Examples with DefaultGraphWalker org.apache.hadoop.hive.ql.lib.DefaultGraphWalker used on opensource projects

Example 11 with DefaultGraphWalker

use of org.apache.hadoop.hive.ql.lib.DefaultGraphWalker in project hive by apache.

the class SamplePruner method transform.

/*
   * (non-Javadoc)
   *
   * @see
   * org.apache.hadoop.hive.ql.optimizer.Transform#transform(org.apache.hadoop
   * .hive.ql.parse.ParseContext)
   */
@Override
public ParseContext transform(ParseContext pctx) throws SemanticException {
    // create a the context for walking operators
    SamplePrunerCtx samplePrunerCtx = new SamplePrunerCtx(pctx.getOpToSamplePruner());
    Map<Rule, NodeProcessor> opRules = new LinkedHashMap<Rule, NodeProcessor>();
    opRules.put(new RuleRegExp("R1", "(" + TableScanOperator.getOperatorName() + "%" + FilterOperator.getOperatorName() + "%" + FilterOperator.getOperatorName() + "%|" + TableScanOperator.getOperatorName() + "%" + FilterOperator.getOperatorName() + "%)"), getFilterProc());
    // The dispatcher fires the processor corresponding to the closest matching
    // rule and passes the context along
    Dispatcher disp = new DefaultRuleDispatcher(getDefaultProc(), opRules, samplePrunerCtx);
    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);
    return pctx;
}

Example 12 with DefaultGraphWalker

use of org.apache.hadoop.hive.ql.lib.DefaultGraphWalker in project hive by apache.

the class ReduceSinkDeDuplication method transform.

@Override
public ParseContext transform(ParseContext pctx) throws SemanticException {
    pGraphContext = pctx;
    // generate pruned column list for all relevant operators
    ReduceSinkDeduplicateProcCtx cppCtx = new ReduceSinkDeduplicateProcCtx(pGraphContext);
    // for auto convert map-joins, it not safe to dedup in here (todo)
    boolean mergeJoins = !pctx.getConf().getBoolVar(HIVECONVERTJOIN) && !pctx.getConf().getBoolVar(HIVECONVERTJOINNOCONDITIONALTASK) && !pctx.getConf().getBoolVar(ConfVars.HIVE_CONVERT_JOIN_BUCKET_MAPJOIN_TEZ) && !pctx.getConf().getBoolVar(ConfVars.HIVEDYNAMICPARTITIONHASHJOIN);
    // If multiple rules can be matched with same cost, last rule will be choosen as a processor
    // see DefaultRuleDispatcher#dispatch()
    Map<Rule, NodeProcessor> opRules = new LinkedHashMap<Rule, NodeProcessor>();
    opRules.put(new RuleRegExp("R1", RS + "%.*%" + RS + "%"), ReduceSinkDeduplicateProcFactory.getReducerReducerProc());
    opRules.put(new RuleRegExp("R2", RS + "%" + GBY + "%.*%" + RS + "%"), ReduceSinkDeduplicateProcFactory.getGroupbyReducerProc());
    if (mergeJoins) {
        opRules.put(new RuleRegExp("R3", JOIN + "%.*%" + RS + "%"), ReduceSinkDeduplicateProcFactory.getJoinReducerProc());
    }
    // TODO RS+JOIN
    // The dispatcher fires the processor corresponding to the closest matching
    // rule and passes the context along
    Dispatcher disp = new DefaultRuleDispatcher(ReduceSinkDeduplicateProcFactory.getDefaultProc(), opRules, cppCtx);
    GraphWalker ogw = new DefaultGraphWalker(disp);
    // Create a list of topop nodes
    ArrayList<Node> topNodes = new ArrayList<Node>();
    topNodes.addAll(pGraphContext.getTopOps().values());
    ogw.startWalking(topNodes, null);
    return pGraphContext;
}

Example 13 with DefaultGraphWalker

use of org.apache.hadoop.hive.ql.lib.DefaultGraphWalker in project hive by apache.

the class PartitionConditionRemover method transform.

/*
   * (non-Javadoc)
   *
   * @see
   * org.apache.hadoop.hive.ql.optimizer.Transform#transform(org.apache.hadoop
   * .hive.ql.parse.ParseContext)
   */
@Override
public ParseContext transform(ParseContext pctx) throws SemanticException {
    // create a the context for walking operators
    List<PcrOpWalkerCtx.OpToDeleteInfo> opToRemove = new ArrayList<PcrOpWalkerCtx.OpToDeleteInfo>();
    PcrOpWalkerCtx opWalkerCtx = new PcrOpWalkerCtx(pctx, opToRemove);
    Map<Rule, NodeProcessor> opRules = new LinkedHashMap<Rule, NodeProcessor>();
    opRules.put(new RuleRegExp("R1", "(" + TableScanOperator.getOperatorName() + "%" + FilterOperator.getOperatorName() + "%)|(" + TableScanOperator.getOperatorName() + "%" + FilterOperator.getOperatorName() + "%" + FilterOperator.getOperatorName() + "%)"), PcrOpProcFactory.getFilterProc());
    // The dispatcher fires the processor corresponding to the closest matching
    // rule and passes the context along
    Dispatcher disp = new DefaultRuleDispatcher(PcrOpProcFactory.getDefaultProc(), 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);
    for (PcrOpWalkerCtx.OpToDeleteInfo entry : opToRemove) {
        entry.getParent().removeChildAndAdoptItsChildren(entry.getOperator());
    }
    return pctx;
}

Example 14 with DefaultGraphWalker

use of org.apache.hadoop.hive.ql.lib.DefaultGraphWalker in project hive by apache.

the class SparkCompiler method runJoinOptimizations.

private void runJoinOptimizations(OptimizeSparkProcContext procCtx) throws SemanticException {
    ParseContext pCtx = procCtx.getParseContext();
    Map<Rule, NodeProcessor> opRules = new LinkedHashMap<Rule, NodeProcessor>();
    opRules.put(new TypeRule(JoinOperator.class), new SparkJoinOptimizer(pCtx));
    opRules.put(new TypeRule(MapJoinOperator.class), new SparkJoinHintOptimizer(pCtx));
    opRules.put(new RuleRegExp("Disabling Dynamic Partition Pruning By Size", SparkPartitionPruningSinkOperator.getOperatorName() + "%"), new SparkRemoveDynamicPruningBySize());
    // The dispatcher fires the processor corresponding to the closest matching
    // rule and passes the context along
    Dispatcher disp = new DefaultRuleDispatcher(null, opRules, procCtx);
    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);
}

Example 15 with DefaultGraphWalker

use of org.apache.hadoop.hive.ql.lib.DefaultGraphWalker in project hive by apache.

the class AnnotateReduceSinkOutputOperator method transform.

@Override
public ParseContext transform(ParseContext pctx) throws SemanticException {
    // 1. We apply the transformation
    Map<Rule, NodeProcessor> opRules = new LinkedHashMap<Rule, NodeProcessor>();
    opRules.put(new RuleRegExp("R1", "(" + ReduceSinkOperator.getOperatorName() + "%)"), new ReduceSinkOutputOperatorAnnotator());
    GraphWalker ogw = new DefaultGraphWalker(new DefaultRuleDispatcher(null, opRules, null));
    ArrayList<Node> topNodes = new ArrayList<Node>();
    topNodes.addAll(pctx.getTopOps().values());
    ogw.startWalking(topNodes, null);
    return pctx;
}