use of org.apache.hadoop.hive.ql.lib.SemanticNodeProcessor in project hive by apache.
the class AnnotateRunTimeStatsOptimizer method resolve.
@Override
public PhysicalContext resolve(PhysicalContext pctx) throws SemanticException {
Map<SemanticRule, SemanticNodeProcessor> opRules = new LinkedHashMap<SemanticRule, SemanticNodeProcessor>();
SemanticDispatcher disp = new AnnotateRunTimeStatsDispatcher(pctx, opRules);
SemanticGraphWalker ogw = new DefaultGraphWalker(disp);
ArrayList<Node> topNodes = new ArrayList<Node>();
topNodes.addAll(pctx.getRootTasks());
ogw.startWalking(topNodes, null);
return pctx;
}
use of org.apache.hadoop.hive.ql.lib.SemanticNodeProcessor in project hive by apache.
the class BucketingSortingInferenceOptimizer method inferBucketingSorting.
/**
* For each map reduce task, if it has a reducer, infer whether or not the final output of the
* reducer is bucketed and/or sorted
*
* @param mapRedTasks
* @throws SemanticException
*/
private void inferBucketingSorting(List<ExecDriver> mapRedTasks) throws SemanticException {
for (ExecDriver mapRedTask : mapRedTasks) {
// of the outputs of intermediate map reduce jobs.
if (!mapRedTask.getWork().isFinalMapRed()) {
continue;
}
if (mapRedTask.getWork().getReduceWork() == null) {
continue;
}
Operator<? extends OperatorDesc> reducer = mapRedTask.getWork().getReduceWork().getReducer();
// uses sampling, which means it's not bucketed
boolean disableBucketing = mapRedTask.getWork().getMapWork().getSamplingType() > 0;
BucketingSortingCtx bCtx = new BucketingSortingCtx(disableBucketing);
// RuleRegExp rules are used to match operators anywhere in the tree
// RuleExactMatch rules are used to specify exactly what the tree should look like
// In particular, this guarantees that the first operator is the reducer
// (and its parent(s) are ReduceSinkOperators) since it begins walking the tree from
// the reducer.
Map<SemanticRule, SemanticNodeProcessor> opRules = new LinkedHashMap<SemanticRule, SemanticNodeProcessor>();
opRules.put(new RuleRegExp("R1", SelectOperator.getOperatorName() + "%"), BucketingSortingOpProcFactory.getSelProc());
// Matches only GroupByOperators which are reducers, rather than map group by operators,
// or multi group by optimization specific operators
opRules.put(new RuleExactMatch("R2", new String[] { GroupByOperator.getOperatorName() }), BucketingSortingOpProcFactory.getGroupByProc());
// Matches only JoinOperators which are reducers, rather than map joins, SMB map joins, etc.
opRules.put(new RuleExactMatch("R3", new String[] { JoinOperator.getOperatorName() }), BucketingSortingOpProcFactory.getJoinProc());
opRules.put(new RuleRegExp("R5", FileSinkOperator.getOperatorName() + "%"), BucketingSortingOpProcFactory.getFileSinkProc());
opRules.put(new RuleRegExp("R7", FilterOperator.getOperatorName() + "%"), BucketingSortingOpProcFactory.getFilterProc());
opRules.put(new RuleRegExp("R8", LimitOperator.getOperatorName() + "%"), BucketingSortingOpProcFactory.getLimitProc());
opRules.put(new RuleRegExp("R9", LateralViewForwardOperator.getOperatorName() + "%"), BucketingSortingOpProcFactory.getLateralViewForwardProc());
opRules.put(new RuleRegExp("R10", LateralViewJoinOperator.getOperatorName() + "%"), BucketingSortingOpProcFactory.getLateralViewJoinProc());
// Matches only ForwardOperators which are preceded by some other operator in the tree,
// in particular it can't be a reducer (and hence cannot be one of the ForwardOperators
// added by the multi group by optimization)
opRules.put(new RuleRegExp("R11", ".+" + ForwardOperator.getOperatorName() + "%"), BucketingSortingOpProcFactory.getForwardProc());
// Matches only ForwardOperators which are reducers and are followed by GroupByOperators
// (specific to the multi group by optimization)
opRules.put(new RuleExactMatch("R12", new String[] { ForwardOperator.getOperatorName(), GroupByOperator.getOperatorName() }), BucketingSortingOpProcFactory.getMultiGroupByProc());
// The dispatcher fires the processor corresponding to the closest matching rule and passes
// the context along
SemanticDispatcher disp = new DefaultRuleDispatcher(BucketingSortingOpProcFactory.getDefaultProc(), opRules, bCtx);
SemanticGraphWalker ogw = new PreOrderWalker(disp);
// Create a list of topop nodes
ArrayList<Node> topNodes = new ArrayList<Node>();
topNodes.add(reducer);
ogw.startWalking(topNodes, null);
mapRedTask.getWork().getMapWork().getBucketedColsByDirectory().putAll(bCtx.getBucketedColsByDirectory());
mapRedTask.getWork().getMapWork().getSortedColsByDirectory().putAll(bCtx.getSortedColsByDirectory());
}
}
use of org.apache.hadoop.hive.ql.lib.SemanticNodeProcessor in project hive by apache.
the class CorrelationOptimizer method transform.
/**
* Detect correlations and transform the query tree.
*
* @param pctx
* current parse context
* @throws SemanticException
*/
@Override
public ParseContext transform(ParseContext pctx) throws SemanticException {
pCtx = pctx;
if (HiveConf.getBoolVar(pCtx.getConf(), HiveConf.ConfVars.HIVECONVERTJOIN)) {
findPossibleAutoConvertedJoinOperators();
}
// detect correlations
CorrelationNodeProcCtx corrCtx = new CorrelationNodeProcCtx(pCtx);
Map<SemanticRule, SemanticNodeProcessor> opRules = new LinkedHashMap<SemanticRule, SemanticNodeProcessor>();
opRules.put(new RuleRegExp("R1", ReduceSinkOperator.getOperatorName() + "%"), new CorrelationNodeProc());
SemanticDispatcher disp = new DefaultRuleDispatcher(getDefaultProc(), opRules, corrCtx);
SemanticGraphWalker ogw = new DefaultGraphWalker(disp);
// Create a list of topOp nodes
List<Node> topNodes = new ArrayList<Node>();
topNodes.addAll(pCtx.getTopOps().values());
ogw.startWalking(topNodes, null);
// We have finished tree walking (correlation detection).
// We will first see if we need to abort (the operator tree has not been changed).
// If not, we will start to transform the operator tree.
abort = corrCtx.isAbort();
if (abort) {
LOG.info("Abort. Reasons are ...");
for (String reason : corrCtx.getAbortReasons()) {
LOG.info("-- " + reason);
}
} else {
// transform the operator tree
LOG.info("Begain query plan transformation based on intra-query correlations. " + corrCtx.getCorrelations().size() + " correlation(s) to be applied");
for (IntraQueryCorrelation correlation : corrCtx.getCorrelations()) {
QueryPlanTreeTransformation.applyCorrelation(pCtx, corrCtx, correlation);
}
}
return pCtx;
}
use of org.apache.hadoop.hive.ql.lib.SemanticNodeProcessor in project hive by apache.
the class PcrExprProcFactory method walkExprTree.
/**
* Remove partition conditions when necessary from the the expression tree.
*
* @param tabAlias
* the table alias
* @param parts
* the list of all pruned partitions for the table
* @param vcs
* virtual columns referenced
* @param pred
* expression tree of the target filter operator
* @return the node information of the root expression
* @throws SemanticException
*/
public static NodeInfoWrapper walkExprTree(String tabAlias, ArrayList<Partition> parts, List<VirtualColumn> vcs, ExprNodeDesc pred) throws SemanticException {
// Create the walker, the rules dispatcher and the context.
PcrExprProcCtx pprCtx = new PcrExprProcCtx(tabAlias, parts, vcs);
Map<SemanticRule, SemanticNodeProcessor> exprRules = new LinkedHashMap<SemanticRule, SemanticNodeProcessor>();
exprRules.put(new RuleRegExp("R1", ExprNodeColumnDesc.class.getName() + "%"), getColumnProcessor());
exprRules.put(new RuleRegExp("R2", ExprNodeFieldDesc.class.getName() + "%"), getFieldProcessor());
exprRules.put(new RuleRegExp("R5", ExprNodeGenericFuncDesc.class.getName() + "%"), getGenericFuncProcessor());
// The dispatcher fires the processor corresponding to the closest matching
// rule and passes the context along
SemanticDispatcher disp = new DefaultRuleDispatcher(getDefaultExprProcessor(), exprRules, pprCtx);
SemanticGraphWalker egw = new ExpressionWalker(disp);
List<Node> startNodes = new ArrayList<Node>();
startNodes.add(pred);
HashMap<Node, Object> outputMap = new HashMap<Node, Object>();
egw.startWalking(startNodes, outputMap);
// Return the wrapper of the root node
return (NodeInfoWrapper) outputMap.get(pred);
}
use of org.apache.hadoop.hive.ql.lib.SemanticNodeProcessor in project hive by apache.
the class ReduceSinkJoinDeDuplication method transform.
@Override
public ParseContext transform(ParseContext pctx) throws SemanticException {
pGraphContext = pctx;
ReduceSinkJoinDeDuplicateProcCtx cppCtx = new ReduceSinkJoinDeDuplicateProcCtx(pGraphContext);
Map<SemanticRule, SemanticNodeProcessor> opRules = new LinkedHashMap<SemanticRule, SemanticNodeProcessor>();
opRules.put(new RuleRegExp("R1", ReduceSinkOperator.getOperatorName() + "%"), ReduceSinkJoinDeDuplicateProcFactory.getReducerMapJoinProc());
SemanticDispatcher disp = new DefaultRuleDispatcher(ReduceSinkJoinDeDuplicateProcFactory.getDefaultProc(), opRules, cppCtx);
SemanticGraphWalker ogw = new ForwardWalker(disp);
// Create a list of topop nodes
ArrayList<Node> topNodes = new ArrayList<Node>();
topNodes.addAll(pGraphContext.getTopOps().values());
ogw.startWalking(topNodes, null);
return pGraphContext;
}
Aggregations