use of org.apache.hadoop.hive.ql.lib.Dispatcher in project hive by apache.
the class HiveOpConverterPostProc method transform.
@Override
public ParseContext transform(ParseContext pctx) throws SemanticException {
// 0. We check the conditions to apply this transformation,
// if we do not meet them we bail out
final boolean cboEnabled = HiveConf.getBoolVar(pctx.getConf(), HiveConf.ConfVars.HIVE_CBO_ENABLED);
final boolean returnPathEnabled = HiveConf.getBoolVar(pctx.getConf(), HiveConf.ConfVars.HIVE_CBO_RETPATH_HIVEOP);
final boolean cboSucceeded = pctx.getContext().isCboSucceeded();
if (!(cboEnabled && returnPathEnabled && cboSucceeded)) {
return pctx;
}
// 1. Initialize aux data structures
this.pctx = pctx;
this.aliasToOpInfo = new HashMap<String, Operator<? extends OperatorDesc>>();
// 2. Trigger transformation
Map<Rule, NodeProcessor> opRules = new LinkedHashMap<Rule, NodeProcessor>();
opRules.put(new RuleRegExp("R1", JoinOperator.getOperatorName() + "%"), new JoinAnnotate());
opRules.put(new RuleRegExp("R2", TableScanOperator.getOperatorName() + "%"), new TableScanAnnotate());
Dispatcher disp = new DefaultRuleDispatcher(null, opRules, null);
GraphWalker ogw = new ForwardWalker(disp);
List<Node> topNodes = new ArrayList<Node>();
topNodes.addAll(pctx.getTopOps().values());
ogw.startWalking(topNodes, null);
return pctx;
}
use of org.apache.hadoop.hive.ql.lib.Dispatcher in project hive by apache.
the class ColumnPruner method transform.
/**
* Transform the query tree. For each table under consideration, check if all
* columns are needed. If not, only select the operators needed at the
* beginning and proceed.
*
* @param pactx
* the current parse context
*/
@Override
public ParseContext transform(ParseContext pactx) throws SemanticException {
pGraphContext = pactx;
// generate pruned column list for all relevant operators
ColumnPrunerProcCtx cppCtx = new ColumnPrunerProcCtx(pactx);
// create a walker which walks the tree in a DFS manner while maintaining
// the operator stack. The dispatcher
// generates the plan from the operator tree
Map<Rule, NodeProcessor> opRules = new LinkedHashMap<Rule, NodeProcessor>();
opRules.put(new RuleRegExp("R1", FilterOperator.getOperatorName() + "%"), ColumnPrunerProcFactory.getFilterProc());
opRules.put(new RuleRegExp("R2", GroupByOperator.getOperatorName() + "%"), ColumnPrunerProcFactory.getGroupByProc());
opRules.put(new RuleRegExp("R3", ReduceSinkOperator.getOperatorName() + "%"), ColumnPrunerProcFactory.getReduceSinkProc());
opRules.put(new RuleRegExp("R4", SelectOperator.getOperatorName() + "%"), ColumnPrunerProcFactory.getSelectProc());
opRules.put(new RuleRegExp("R5", CommonJoinOperator.getOperatorName() + "%"), ColumnPrunerProcFactory.getJoinProc());
opRules.put(new RuleRegExp("R6", MapJoinOperator.getOperatorName() + "%"), ColumnPrunerProcFactory.getMapJoinProc());
opRules.put(new RuleRegExp("R7", TableScanOperator.getOperatorName() + "%"), ColumnPrunerProcFactory.getTableScanProc());
opRules.put(new RuleRegExp("R8", LateralViewJoinOperator.getOperatorName() + "%"), ColumnPrunerProcFactory.getLateralViewJoinProc());
opRules.put(new RuleRegExp("R9", LateralViewForwardOperator.getOperatorName() + "%"), ColumnPrunerProcFactory.getLateralViewForwardProc());
opRules.put(new RuleRegExp("R10", PTFOperator.getOperatorName() + "%"), ColumnPrunerProcFactory.getPTFProc());
opRules.put(new RuleRegExp("R11", ScriptOperator.getOperatorName() + "%"), ColumnPrunerProcFactory.getScriptProc());
opRules.put(new RuleRegExp("R12", LimitOperator.getOperatorName() + "%"), ColumnPrunerProcFactory.getLimitProc());
opRules.put(new RuleRegExp("R13", UnionOperator.getOperatorName() + "%"), ColumnPrunerProcFactory.getUnionProc());
// The dispatcher fires the processor corresponding to the closest matching
// rule and passes the context along
Dispatcher disp = new DefaultRuleDispatcher(ColumnPrunerProcFactory.getDefaultProc(), opRules, cppCtx);
GraphWalker ogw = new ColumnPrunerWalker(disp);
// Create a list of topop nodes
ArrayList<Node> topNodes = new ArrayList<Node>();
topNodes.addAll(pGraphContext.getTopOps().values());
ogw.startWalking(topNodes, null);
// set it back so that column pruner in the optimizer will not do the
// view column authorization again even if it is triggered again.
pGraphContext.setNeedViewColumnAuthorization(false);
return pGraphContext;
}
use of org.apache.hadoop.hive.ql.lib.Dispatcher in project hive by apache.
the class MapJoinProcessor method transform.
/**
* 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.
*
* @param pactx
* current parse context
*/
@Override
public ParseContext transform(ParseContext pactx) throws SemanticException {
List<MapJoinOperator> listMapJoinOps = new ArrayList<MapJoinOperator>();
// traverse all the joins and convert them if necessary
if (pactx.getJoinOps() != null) {
Set<JoinOperator> joinMap = new HashSet<JoinOperator>();
Set<MapJoinOperator> mapJoinMap = pactx.getMapJoinOps();
if (mapJoinMap == null) {
mapJoinMap = new HashSet<MapJoinOperator>();
pactx.setMapJoinOps(mapJoinMap);
}
Iterator<JoinOperator> joinCtxIter = pactx.getJoinOps().iterator();
while (joinCtxIter.hasNext()) {
JoinOperator joinOp = joinCtxIter.next();
int mapJoinPos = mapSideJoin(joinOp);
if (mapJoinPos >= 0) {
MapJoinOperator mapJoinOp = generateMapJoinOperator(pactx, joinOp, mapJoinPos);
listMapJoinOps.add(mapJoinOp);
mapJoinOp.getConf().setQBJoinTreeProps(joinOp.getConf());
mapJoinMap.add(mapJoinOp);
} else {
joinOp.getConf().setQBJoinTreeProps(joinOp.getConf());
joinMap.add(joinOp);
}
}
// store the new joinContext
pactx.setJoinOps(joinMap);
}
// Go over the list and find if a reducer is not needed
List<AbstractMapJoinOperator<? extends MapJoinDesc>> listMapJoinOpsNoRed = new ArrayList<AbstractMapJoinOperator<? extends MapJoinDesc>>();
// create a walker which walks the tree in a DFS manner while maintaining
// the operator stack.
// The dispatcher generates the plan from the operator tree
Map<Rule, NodeProcessor> opRules = new LinkedHashMap<Rule, NodeProcessor>();
opRules.put(new RuleRegExp("R0", MapJoinOperator.getOperatorName() + "%"), getCurrentMapJoin());
opRules.put(new RuleRegExp("R1", MapJoinOperator.getOperatorName() + "%.*" + FileSinkOperator.getOperatorName() + "%"), getMapJoinFS());
opRules.put(new RuleRegExp("R2", MapJoinOperator.getOperatorName() + "%.*" + ReduceSinkOperator.getOperatorName() + "%"), getMapJoinDefault());
opRules.put(new RuleRegExp("R4", MapJoinOperator.getOperatorName() + "%.*" + UnionOperator.getOperatorName() + "%"), getMapJoinDefault());
// The dispatcher fires the processor corresponding to the closest matching
// rule and passes the context along
Dispatcher disp = new DefaultRuleDispatcher(getDefault(), opRules, new MapJoinWalkerCtx(listMapJoinOpsNoRed, pactx));
GraphWalker ogw = new GenMapRedWalker(disp);
ArrayList<Node> topNodes = new ArrayList<Node>();
topNodes.addAll(listMapJoinOps);
ogw.startWalking(topNodes, null);
pactx.setListMapJoinOpsNoReducer(listMapJoinOpsNoRed);
return pactx;
}
use of org.apache.hadoop.hive.ql.lib.Dispatcher in project hive by apache.
the class PrunerUtils method walkOperatorTree.
/**
* Walk operator tree for pruner generation.
*
* @param pctx
* @param opWalkerCtx
* @param filterProc
* @param defaultProc
* @throws SemanticException
*/
public static void walkOperatorTree(ParseContext pctx, NodeProcessorCtx opWalkerCtx, NodeProcessor filterProc, NodeProcessor defaultProc) throws SemanticException {
Map<Rule, NodeProcessor> opRules = new LinkedHashMap<Rule, NodeProcessor>();
// Build regular expression for operator rule.
// "(TS%FIL%)|(TS%FIL%FIL%)"
String tsOprName = TableScanOperator.getOperatorName();
String filtOprName = FilterOperator.getOperatorName();
opRules.put(new RuleExactMatch("R1", new String[] { tsOprName, filtOprName, filtOprName }), filterProc);
opRules.put(new RuleExactMatch("R2", new String[] { tsOprName, filtOprName }), filterProc);
// The dispatcher fires the processor corresponding to the closest matching
// rule and passes the context along
Dispatcher disp = new DefaultRuleDispatcher(defaultProc, opRules, opWalkerCtx);
GraphWalker ogw = new DefaultGraphWalker(disp);
// Create a list of topop nodes
ArrayList<Node> topNodes = new ArrayList<Node>();
topNodes.addAll(pctx.getTopOps().values());
ogw.startWalking(topNodes, null);
}
use of org.apache.hadoop.hive.ql.lib.Dispatcher in project hive by apache.
the class PrunerUtils method walkExprTree.
/**
* Walk expression tree for pruner generation.
*
* @param pred
* @param ctx
* @param colProc
* @param fieldProc
* @param genFuncProc
* @param defProc
* @return
* @throws SemanticException
*/
public static Map<Node, Object> walkExprTree(ExprNodeDesc pred, NodeProcessorCtx ctx, NodeProcessor colProc, NodeProcessor fieldProc, NodeProcessor genFuncProc, NodeProcessor defProc) throws SemanticException {
// create a walker which walks the tree in a DFS manner while maintaining
// the operator stack. The dispatcher
// generates the plan from the operator tree
Map<Rule, NodeProcessor> exprRules = new LinkedHashMap<Rule, NodeProcessor>();
exprRules.put(new TypeRule(ExprNodeColumnDesc.class), colProc);
exprRules.put(new TypeRule(ExprNodeFieldDesc.class), fieldProc);
exprRules.put(new TypeRule(ExprNodeGenericFuncDesc.class), genFuncProc);
// The dispatcher fires the processor corresponding to the closest matching
// rule and passes the context along
Dispatcher disp = new DefaultRuleDispatcher(defProc, exprRules, ctx);
GraphWalker egw = new DefaultGraphWalker(disp);
List<Node> startNodes = new ArrayList<Node>();
startNodes.add(pred);
HashMap<Node, Object> outputMap = new HashMap<Node, Object>();
egw.startWalking(startNodes, outputMap);
return outputMap;
}
Aggregations