp.getConf().getLeftAlias() :
joinOp.getConf().getRightAliases()[bigTablePosition - 1];
joinAlias = QB.getAppendedAliasFromId(joinOp.getConf().getId(), joinAlias);
Map<Byte, List<ExprNodeDesc>> keyExprMap = new HashMap<Byte, List<ExprNodeDesc>>();
List<Operator<? extends OperatorDesc>> parentOps = joinOp.getParentOperators();
// get the join keys from parent ReduceSink operators
for (Operator<? extends OperatorDesc> parentOp : parentOps) {
ReduceSinkDesc rsconf = ((ReduceSinkOperator)parentOp).getConf();
Byte tag = (byte) rsconf.getTag();
List<ExprNodeDesc> keys = rsconf.getKeyCols();
keyExprMap.put(tag, keys);
}
context.setKeyExprMap(keyExprMap);
// Make a deep copy of the aliases so that they are not changed in the context
String[] joinSrcs = joinOp.getConf().getBaseSrc();
String[] srcs = new String[joinSrcs.length];
for (int srcPos = 0; srcPos < joinSrcs.length; srcPos++) {
joinSrcs[srcPos] = QB.getAppendedAliasFromId(joinOp.getConf().getId(), joinSrcs[srcPos]);
srcs[srcPos] = new String(joinSrcs[srcPos]);
}
// Given a candidate map-join, can this join be converted.
// The candidate map-join was derived from the pluggable sort merge join big
// table matcher.
return checkConvertBucketMapJoin(
context,
joinOp.getConf().getAliasToOpInfo(),
keyExprMap,
joinAlias,
Arrays.asList(srcs));
}
// Convert the join operator to a bucket map-join join operator
protected MapJoinOperator convertJoinToBucketMapJoin(
JoinOperator joinOp,
SortBucketJoinProcCtx joinContext) throws SemanticException {
MapJoinOperator mapJoinOp = new MapJoinProcessor().convertMapJoin(
pGraphContext.getConf(),
joinOp,
joinOp.getConf().isLeftInputJoin(),
joinOp.getConf().getBaseSrc(),
joinOp.getConf().getMapAliases(),
joinContext.getBigTablePosition(),
false,
false);
// Remove the join operator from the query join context
// Data structures coming from QBJoinTree
mapJoinOp.getConf().setQBJoinTreeProps(joinOp.getConf());
//
pGraphContext.getMapJoinOps().add(mapJoinOp);
pGraphContext.getJoinOps().remove(joinOp);
convertMapJoinToBucketMapJoin(mapJoinOp, joinContext);
return mapJoinOp;
}
// Convert the join operator to a sort-merge join operator
protected void convertJoinToSMBJoin(
JoinOperator joinOp,
SortBucketJoinProcCtx smbJoinContext) throws SemanticException {
MapJoinOperator mapJoinOp = convertJoinToBucketMapJoin(joinOp, smbJoinContext);
SMBMapJoinOperator smbMapJoinOp =
convertBucketMapJoinToSMBJoin(mapJoinOp, smbJoinContext);
smbMapJoinOp.setConvertedAutomaticallySMBJoin(true);
}
}